Empire Teas CRM — PRD History
Sidecar to
project-prd.md. Carries audit trail only — questions that have been answered (with the answer folded into the PRD body) and PRD revisions.
What lives here
- Resolved open questions — every OQ that was once in
project-prd-signals.mdand has since been answered. The answer itself has already been folded intoproject-prd.md(or into an AD in.claude/CLAUDE.mdfor architectural decisions). The entry here is the historical record with a pointer to where the answer now lives. - PRD revisions — every meaningful change to
project-prd.mdafter first sign-off. The PRD body itself no longer carries a## Revisionssection; entries are appended here instead.
What does NOT live here
- Open / partial OQs — those live in
project-prd-signals.md. - The live contract — lives in
project-prd.md.
How to add an entry
The OQ resolution procedure and the revision convention are defined in .claude/rules/prd.md. Three-step lift on resolution: fold the answer into the PRD body, move the OQ row from project-prd-signals.md to ## Resolved Open Questions below, append a ### Rev N entry under ## Revisions.
Append-only. No curation in v1 — if this file becomes hard to navigate later, that’s a signal to revisit the lifecycle, not to silently prune.
Resolved Open Questions
Each entry preserves the original question and records where the answer was folded.
Resolved in Rev N: see §Xorsee AD #Yis the canonical pointer format.
| # | Section | Topic | Question | Resolution | Resolved in | Date |
|---|---|---|---|---|---|---|
| OQ-31 | §Domain Model / §Functional Surface > F26 | “Region-linked” bulk pricing — what “region” means | The 2026-07-16 client meeting raised bulk pricing as price-sensitive, weekly-changing, and “connected to region.” Which region — the tea’s growing origin, or the market it’s sold into? | Destination market (the market Empire sells into) — confirmed directly by Dinil. Source region/grade remain product attributes on the cost side, not the pricing key. | Rev 1: see §Domain Model, §Functional Surface > F26 | 2026-07-17 |
| OQ-37 | §Functional Surface > F26 / F7 / F8 | Private-label pricing scope | Does private-label pricing share the same region-linked, weekly-updated model as bulk, or is it always quoted bespoke per deal? | Bulk-only. The region-linked, periodically-updated pricing model applies to bulk; private-label continues to be quoted bespoke per deal (single price, with additional cost lines available on the quote). The 2026-07-16 meeting confirmed only bulk pricing is price-sensitive. Revised by Rev 27 (2026-09-15): private-label items are buying standards too — the standard is the sellable record for both lines and there is no products catalog; whether private-label standards are weekly-priced is OQ-61. | Rev 2: see §Functional Surface > F26 — revised Rev 27 | 2026-07-21 |
| OQ-3 | §Users and Access | Teams using the CRM | Which teams/user types use the CRM, and are the four assumed roles correct? | Four core roles confirmed (Sales Rep, Sales Manager, R&D/tea taster, Admin), including that tea tasters use the system. Executive/Management read-only planned as a fifth role — split out as assumption OQ-47. QA/Compliance and Export-Documentation Coordinator are not separate V1 roles (covered by Sales Manager + Admin). | Rev 4: see §Users and Access | 2026-08-02 |
| OQ-4 | §Business Specifics / §Scope & Boundaries | External data stores | What sales data is still tracked outside the CRM? | Confirmed — sales data also lives in Excel, WhatsApp, and personal email; migration scope spans these non-Pipedrive sources. | Rev 4: see §Business Specifics | 2026-08-02 |
| OQ-14 | §Integration Points / §Functional Surface > F12 | Inbound channel | Where do inquiry emails arrive? | Individual rep inboxes plus a shared info@empire.lk. Confirms the inbound feed for F12 extraction; automated trigger stays deferred. Revised by Rev 37 (2026-09-24): info@ remains an inbound channel, but in V1 it reaches the CRM only by forwarding into a rep’s synced mailbox — there is no shared mailbox connection (OQ-52, OQ-54). |
Rev 4: see §Integration Points — revised Rev 37 | 2026-08-02 |
| OQ-20 | §Business Specifics / §Functional Surface > F26 | Own-brand sales & brand attribute | Does Empire sell own-branded goods through this pipeline, or only bulk + private-label? | Yes — own-branded goods (Thurson/Hyson/Akina/CWH) sell through this B2B pipeline, so Product and Deal carry a first-class brand attribute. Closes Risk R-08. Rev 27 (2026-09-15) moved the attribute from Product to Buying Standard — the catalog is removed. |
Rev 4: see §Functional Surface > F26, §Domain Model | 2026-08-02 |
| OQ-21 | §Functional Surface > F4 | Migration strategy | Migrate everything or only cleaned records? Preserve original IDs? | Migrate everything, preserve original Pipedrive IDs for audit, and let AI triage flag junk after import (not dropped beforehand). | Rev 4: see §Functional Surface > F4 | 2026-08-02 |
| OQ-24 | §Functional Surface > F20 | Email accounts & providers | Which email provider? | Microsoft 365. (Gmail/Exchange/IMAP remain supported by the build but are not in use — resolves the SP-ARCH-002 Exchange/IMAP spike trigger to “not needed”.) ⚠️ The parenthetical overstated scope — superseded by Rev 7: Microsoft 365 via Graph is the only provider in V1 scope; the others were never quoted. | Rev 4: see §Integration Points; corrected in Rev 7 | 2026-08-02 |
| OQ-15 | §Non-Functional > Security | External AI permitted | Is external cloud AI permitted, or must the AI layer be self-hosted? | Cloud AI accepted (manager, 2026-07-28) — Anthropic Claude on secure, reputable cloud infrastructure, since served in-account via Amazon Bedrock (AD #19). The residual — formal sign-off of the data-handling arrangements at contract stage — closed with the commercially accepted proposal (2026-08-16), whose Appendix A.6 states those arrangements; no separate approval document or gate (decision 2026-08-19). The AIProvider seam is retained as an architectural safeguard; SP-ARCH-001 stays stood down. Reopened 2026-09-30 (Rev 48) for the location only: may buyer content be processed outside the project’s AWS Region, and outside the US? It was asked of Empire as client question Q18 and never answered; the lead’s default meanwhile was “yes, unless Empire asks for US-only”. Resolved again 2026-11-04 (Rev 82), by the lead’s decision: yes. Buyer content may be processed in any commercial AWS Region, through Bedrock’s Global or geographic (us.) inference profiles. The route for each call is chosen for capacity and cost (ai-provider-strategy.md §4b, §5 row 9). The grounds: every call is made from Empire’s own AWS account; data between Regions stays on the AWS network, encrypted in transit; model providers have no access to prompts or answers; and neither AWS nor the providers use them for training (AWS Bedrock documentation, read 2026-11-04). All four hold on every route, whichever Region processes the request. A US-only route stays a configuration change if Empire ever asks. Q18 is answered with this decision and closed in the client question list, kept there with the answer and the reasons. |
Rev 12: see §Non-Functional > Security; reopened Rev 48; resolved again Rev 82: see §Non-Functional > Security | 2026-08-19; 2026-11-04 |
| OQ-33 | §Functional Surface > F26 / F7 | Bulk price validity window | How long is a bulk price valid; same clock as the quote or separate? | Holds until the next weekly Colombo auction, with a manual override for exceptional cases; the quote shares the same clock — a single validity clock, superseding the earlier two-clock / ~30-day assumption. | Rev 4: see §Domain Model, §Functional Surface > F26 | 2026-08-02 |
| OQ-39 | §Functional Surface > F26 / §Users and Access | Weekly price-update ownership | Who sets and publishes the weekly bulk price? | The Tea Room team sets and publishes it (central, not per-rep). Exact role/permission mapping (is “Tea Room” the R&D/taster role or its own permission) to confirm at kickoff. | Rev 4: see §Functional Surface > F26 | 2026-08-02 |
| OQ-43 | §Domain Model / §Functional Surface > F26 | Pricing currency | USD only, or multi-currency? Separately maintained or converted? | USD base; other currencies are display conversions only, not separately maintained price rows. Currency is not a price-key axis. Refined by Rev 31 (2026-09-17): USD remains the only published currency and the key is unchanged, but the USD figure is the end of an LKR build-up (price → + overhead → converted → + margin) which is now stored beside it. | Rev 4: see §Domain Model, §Functional Surface > F26; refined Rev 31 | 2026-08-02 |
| OQ-44 | §Functional Surface > F26 / F7 | Incoterm basis of published price | FOB Colombo with per-deal build-up, or separate CIF prices? | FOB Colombo (“depends on the client, mostly FOB”); CIF and other Incoterms are handled per-quote via the cost build-up (additional cost lines), not as separately maintained CIF prices. | Rev 4: see §Functional Surface > F26 | 2026-08-02 |
| OQ-46 | §Functional Surface > F26 | Current weekly-price artifact | How is the weekly price recorded/communicated today? | Via the online auction system Empire uses — both the migration seed for price history and the reference for the weekly “publish this week’s price” workflow. | Rev 4: see §Functional Surface > F26 | 2026-08-02 |
| OQ-42 | §Functional Surface > F26 | Bulk price key / granularity | When Empire sets a bulk price, what is it attached to — grade, family, or ad-hoc; and can two buyers hold different list prices? | The priced unit is a client-specific Buying Standard — a grade/blend spec a client buys (e.g. BOP1, POSOS) with a unique running number never reused across clients; a client buys 4–5, each with its own weekly price. Confirmed 2026-08-03. | Rev 6: see §Domain Model (Buying Standard), §Functional Surface > F26 | 2026-08-03 |
| OQ-38 | §Domain Model / §Functional Surface > F26 | Destination markets | Which markets get a different price for the same tea? | Price differs completely by destination market (Europe high, Middle East/Eastern nominal). Market is a first-class price driver. Revised by Rev 23 (2026-09-11): one price per standard per week; destination market is a customer attribute and a quote field, not a price key. | Rev 6: see §Domain Model, §Functional Surface > F26 — revised Rev 23 | 2026-08-03 |
| OQ-45 | §Functional Surface > F26 / F7 | Volume-based price tiering | Formal volume tiers, or rep discretion? | No formal tiers. Order size moves the per-kg price only nominally (1–2 cents); not a pricing dimension. Price drops come from value addition, not volume. | Rev 6: see §Functional Surface > F26 | 2026-08-03 |
| OQ-40 | §Functional Surface > F7 / §Users and Access | Pricing/margin approval authority | Who approves quote pricing/margin? | Reps quote freely at or above a required gross-profit % floor; sub-floor quotes require Managing Director approval. The CRM holds cost and computes GP%. | Rev 6: see §Functional Surface > F7, §Users and Access | 2026-08-03 |
| OQ-10 | §Domain Model / §Functional Surface > F7 | Payment terms | Which terms are standard; field on quote or ERP? | Terms are LC / Open Account / DA / CAD (+ advance %), set per customer by finance and held in the ERP. The CRM shows a simple payment-terms field on the quote/account; detailed administration stays in the ERP. | Rev 6: see §Domain Model (Quote), §Functional Surface > F7 | 2026-08-03 |
| OQ-11 | §Functional Surface > F6 | Sample rounds | How many rounds does a typical private-label deal take? | 1–2 when matching an existing client sample; 6–7 when developing a new standard; almost always closed before 10. Reminder defaults calibrated to this; unlimited rounds supported. | Rev 6: see §Functional Surface > F6 | 2026-08-03 |
| OQ-47 | §Users and Access | Executive read-only role + permission levels | Include an Executive read-only role? | Yes. Plus the client wants granular permission levels — read-only, edit, AI-use, and execute — layered on the roles. Revised by Rev 37 (2026-09-24, AD #34): the levels are met by a view and a manage key per entity plus a short list of special keys, with no category axis (the fold was restated before at Rev 17, Rev 21 and Rev 24). |
Rev 6: see §Users and Access — revised Rev 37 | 2026-08-03 |
| OQ-48 | §Scope & Boundaries / §Functional Surface | Trade-show lead-capture app in V1? | Is the proposal §4 trade-show lead-capture companion app part of V1, and at what scope? | Yes — full scope inside V1 (Option C, decided 2026-08-18), folded in as F36: a PWA in its own repository at its own URL (AD #20/#22) — capture + AI card read + research/scoring on the AIProvider seam + SES email notifications, feeding the F11 intake; parallel work-stream weeks 8–13 with a capture-and-feed descope lever at the mid-November checkpoint. The §Deferred trade-fair-intake row was superseded (per-lead capture = F36; organizer bulk lists stay F23+F11). Prototype published read-only as forge-et-pwa-poc. Originally parked by Dinil 2026-08-06; scope settled 2026-08-18; folded 2026-08-19. |
Rev 13: see §In Scope (V1), §Functional Surface > F36 | 2026-08-19 |
| OQ-50 | §Constraints / commercial | Reuse rights over the generic CRM layer | May the generic (non-tea) layer be reused on future projects? | Moot — the reuse is no longer pursued. Assumed yes on 2026-08-06 (advice: no obstacle), and the full core/domain separation proceeded on that basis. AD #29 (2026-09-07) stopped extending the split once the reuse was no longer expected; AD #31 (2026-09-09, FS-14) then collapsed it outright — one module, one public schema, feature-first packages. No reusable generic layer results from this engagement, so the rights question has nothing to attach to; the OQ is closed rather than answered. |
Rev 22: see AD #29 / AD #31 | 2026-09-10 |
| OQ-58 | §Users and Access / §Functional Surface > F5 | Owner-initiated sharing of a lead or deal | Empire (2026-09-09): the owner can share a lead or deal with a colleague, who then sees everything on it. A third visibility path beside ownership and the reporting hierarchy — per-record, owner-granted. Decide: view or edit; whether the share covers linked correspondence; who may revoke; whether a manager’s team reach already covers the common case. |
Owner + shared, not co-ownership. A record_share row per (record, colleague) beside a singular owner. A share widens reach, never authority — no access level on the share; the recipient’s role decides what they may do. Covers the record and everything attached including record-linked mail, but never widens mailbox.read. Not transitive; survives owner reassignment; revocable by owner / sharer / covering manager; shown on the record, not in the per-user viewer. A manager needs no share — team already reaches their reports — so the case it serves is peer-to-peer and cross-line. Full decision record: .forge/design/data/record-sharing-model.md. Revised by Rev 37 (2026-09-24, AD #35): sharing is transitive; sharing, removing a share and reassigning the owner all need the entity’s manage key, and anyone with manage access to the record may remove any share on it; the picker offers only active users whose role holds the entity’s view key and who do not already reach the record; shares roll up through team; the table carries one nullable foreign key per entity type with shared_with_user_id, shared_by_user_id and shared_at; and the rule is built once as the F5-f substrate, used by both list queries and single-record checks. “Never widens mailbox.read” now reads: no key, at any scope, reaches another person’s mailbox. |
Rev 29: see §Multi-Tenancy / Org Hierarchy — revised Rev 37 | 2026-09-16 |
| OQ-64 | §Users and Access > Per-Role Capability Matrix / §Functional Surface > F1, F32 | Delete an account or contact, or archive only? | Raised 2026-09-18 by the F1-a spec (FR-A7): the PRD matrix granted D on Account and Contact to managers and administrators, client/roles-and-permissions.md promised managers a delete, F32-b listed delete as a bulk action, while F1-a ships archive only. |
Archive only in V1 (Tharinda, 2026-09-24). The matrix’s D on Account and Contact becomes archive; there is no hard delete, and a true delete would be a change request. client/roles-and-permissions.md is corrected to say managers archive. F32-b’s bulk action is archive, not delete. |
Rev 36: see §Users and Access > Per-Role Capability Matrix | 2026-09-24 |
| OQ-65 | §Users and Access / §Functional Surface > F1, F5 | No owning feature for the Administrator’s reference-data surface | Raised 2026-09-18 by the F1-a spec (FR-R2, Q-4): the Administrator manages “dropdown lists”, but no feature owns an administration screen, and F1-a ships the account vocabularies as code-declared values. Options: (a) code-declared for V1; (b) an admin-editable reference_value table. |
(a) for V1 (Tharinda, 2026-09-24). The account dropdown lists stay code-declared, changed by a developer release; there is no administration screen for them in V1, so no feature needs to own one. The certification list stays a table. Empire is told in the weekly update. | Rev 36: see §Users and Access > Roles (Administrator) | 2026-09-24 |
| OQ-51 | §Users and Access / §Functional Surface > F5 | Admin user provisioning & offboarding | How does an Administrator create a new user and assign their role — an invite email (an SES template flow), admin pre-creation (the account exists; the person signs in with their company Microsoft account), or auto-provision on first Microsoft sign-in behind a domain allow-list (which needs a default-role rule)? And the role and offboarding decisions around it: is the role mandatory at creation; one role or several; may the role change later; who may create an Administrator; last-admin protection; what happens to a deactivated user’s owned records and mailbox connection; who authors roles and sets manager_id; and how a user with reports is offboarded. Raised 2026-08-20 from client notes. Narrowed by Rev 20 (login is Microsoft-only, so no CRM password exists); widened by Rev 21 (role authoring, hierarchy assignment); answered in part by Rev 23 (a deal or quote always has an active owner, so records are reassigned before deactivation) and Rev 24 (one role per user, no templates, roles authored by Administrators); Gate 3 default 2026-09-17: admin pre-creation. |
Admin pre-creation only (Tharinda, 2026-09-24; AD #34). The holder of access.manage creates each user with one mandatory role; there is no invitation email and no auto-provisioning. Sign-in succeeds only when the Microsoft-verified email matches exactly one active user. Only the holder of access.manage creates and edits users, assigns roles, sets managers, authors roles, and deactivates or reactivates users; users are never deleted. Deactivation ends the user’s sessions at once, and is refused for oneself, when it would leave no active holder of access.manage, while the user has direct reports, and while they own any lead, account, deal or quote. Changing a user’s email clears their Microsoft identity link. Role and permission changes apply on the holder’s next request, with no sign-out. The first administrator in every environment is seeded by a Flyway migration (an editable IT Admin role and one user, the email set per environment); there is no superadmin. Each user’s manager is set on the user record by the holder of access.manage. |
Rev 37: see §Functional Surface > F5, §Domain Model (User); AD #34 | 2026-09-24 |
| OQ-52 | §Functional Surface > F20 / §Users and Access | Shared info@ inbox — UI placement, visibility, assignment |
The data model was settled (OQ-14 — an unowned MailboxConnection with its own assignment rule and rate bucket); the user-facing half was not. (a) UI — does shared mail appear in the in-CRM Inbox beside personal mail, or on a separate screen? (b) Visibility — who may read the shared inbox? © Assignment and triage handoff — which rule assigns an info@ inquiry, and does it raise a lead in the F11 triage queue? Raised 2026-08-20 from client notes. Answered in part by Rev 23: an info@ inquiry about a region goes to the rep in charge of that region, and two copies of one inquiry must yield one lead; whether info@ is a shared mailbox or a distribution list stayed unanswered. |
V1 handles info@ by forwarding (Tharinda, 2026-09-24). Someone forwards an inquiry to a rep; the rep’s own synced mailbox brings it in; the rep creates the lead by hand or with F12 extraction. There is no shared-inbox screen, visibility rule or assignment rule, and no unowned mailbox connection, so parts (a) to © have nothing to attach to; the region → rep mapping recorded at Rev 23 is not built, because whoever forwards the inquiry chooses the rep. If Empire wants, info@ can be added as an ordinary CRM user, which needs nothing built, provided it is a licensed mailbox that can sign in to Microsoft itself (not a sign-in-disabled Microsoft 365 shared mailbox); its actions are then attributed to info@, not to the person using it. F20-e (the shared inquiries inbox) leaves V1. |
Rev 37: see §Functional Surface > F20, §Integration Points, §Domain Model (MailboxConnection) | 2026-09-24 |
| OQ-54 | §Functional Surface > F20 / §Scope & Boundaries | Shared info@ inbox — is it funded V1 scope? |
Four documents diverged: the internal PRD and architecture.md carried the shared info@ inbox as first-class, and the issued dependency register asked Empire which shared address to connect, while the accepted proposal scopes mailboxes only as “Each user connects their own Microsoft 365 mailbox” and the client-facing PRD describes per-user sync only. Is the shared inbox (a) inside accepted V1 scope as part of “email in the CRM”, or (b) additive scope needing a change request? Raised 2026-08-20 as a commercial call. |
Not V1 scope (Tharinda, 2026-09-24). info@ is handled by forwarding (OQ-52), and F20-e leaves V1. This matches the accepted proposal’s per-user mailbox scope, so no change request arises and the proposal needs no post-issue note on this point. The issued dependency register and client question Q6, which told Empire the shared address would be connected, are to be corrected and Empire told. |
Rev 37: see §Functional Surface > F20, §Out of Scope | 2026-09-24 |
| OQ-56 | §Users and Access > Multi-Tenancy / Org Hierarchy / §Functional Surface > F5 | Reporting-hierarchy policy — four client-facing questions | Raised 2026-09-07 with Rev 21 / AD #30. (a) Are records with no owner (the shared info@ inbox’s leads) visible to every team/all holder, or worked from an explicit Unassigned queue? Default put to Empire: both. (b) Deactivating a manager who has reports — auto-reparent, or block until every report is reassigned? Default: block. © Is team recursive, or direct reports only? Default: recursive. (d) May a manager edit a report’s records, or only view them? Default: view and edit at team, with the mailbox at own regardless. Put to the client 2026-09-07 in client/roles-and-permissions.md §2 (A-4–A-8) and client/open-questions.md Q1. |
Answered (Tharinda, 2026-09-24; AD #35). (a) Unowned records are visible only to all-scope holders; F11 gives them an Unassigned filter, with no separate queue — this differs from the default put to Empire, which also showed them to team holders and listed them in a queue. Only a lead can be unowned (its owner is nullable for import edge cases). (b) Deactivation is refused while the user has direct reports — give them another manager first; deactivated users stay in the reporting tree, and shares to them stay, shown as deactivated and removable. © team is recursive through all levels, and includes records shared with anyone below the holder. (d) Edit versus view is configuration — manage at team or view at team — and nothing in code decides it. No key at any scope reaches another person’s mailbox. |
Rev 37: see §Users and Access > Multi-Tenancy / Org Hierarchy; AD #35 | 2026-09-24 |
| OQ-53 | §Users and Access > Per-Role Capability Matrix | Recommended Tea Room and Executive role contents | Raised with Rev 21 / AD #30 and restated at Rev 24 and Rev 29: (a) the permission-level semantics (read-only / edit / AI-use / execute) and how the Managing Director’s approval right is expressed; (b) which permissions the recommended Tea Room and Executive / Management roles hold, including whether the Tea Room can view leads and deals so it can be brought onto one by a share. | Resolved — not required (Tharinda, 2026-09-24). (a) is superseded by AD #34: view / manage per entity plus a few special keys, no category axis; the Managing Director’s approval is the special key quote.approve_sub_floor, given to that role by configuration only. (b) is not needed: in the lead’s words, “we dont need to worry about exact contents for any role since it is configurable … we dont even know if Tea Room / Executive role will be there, whether that has a different name”. Empire’s administrator builds every role (AD #34); the role tables in client/roles-and-permissions.md stay as optional advice for that, and local sample roles give developers something to test with. |
Rev 37: see §Users and Access > Per-Role Capability Matrix; AD #34 | 2026-09-24 |
| OQ-6 | §Functional Surface > F4 | Migration volume | How many total records need migration (leads, contacts, orgs, deals, notes, activities)? Sizes the migration effort (weekend job vs full script with rollback). Reclassified 2026-07-22 — read record counts directly from the Pipedrive export during the audit. Leads were counted on 2026-09-25 (692, all not archived); organisations, contacts, notes and activities were still to count. | Counted from the Pipedrive exports received 2026-09-28 (all three re-exported with every hidden column ticked): 692 leads, 4,235 organisations, 4,977 people. Notes and activities no longer need counting, because they are not migrated (Rev 44), and deals are not migrated (Rev 32). About 9,900 records in all — one import job, not a weekend project. Column-level findings are in .forge/discovery/source-schemas/ and the F4-a spec. |
Rev 44: see §Functional Surface > F4 | 2026-09-28 |
| OQ-5 | §Domain Model / §Functional Surface > F11 | Lead vs deal distinction | Do you use both leads and deals in Pipedrive today, and what is the distinction? Confirms the lead → opportunity boundary and the first screen a rep sees. Reclassified 2026-07-22 — directly observable from Leads-inbox usage and conversion patterns in the Pipedrive audit. | Both are used (Dinil, 2026-10-05, from the Pipedrive audit): Empire keeps unqualified inquiries as leads in Pipedrive’s Leads Inbox (692 in the 2026-09-28 export, OQ-6) and works the inquiries a rep pursues as deals on a pipeline. The boundary the PRD already models stands: a lead lives in the triage inbox, and becomes a deal only when a person converts it (F11-e). | Rev 51: see §Domain Model › Key Entities › Lead | 2026-10-05 |
| OQ-60 | §Functional Surface > F36 / F20 | Same-day generic email from the capture app | Empire (2026-09-09): after meeting someone at the stand, send them a generic templated email the same day. Decide the sender (the rep’s own mailbox through Graph, or a generic address via SES), per lead at capture time or an end-of-day list, and the approval shape (HITL), with an idempotent send (AD #13). Raised 2026-09-11 (PRD Rev 23); asked of Empire as Q15 on 2026-09-27, unanswered. | Closed as not needed (lead, 2026-10-11): the feature it served, F36-f, is dropped from V1. The same message can go to a show’s contacts as one bulk email to a saved contact list (F37-b, Phase 2), or a rep emails a lead from the CRM (F20). Q15 is withdrawn from the client question list. | Rev 63: see §Out of Scope, §Functional Surface > F36 | 2026-10-11 |
| OQ-69 | §Functional Surface > F20 | Must reps attach a new file to an email at launch? | Raised 2026-10-12 by the F20-c spec (Q-2), when the draft left attaching a new file to an outgoing email out of V1 although Pipedrive’s composer allows it; the question was to go to Empire with the default “not in V1”. | Withdrawn the same day, never sent to Empire (Tharinda, 2026-10-12): attaching a new file is built in Phase 1 as its own feature, F20-h (Email attachments), on F20-c’s dialog and send path, so the features waiting on F20-c do not wait for it. | Rev 68: see §Functional Surface > F20 | 2026-10-12 |
| OQ-32 | §Domain Model / §Functional Surface > F26 | Bulk price granularity | Is a bulk price set per product/SKU, per product × grade, per product × destination market, or per product × grade × market — or negotiated ad-hoc per deal? Fixes the price key behind the F26 catalog’s price-history model. Raised 2026-07-17. | Withdrawn, never sent to Empire: removed from the signals file on 2026-07-22 without a history row; the same question was asked again as OQ-42, which holds the answer. | Rev 69: withdrawn | 2026-07-22 |
| OQ-34 | §Domain Model / §Functional Surface > F26 | Pricing currency | Are bulk export prices quoted in USD (with LKR auction cost as an internal cost basis), and does the catalog need to store both? Raised 2026-07-17. | Withdrawn, never sent to Empire: removed on 2026-07-17, when the pricing questions were cut to those blocking the F26 schema, without a history row; the same question was asked again as OQ-43, which holds the answer. | Rev 69: withdrawn | 2026-07-17 |
| OQ-35 | §Functional Surface > F26 | Weekly price-update ownership | Which role updates bulk prices week to week, on what cadence, and from what input (broker market report, negotiation, gut feel)? Confirms the weekly bulk-price-update workflow and whether an auction-reference feed is worth building later. Raised 2026-07-17. | Withdrawn, never sent to Empire: removed on 2026-07-17, when the pricing questions were cut to those blocking the F26 schema, without a history row; the same question was asked again as OQ-39, which holds the answer. | Rev 69: withdrawn | 2026-07-17 |
| OQ-36 | §Functional Surface > F26 / F17 | Price-history retention | How far back must bulk price history be queryable, and is retroactive correction of a past price ever needed? Sets the history-retention/audit requirement behind the F26 price-trend view and F17 analytics. Raised 2026-07-17. | Withdrawn, never sent to Empire: removed on 2026-07-17, when the pricing questions were cut to those blocking the F26 schema, without a history row, and not asked again. Prices are effective-dated rows that are never updated in place (AD #14). | Rev 69: withdrawn | 2026-07-17 |
| OQ-62 | §In Scope > F37 / §Functional Surface > F20, F26 / §NFR > Marketing consent | Bulk email — consent, unsubscribe, sender identity and audience | Raised 2026-09-16 with PRD Rev 28, for one message sent to many contacts (F37-b; the weekly price broadcast, F37-a, was clear of it). (a) Consent and unsubscribe: does a contact carry an opt-in or marketing-consent state, who sets it, and what is the unsubscribe mechanism (a link in every newsletter writing back a suppression flag was the default)? (b) Suppression: bounces and complaints must stop being sent to, so where is that state held? © Sender identity: a rep’s own Microsoft 365 mailbox through Graph (throttling, the rep’s own reputation, every reply in one inbox) or a shared address through SES (deliverability, reply routing, Empire’s SPF/DKIM, and a permission key the catalog lacked, AD #34)? (d) Audience: who is a newsletter addressed to, as a simple, statable selection with no segmentation builder (F28)? Empire’s request of 2026-10-09 answered (d) in part: a saved list filtered by market, customer type, tag or field (F32). (a)–© were left out of the reply sent that day, and were never sent. | Decided by the delivery team, not sent to Empire (the lead’s rule of 2026-11-13, No questions to Empire before UAT; the lead chose each option in F37-b’s spec authoring, 2026-11-19). © Each email goes from the mailbox of the person who confirms the send, through Graph, one email per contact, at most 20 a minute as F37-a does and at most 5,000 recipients in any 24 hours counted with the sender’s price broadcasts, so replies reach that person; a shared address through SES is not built. (a) Every contact on the list receives it unless they have unsubscribed, with no opt-in tick, because Empire’s contacts are business contacts; every bulk email carries Empire’s postal address and an unsubscribe link that needs no sign-in and opens a page with one Unsubscribe button. (b) Bounce reports in the sender’s mailbox are read: a bounced address shows not delivered, delivered is the number sent less the number bounced, and an unsubscribed or bounced address is left out of later bulk sends until someone who can edit the contact clears it at the contact’s request, audited; complaints are not handled, because Microsoft Graph gives a mailbox no complaint reports. (d) The audience is a saved contact list (F32-d), the contacts in the sender’s reach. Empire is asked to confirm the choices at the Phase 1 hand-over (owed.md); a change it asks for is made as ordinary work. Detail: F37-b spec. |
Rev 103: see §In Scope > F37, §NFR > Marketing consent | 2026-11-19 |
Revisions
One
### Rev N — YYYY-MM-DDheading per revision, newest at the bottom. Body is free-form prose: what changed, why, and (if applicable) which OQ this resolved.
Rev 1 — 2026-07-17
- Changed: §Domain Model (Product entity) and §Functional Surface > F26 updated to record that bulk tea pricing is price-sensitive, changes weekly, needs price history and validity/expiry, and that “region-linked” pricing means priced per destination market (not the tea’s growing origin). Added Risk R-17 (bulk-pricing model unconfirmed) to §Risks.
- Why: The 2026-07-16 client meeting surfaced bulk pricing as the priority new requirement post-Gate-1.
bulk-tea-pricing.md(R&D research) and a client-facing prototype (.forge/design/prototypes/crm/) explored the domain and a candidate CRM data model; Dinil confirmed directly that “region” means destination market, not growing origin. - Resolves: the “what does region mean” question — logged directly here as OQ-31 rather than passing through
project-prd-signals.md, since it was raised and answered within the same session. - New open questions raised: OQ-33, OQ-37 added to
project-prd-signals.md— validity window and private-label pricing scope, kept deliberately narrow to the questions that block finalizing the F26 schema. Other candidate questions surfaced by the research (currency, weekly-update ownership, history retention, volume/quality tiering) and the two smaller meeting items (CC/BCC email, lead-source tracking) were considered but not logged as signals. - Approved by: Dinil
Rev 2 — 2026-07-21
- Changed: §Functional Surface > F26 (and Risk R-17) now state that the region-linked, weekly-updated bulk-pricing model is bulk-only — private-label stays quoted bespoke per deal. R-17 narrowed accordingly: its remaining exposure is the validity-window length (OQ-33). F26 cross-references updated to point at the still-open pricing questions (OQ-33 validity window, OQ-38 destination-market list, OQ-39 weekly-price-update ownership).
- Why: While finalizing the client kick-off question set (2026-07-21), private-label pricing scope was assessed as already-settled by the 2026-07-16 meeting (only bulk pricing is price-sensitive), so it was removed from the client questionnaire and resolved by decision rather than asked.
- Resolves: OQ-37 (moved to
## Resolved Open Questionsabove). - Still open: OQ-33 (bulk price validity window) remains a client question. New signals OQ-38 (destination-market list), OQ-39 (weekly-price-update ownership), OQ-40 (pricing-approval authority) and OQ-41 (core Pipedrive reliance) were added to
project-prd-signals.mdin the same session. - Approved by: Dinil
Rev 3 — 2026-07-20
- Changed: Three more confirmed items from the 2026-07-16 client meeting, landed as direct PRD-body edits (no signal needed — none were ambiguous):
- §Functional Surface > F20 — CC/BCC recipient fields added as a confirmed capability on every compose/reply.
- §Functional Surface > F11 — lead source promoted to a first-class, filterable field (Website / Trade Fair / Referral / Email / Other), set on lead creation.
- §Functional Surface > F17 — a leads-by-source dashboard breakdown, fed by F11’s lead-source field.
- §Domain Model > Lead entity — updated to record the first-class lead-source field.
- Why: All three were explicit, unambiguous asks from the 2026-07-16 meeting (Pipedrive parity gap for CC/BCC; “track where leads came from and use it for analytics” for lead source) — verified against Pipedrive’s own documentation before writing (CC exists there; BCC on a manually-composed email does not), then built directly into both the prototype and the PRD body rather than logged as open questions.
- Resolves: n/a — these were never logged as open questions; this Rev exists solely to close the audit-trail gap flagged in PR review (the four edits above landed without a corresponding Rev entry).
- Approved by: Dinil
Rev 4 — 2026-08-02
- Changed: Folded the confirmed answers from the client’s pre-kickoff response round (returned 2026-07-22) into the PRD body:
- §Business Specifics + §Domain Model (Product) + F26 — own-brand goods sell through the pipeline, so Product/Deal carry a
brandattribute (OQ-20). - §Domain Model (Product) + F26 — a bulk price holds until the next weekly auction (a single clock shared with the quote, manual override for exceptions) (OQ-33); the Tea Room team owns the weekly publish (OQ-39), seeded from their online auction system (OQ-46); the published price is FOB Colombo in USD, with CIF/other currencies handled per-quote (OQ-44, OQ-43).
- §Users and Access — four V1 roles confirmed incl. the R&D/tea-taster (OQ-3); Executive read-only planned as a fifth (recorded as assumption OQ-47).
- F4 — migration confirmed: migrate everything, preserve original IDs, AI-triage after import (OQ-21).
- §Integration Points — Microsoft 365 confirmed provider (OQ-24); inbound = individual inboxes +
info@empire.lk(OQ-14); the ERP row updated for the current-plus-new ERP landscape and the start-with-current intent (OQ-2, still an assumption). - §Risks — R-08 (own-brand ambiguity) and R-17 (bulk-pricing validity) closed.
- §Business Specifics + §Domain Model (Product) + F26 — own-brand goods sell through the pipeline, so Product/Deal carry a
- Why: The client returned written answers to the pre-kickoff questionnaire. The confirmed, unambiguous ones are now contract, so the project proposal can be built on an accurate baseline. Ambiguous or unanswered items are recorded as working assumptions in
project-prd-signals.md(◐ partial · ASSUMED), not folded as fact — the 27 July follow-up email covering them is still outstanding. - Resolves: OQ-3, OQ-4, OQ-14, OQ-20, OQ-21, OQ-24, OQ-33, OQ-39, OQ-43, OQ-44, OQ-46 (moved to
## Resolved Open Questionsabove). - Assumptions recorded (pending client confirmation): OQ-9, OQ-10, OQ-11, OQ-16, OQ-38, OQ-40, OQ-42, OQ-45, and new OQ-47 (Executive read-only role). Highest-impact: OQ-42 (buyer-specific pricing). Each is surfaced in the proposal’s Assumptions section and confirmed at contract/kickoff.
- ERP + AI hosting — accepted decisions (manager, 2026-07-28), not open client questions: current ERP confirmed as IFS; proceed assuming IFS APIs are available and integrate against it, with a manual-entry fallback if they are not (Risk R-01 → accepted). AI runs on secure cloud infrastructure, no on-prem/self-hosted model, with data-handling approved at the contract stage (Risk R-02 → accepted; the self-hosted-AI spike SP-ARCH-001 stands down unless contract approval fails, and SP-ARCH-002 Exchange/IMAP is moot under Microsoft 365). Neither was asked in the 28 July closing questionnaire. OQ-2 and OQ-15 stay
◐ partialonly for the narrow residuals (IFS APIs provided at start; client’s contract-stage sign-off). - Still audit-derivable (not asked in writing): OQ-1, 5, 6, 7, 8, 12, 13, 17, 22, 23, 25, 26, 27, 28, 29, 30, 41 — confirmed during the Pipedrive audit.
- Approved by: Dinil
Rev 5 — 2026-08-02
- Changed: Reconciled the CRM→ERP scope with Architecture Decision #10. Rev 4 had described V1 as integrating against IFS (assuming APIs, manual fallback), which contradicted AD #10 (manual handoff in V1, no live integration until OQ-2 resolves) and the §Deferred list. §Business Specifics, §Integration Points, and Risk R-01 now state: V1 ships the manual handoff document on the
ErpHandoffServiceseam; automated integration against IFS is a separately-priced follow-on offered once IFS API availability is confirmed. OQ-2 (signals) updated to match. - Why: Flagged in the proposal PR review as a V1 scope contradiction (the proposal §4 deferred automated sync while §9 integrated against IFS, mirroring the PRD split and conflicting with AD #10). Settled 2026-08-02 as a fixed-price scope decision: automated integration is out of the V1 fixed price and offered as a priced follow-on, keeping AD #10 intact.
- Resolves: the ERP-scope contradiction. OQ-2 remains ◐ partial for the narrow follow-on question (whether IFS exposes usable APIs).
- Approved by: Dinil
Rev 6 — 2026-08-03
- Changed: Folded the confirmed answers from the 2026-08-03 requirements-clarification meeting (Praveen, Kavindu) into the PRD body:
- §Domain Model — added the Buying Standard entity (the client-specific priced unit with a unique running number, OQ-42); the Product entity now holds cost for gross-profit computation, records no volume tiers (OQ-45) and market-driven pricing (OQ-38); the Quote entity gains the gross-profit-% gate with Managing Director sub-floor approval (OQ-40) and a payment-terms field (OQ-10).
- §Users and Access — added the distinct Tea Room role (pricing team), confirmed the Executive read-only role, and recorded the gross-profit-% floor + MD approval model and the granular permission levels (read-only / edit / AI-use / execute) (OQ-40, OQ-47, OQ-39).
- F6 — typical sample rounds (1–2 match, 6–7 new standard, < 10) (OQ-11).
- F7 — approval gate rewritten to the gross-profit-% floor + MD sub-floor approval (OQ-40).
- F26 — bulk pricing rewritten to the client-specific Buying Standard model, no volume tiers, market-driven, cost held for GP%; the tea-auction public API noted as a confirmed-feasible follow-on (OQ-42, OQ-38, OQ-45).
- §Integration Points — added the tea-auction public API (follow-on).
- Why: The 2026-08-03 client meeting answered the closing pre-kickoff questions. Confirmed items are now contract; the pricing model in particular was reshaped — the priced unit is a client-specific standard, not a generic product × market table.
- Resolves: OQ-10, OQ-11, OQ-38, OQ-40, OQ-42, OQ-45, OQ-47 (moved to
## Resolved Open Questionsabove). - Still open: OQ-9 (product-held certifications) — storage being clarified (client action items: Kavi Amali / QC). The tea-auction feed is a scoped follow-on, not V1. OQ-15/OQ-16 (AI) and OQ-2 (ERP follow-on) remain as previously recorded.
- Approved by: Dinil
Rev 7 — 2026-08-06
- Changed: Narrowed the email-provider scope in five places to match the issued project proposal:
- §Scope and Boundaries > In Scope (V1) > F20 — the summary bullet’s provider list “(Gmail/Outlook/Exchange/IMAP)” → Microsoft 365 via Microsoft Graph, the only provider in V1 scope.
- §Functional Surface > F20 — same narrowing in the feature paragraph, with the out-of-scope providers and the change-request route stated explicitly.
- §Integration Points — the Microsoft 365 row no longer says the other providers are “supported but not in use”; it states Graph is the sole in-scope provider and the others are change-request territory behind the
MailboxProviderseam. - §Constraints (email/calendar/contact connectivity) — “must integrate Gmail/Outlook/Exchange/IMAP” → must integrate Microsoft 365.
- §Risks > R-12 — the multi-provider component of the risk is retired (
SP-ARCH-002stands down). R-12 stays open (reduced), now scoped to what actually remains: sync/template/tracking build size, SPF/DKIM deliverability, and silent sync failure when a ~3-day Graph subscription lapses. - The OQ-24 row in
## Resolved Open Questionsis annotated as superseded on this point (its resolution parenthetical said the others “remain supported by the build”). The historical resolution text itself is preserved — this file is append-only — with a forward pointer to this revision.
- Why: Found on 2026-08-06 while refreshing
architecture.mdagainst the PRD. The PRD implied a four-provider email build in two places, butproject-proposal.md— already issued to the client — names Microsoft 365 exclusively in six places (§4 scope, the §9 diagram and narrative, §Client Obligations, and the Appendix A third-party-services and cost tables, the last specifying “synchronisation via Microsoft Graph”). Gmail, Exchange and IMAP appear nowhere in it and were never quoted. Internal documents were claiming more scope than was sold — the class of drift that surfaces as a dispute at UAT. - Impact: No change to what will be delivered (every Empire user is on Microsoft 365, so the built system is unaffected). It removes ~3 provider integrations of unfunded implied scope, and confirms SP-ARCH-002 correctly stands down.
architecture.mdRev 1b carries the matching change and adds theMailboxProviderseam so a future provider stays a cheap addition. - Resolves: the PRD-vs-proposal email-scope contradiction. OQ-24 stays resolved; only its scope parenthetical was wrong.
- Approved by: Dinil, 2026-08-06 — confirmed at the Gate 2 Run 2 verdict.
Rev 8 — 2026-08-18
- Changed: Three stale-fact corrections in the PRD body, none altering requirements or scope:
- F21 / F22 (In-Scope bullets) — “Google/Microsoft/Outlook calendars” and “a connected Google/Microsoft account” → Microsoft 365 via Graph only. Rev 7 narrowed the provider scope in five places but missed these two bullets; they now match AD #16 and the issued proposal.
- §NFR Observability — the AI-cost parenthetical “negligible — ~$7/yr projected” (a June research figure that pre-dated drafts and daily briefings entering scope) → the committed
proposal § A.3.2numbers: ≈ $25–30/month expected, budget $50, billing alert at $150, verified against official provider pricing on 2026-08-18.
- Why: Surfaced during the 18 Aug docs review — both are the L-001 drift class (internal doc claiming something different from the issued proposal). The $7/yr figure had already propagated into an onboarding guide and been queried by the team.
- Resolves: n/a — no OQ; stale-fact corrections only.
- Approved by: pending — raised by Tharinda; to be confirmed by Dinil at next review.
- Addendum 2 (same day, decision wave): §Phasing gains the delivery-consolidation note — Phase 1 = P0 + P1a, Phase 2 = P1b + P1c, a client UAT round after each (scope per bucket unchanged; the enhancements bucket relabelled Post-V1 to keep “Phase 2” unambiguous) — and §NFR Security’s AI-data-handling bullet now records the Bedrock-in-account serving decision (AD #19) and points the approval deadline at the dependencies document. Approval rides with Rev 8 — same reviewer, same class.
- Addendum (same day, evening docs sweep): five further stale-fact corrections of the same class, none altering requirements or scope — F21/F22 Functional-Surface paragraphs still said Google/Microsoft (the In-Scope bullets were fixed in the morning pass but the fuller feature descriptions were missed — the exact miss-class Rev 8 exists to fix) → Microsoft 365 via Graph only; §NFR Security AI-data-handling restated from “unresolved binary decision” to accepted-2026-07-28 with contract-stage formal sign-off residual (matches R-02 in the same file); §Constraints Tech Stack list of “open” items → settled pointer at AD #1–#7/#16; §Constraints Deployment “provider/region open” → AWS settled (AD #4), region-only open (AD-O-9); F5 role list widened from four roles to all six per §Users. Header status/date also corrected (was “draft / 2026-08-03”). Approval rides with Rev 8 — same reviewer, same class.
- Filing note (2026-08-18, evening sweep): the block below — the 2026-07-22 ERP-presumption correction — predates Rev 8 and was misfiled into this entry during consolidation. It is kept in place to preserve the record; its closing “Approved by: Dinil” belongs to that block alone, and Rev 8’s own approval status is the “pending” line above.
- Changed: §Business Specifics ERP line updated to state Empire runs a current ERP today in addition to the new ERP it’s adopting — both now understood to exist, rather than only “a new ERP” coming. The unconfirmed specifics (which current ERP, either’s API, timeline, integration sequencing) are unchanged as open. OQ-2 in
project-prd-signals.mdreworded to a three-part question covering both ERPs and the integrate-current-first-vs-hold-at-manual-handoff decision. - Why: While preparing the client kick-off question set (2026-07-22), the ERP question was found to have presumed only one (“the new”) ERP existed, when Empire actually runs a current ERP and has a new one incoming — a correction that needed to land in the PRD body, not just the question wording.
- Resolves: n/a — OQ-2 remains open; this Rev only corrects the presumption baked into the PRD body and the question, it does not answer either.
- Approved by: Dinil
Rev 9 — 2026-08-18 (evening sweep)
- Changed: Alignment sweep after the 18–19 Aug decision wave — the PRD had missed the pass every other document received. Four clusters:
- AI serving facts: §Integration Points row retitled Amazon Bedrock (Claude) — in-account (Singapore), degrade-to-manual, Altrium-held evaluation keys as the only non-Bedrock call path; §Constraints Tech Stack provider line and §NFR Security AI-data-handling bullet restated Bedrock-only with IAM-role auth (no AI API key exists in the deployment); §Constraints Deployment names Singapore
ap-southeast-1(AD-O-9 closed) and replaces dev-doubles-as-UAT with the temporary-per-round UAT model (dev is build-only). - Resolved-elsewhere decisions un-gated in place: OQ-20’s brand attribute no longer reads as gated in F26 / §Deferred / P-20 (resolved yes at Rev 4); the §Deferred automated-inbound-trigger row drops the non-Graph channel list (OQ-14 settled the channels).
- Domain model: new MailboxConnection entity (owner nullable — per-user, or the shared
info@inbox as an unowned connection with its own assignment rule and rate bucket); User and Email relationships repointed to it; F20’s body states the shared-inbox connection explicitly. - Phasing: F18 moved P1b → P1a — the issued proposal names stale/dormant alerts in 1a explicitly (§7), so the client-facing timeline was right and the internal map wrong (L-001); F16 moved out of the Post-V1 bucket into P1b, scheduled last, as the first descope lever (it is in-scope V1); Post-V1 row now also names F35; the feature-paragraph range reads F1–F26 + F31–F34; Gate-table anchors use GitHub’s double-hyphen slugs; §Constraints “settled at Gate 2” now points at the AD table instead of enumerating a stale range (the L-002 pattern that let AD #19/#20 slip).
- AI serving facts: §Integration Points row retitled Amazon Bedrock (Claude) — in-account (Singapore), degrade-to-manual, Altrium-held evaluation keys as the only non-Bedrock call path; §Constraints Tech Stack provider line and §NFR Security AI-data-handling bullet restated Bedrock-only with IAM-role auth (no AI API key exists in the deployment); §Constraints Deployment names Singapore
- Why: Same-day (18 Aug, evening) fresh-perspective repo sweep with four parallel reviewers. Same L-001/L-002 drift classes as Rev 7/Rev 8, this time concentrated in the PRD.
- Resolves: n/a — no OQ; consistency corrections plus one sequencing correction (F18) settled by the proposal.
- Approved by: pending — raised by Tharinda; to be confirmed by Dinil at next review.
Rev 10 — 2026-08-18 (late evening)
- Changed: two stack facts in §Constraints and one explanatory addition to F20, following the same-evening team decisions:
- Frontend framework: TypeScript + React SPA on Vite — Next.js/SSR dropped (AD #21: internal authenticated app, no SEO need); both frontends statically hosted on S3 + CloudFront (AD #22), the backend remaining the only container service.
- F20 now states the mailbox model in the team’s own reference frame: per-user connections plus the shared inquiries inbox is the standard CRM shape and exactly how Pipedrive works today, with a pointer to the
m365-mailbox-integration.mdresearch note.
- Why: framework/hosting decided by the team 18 Aug (evening); the F20 note answers “how do CRMs actually do email?” in the document a spec author reads first.
- Resolves: n/a — no OQ; the framework change is recorded as AD #21/#22 in
CLAUDE.md. - Approved by: pending — raised by Tharinda; to be confirmed by Dinil at next review.
Rev 11 — 2026-08-18 (night)
- Changed: AWS region in §Integration Points and §Constraints — Singapore → US East (
us-east-1), decided on cost the same night AD-O-9 first closed. §Constraints > Deployment now states the accepted trade-off (~200ms API round-trip from Sri Lanka; static frontends serve from CloudFront edge; buyer data stays inside the client’s own account in either region) instead of the region-as-data-protection framing. - Why: team decision — cost saving preferred over regional latency;
us-east-1is also Bedrock’s home region, simplifying model availability. - Resolves: n/a — AD-O-9 revision, recorded in
architecture.md § Open Decisionsand Rev 9. - Approved by: pending — raised by Tharinda; to be confirmed by Dinil at next review.
Rev 12 — 2026-08-19
- Changed: §NFR Security AI-data-handling bullet and §Risks R-02 — the contract-stage “formal approval of the AI data-handling document” residual is closed: the data-handling arrangements are those stated in the accepted proposal (Appendix A.6), signed off with its commercial acceptance on 2026-08-16. No separate approval document or gate exists; the
AIProviderseam remains as an architectural safeguard. The client pack was aligned the same day — the client PRD no longer announces an approval step, and the dependencies document now lists only the seven development-blocking inputs, front-loaded to weeks 1–2. - Why: the proposal the client accepted already states the AI arrangements; running a second approval route re-opens a settled question (team decision, 2026-08-19).
- Resolves: OQ-15 residual — row moved to Resolved Open Questions.
- Approved by: Tharinda, 2026-08-19.
Rev 13 — 2026-08-19
- Changed: the trade-show capture app is now in the PRD as F36. §In Scope (V1) gains the theme group Trade-show capture companion (own application) with the F36 bullet (P1); §Functional Surface gains the F36 subsection (own-repo/URL PWA per AD #20/#22, offline-tolerant capture, AI card read, research/scoring on the
AIProviderseam, SES email notifications, feeds F11, parallel stream weeks 8–13, capture-and-feed descope lever); §Phasing gains a Capture stream (parallel) row. §Deferred: the Trade-fair bulk lead intake workflow row is removed — per-lead floor capture is now F36 (V1) and organizer bulk lists were already V1 via F23 import + F11 triage; the Native mobile apps deferred row and the §Out of Scope native-apps bullet now name the F36 PWA as V1’s mobile surface. - Why: the accepted proposal (§4) includes the companion app; scope was settled at full (Option C) on 2026-08-18 (AD #20) and
architecture.md § Trade-Show Capture Applanded the same day — this Rev closes the record-level gap (the PRD previously deferred exactly this surface). - Resolves: OQ-48 — row moved to Resolved Open Questions.
- Approved by: scope — team decision 2026-08-18 (AD #20, matching proposal §4); this drafting rides PR #11 review (Dinil, the OQ owner).
Rev 14 — 2026-08-20
- Changed: §Functional Surface > F5 — the paragraph now names the Administrator user-management surface (create user + assign role at creation, change role later, deactivate/offboard with record reassignment and mailbox release, credential reset, across both login paths per AD #5) and states that its mechanics are undecided and gate the F5 spec. Also records that the prototype carries only a four-role demo switcher, not a management screen.
- Why: the capability matrix already granted the authority (
Users & roles= AdminC,R,U,D,Cfg) andproposal §9commits to Administrator training on “users and roles”, but no document described the surface that implements it — a client note (2026-08-19) surfaced the gap. This adds no new scope: it describes an already-granted, already-costed capability and points at the open questions instead of inventing answers. - Impact: F5’s spec must resolve OQ-51 before drafting requirements; Foundation slice 6 is separately blocked by OQ-53.
- Raises: OQ-51 (admin provisioning & offboarding mechanics, incl. role-assignment rules), OQ-52 (shared
info@inbox UI/visibility/assignment), OQ-53 (permission-level semantics + the two missing capability-matrix role columns — settle before Foundation slice 6). OQ-53 also supersedes the previously untracked Gate 2 Run 2 fail item by giving it an owner. - Approved by: Dinil, 2026-08-20 (directed this session).
Rev 15 — 2026-08-20
- Changed: §Deferred (Post-V1) gains a row for the automated tea-auction price feed — the one item in the accepted proposal’s deferred list that had no row in the PRD’s canonical deferred table. It was already recorded in §Integration Points as a follow-on and carries the
AuctionPriceProviderseam inarchitecture.md, so this is bookkeeping, not a scope change. The row also states the distinction the two sections blur: the price feed is deferred, auction bidding is permanently out of scope. - Why: a scope review (2026-08-20) compared the proposal’s deferred list against §Deferred row by row. Everything matched except the auction feed, which was discoverable only by reading §Integration Points — so anyone answering “is the auction feed in V1?” from the deferred table alone would not have found it.
- Impact: none on scope or estimate — V1 weekly pricing was already manual, and no feature changes. The client-facing PRD gained a matching plain-language entry in its not-included list (the proposal names the feed; that document had not mentioned it at all).
- Not changed, deliberately: Native mobile apps remains listed in both §Out of Scope and §Deferred. The §Out of Scope entry ends with an explicit “(See Deferred.)” cross-reference, so the dual listing is intentional — not built in this project, with the post-V1 trigger recorded in Deferred.
- Approved by: Dinil, 2026-08-20 (directed this session).
Rev 16 — 2026-08-20
- Changed: F27 (chatbot/live chat), F28 (campaigns), F29 (enrichment), F30 (LinkedIn) and F35 (sales goals) are now deferred from V1 outright — no longer audit-contingent. Every “promote to V1 if the Pipedrive audit shows…” clause is removed. Updated across §Deferred (five rows), §Reference-System Classification (P-23–P-26, P-33), the §Out of Scope marketing parenthetical, §Integration Points (LinkedIn, enrichment provider), the two marketing-compliance notes, and §Risks R-13.
features.mdand OQ-23/OQ-29 in signals updated to match; the client-facing PRD states the same position in plain language. - Why: developer decision, 2026-08-20. The audit-contingent framing left V1 scope genuinely undetermined until week 1 against a low-contingency schedule (R-ARCH-001), and an add-on promoted late is the least affordable kind of scope growth. Deferring them regardless makes V1 a fixed target now.
- Consistent with the proposal:
project-proposal.md §4already places all five in “Deferred (post-V1) … not part of this proposal’s timeline or price”; only the “pending the Pipedrive audit” qualifier is being closed off. Nothing that was sold is being withdrawn. - Impact: no feature or estimate change (none were in the V1 plan). R-13 is re-scoped rather than closed — the over-building risk is gone; what remains is an expectation risk against the “nothing lost on migration” promise if the audit shows real reliance, Campaigns being the likeliest. Mitigation is to run the audit early and raise any gap with the client then, as a change-request conversation, rather than letting it surface at UAT.
- OQ-23 / OQ-29 stay open with a narrowed purpose: they no longer decide V1 entry, only post-V1 priority and early warning.
- Client communication owed: Empire holds the pre-change client PRD. This position — and the auction-feed entry from Rev 15 — should be stated when the Pipedrive audit is scheduled.
- Approved by: Dinil, 2026-08-20 (directed this session).
Rev 17 — 2026-08-21
- Changed: §Users and Access > Per-Role Capability Matrix gains a
#### Permission levelssubsection defining the four levels the client asked for (OQ-47) and reconciling them with the matrix’sC/R/U/D/A/Cfgaction axis — the two vocabularies had coexisted since the proposal without ever being joined up. Levels are a second axis over the matrix and are fixed per role, not assigned per user. The section also records the working positions for the matrix’s two unmapped role columns (Tea Room, Executive) and marks them explicitly as awaiting client confirmation. - The load-bearing definition:
AI-usegrants invoking an AI feature and receiving its suggestion, and nothing more;executegrants the outward or committing actions (send buyer email, issue or approve a quote, fire the ERP handoff, publish a price). The two are separable — a user may holdAI-usewithoutexecute, generating a draft they cannot send. This makes the PRD’s human-in-the-loop mandate mechanical rather than procedural: there is no path by which an AI feature performs an outward action, because the permission that performs outward actions is not one the AI layer holds. - Why: OQ-53(a) was a Gate 2 Run 2 fail item with a settle-by of “before Foundation slice 6”. FS-6 builds the
(role, resource, level)model and could not be built from four undefined words. Fixed-per-role was chosen over per-user assignment deliberately, for a materially simpler FS-6 data model at a scale (10–20 users) where per-user grants buy little. - Resolves: OQ-53(a). OQ-53 stays
◐ partial— (b) the Tea Room and Executive columns remain unmapped and are an outstanding client input rather than an internal decision; the Managing Director question (seventh role / flag / grantable right) and the cross-cutting Executive hide-vs-disable rendering rule are also still open. - Impact: FS-6 is substantially unblocked — the level axis is settled and the model core can be built. One model-shape question survives: if Managing Director resolves to a grantable approval right, FS-6 needs a per-user grant alongside the role-fixed levels; if it resolves to a role or a user flag, it does not. F7’s sub-floor gate depends on the same answer.
- Approved by: Tharinda, 2026-08-21 (answered this session).
Rev 18 — 2026-08-21
- Changed: §Users and Access > Per-Role Capability Matrix gains a
#### Named permissions, and the Managing Directorsubsection. Sub-floor quote approval (OQ-40) becomes a named permission —APPROVE_SUB_FLOOR_QUOTE— granted to whichever role Empire decides holds that authority. Recorded as AD #26: an authorization check names the permission it needs, never the role that happens to carry it. F7 asksuser.can(APPROVE_SUB_FLOOR_QUOTE), neveruser.role == ADMIN. - Why: the Managing Director question had three candidate answers (seventh role / user flag / grantable right) and each implied a different FS-6 data model, which was the last thing gating that slice. Naming the permission dissolves the question rather than answering it: who holds it becomes configuration, so Empire may give the MD the Administrator role, a dedicated role, or grant it temporarily to a Sales Manager during an absence — all three are the same code.
- What the design forbids, and why: F7 learning the answer as a role test.
role == ADMINsilently converts every future Administrator into a below-floor-pricing approver, and nothing at the point of granting admin says so. An audit log does not catch this — it records a legitimate approval by a legitimate admin. (An earlier objection, that MD-as-Administrator would make approvals untraceable, was withdrawn: approval records store a user id, so a named MD account is perfectly identifiable. The surviving objection is about the predicate, not the audit trail.) - Also recorded: a note flagging that a single account holding both
Cfgon the margin floor andAPPROVE_SUB_FLOOR_QUOTEcan lower the floor and approve against it — a separation-of-duties point worth raising with Empire when role assignment is confirmed. - Resolves: the Managing Director half of OQ-53. OQ-53 remains
◐ partial, now narrowed to part (b) only — the Tea Room and Executive column mapping, which is an outstanding client input. - Impact: FS-6 is unblocked (
blocked_bycleared). Whichever way Empire assigns the permission, the model is one more entry in(role, resource, level)— no per-user grant table. The cross-cutting hide-vs-disable rendering rule settled the same day indesign/ui/design-system.md §9 > Permission rendering: disable with a reason when the block is contextual, hide when the role can never do it anywhere (the Executive case), never the bare HTMLdisabledattribute. - Approved by: Tharinda, 2026-08-21 (directed this session).
Rev 19 — 2026-09-02
- Changed: §Users and Access > Named permissions, and the Managing Director — the V1 default holder of
APPROVE_SUB_FLOOR_QUOTEmoves from the Administrator role to a dedicatedmanaging_directorrole (the Executive read-only grant set plus the one named permission), held by the Managing Director’s account alone; Administrators and Sales Managers do not hold it. The role table gains the Managing Director as a seeded access profile, explicitly not a seventh client-confirmed role. The “worth raising with Empire” caveat about one Administrator lowering the floor and approving against it is replaced by the separation-of-duties note the new default earns. - Why: the client’s statement of 2026-08-03 (“special approval from the Managing Director”) and the issued proposal’s pricing-approval line (“approved by the Managing Director”) both name the MD and nobody else. The 2026-08-21 default gave the permission to every Administrator for convenience, and the client-facing Q3 and role-access map were putting that widening to Empire for confirmation. A widening of a requirement the client stated and the proposal commits to is not the team’s to propose (L-001: the proposal outranks internal convenience); the default now matches what was asked, and any broadening is Empire’s explicit call.
- Unchanged: AD #26 — the check names the permission, never the role; who holds it remains configuration, which is exactly what makes this a seed change (FS-6 FR-P6, Q-3) rather than a code change.
- Impact: FS-6 spec FR-P6 / AC-8 / Q-3 (seed the
managing_directorrole instead of granting toadministrator);client/open-questions.mdQ3 andclient/roles-and-permissions.md§4.7 / A-25 and its Quotes rows restated to the MD-only plan; F5’s user-management surface must offer the role. Resolves nothing — Q3 and OQ-53 stay open until Empire answers. - Approved by: pending review of harness PR #46 — drafted 2026-09-02 at Tharinda’s direction.
Rev 20 — 2026-09-04
- Changed: §Users and Access / §Functional Surface > F5 — sign-in is via each user’s company Microsoft account only. Google OAuth and the local email/password path are dropped from V1 (AD #28, refining AD #5’s login half; AD #5’s session model — server-side sessions via Spring Session JDBC — stands unchanged). F5’s account-creation option set loses “admin-set temporary password”; credential resets move out of the CRM entirely (they live in Empire’s Microsoft 365). The Gate-2 settled-decisions list’s “dual-path auth” wording updated to match.
- Why: team decision 2026-09-04. OQ-24/AD #16 established Empire is an all-Microsoft-365 shop — every CRM user already holds an M365 account, so the local fallback (AD #5’s stated rationale) protected nobody real while carrying password storage, reset flows, and lockout/anti-enumeration surface. Entra supplies MFA and conditional access. Accepted trade-offs recorded on AD #28: no local break-glass if Entra is unavailable; every user (MD and Executive included) needs an Empire M365 account; the Entra app registration becomes the sole login path’s one hard external dependency.
- Impact: FS-6 spec Rev 1 (FR-A1/FR-U4 removed; FR-A2 Microsoft-only with a profile-scoped
local/teststub login; ACs reworked); OQ-51 narrowed (the temporary-password option retired — the invite / pre-create / auto-provision fork remains open);architecture.mdRev 28;client/client-facing-prd.mdFoundation line,client/open-questions.mdQ2 andclient/roles-and-permissions.mdsign-in rows restated to Microsoft-only; features.md F5/FS-6 rows. Resolves nothing client-facing — Q2 and OQ-51 stay open. - Approved by: Dinil (relaying the team decision, 2026-09-04); spec-side lead re-ack owed from Tharinda (Rev 1 shrinks an approved scope).
Rev 21 — 2026-09-07
- Changed: §Users and Access — role management becomes fully configurable, and a reporting hierarchy enters scope (AD #30, refining AD #15). The six confirmed roles become the built-in role templates; the Per-Role Capability Matrix is restated as the default template set, not fixed truth. § Permission levels: the Rev 17 statement “Levels are fixed per role — an Administrator does not set them per user” is reversed — the level becomes each catalog permission’s category (read / edit / ai-use / execute), a property of the permission key rather than of the role or the grant; the grant tuple is
(role, permission_key, scope)over a code-declared catalog (<resource>.<action>keys with category / label / description / UI-group / scope-aware metadata). § Multi-Tenancy / Org Hierarchy: access is role + record ownership + the reporting hierarchy (app_user.manager_id, nullable self-reference); scope isown/team/all, set per permission; multiple roles union to the most permissive grant. § Deferred: the multi-office row now distinguishes a reporting hierarchy (in scope) from multi-office data partitioning (still deferred); the visibility-groups row re-scoped to ad-hoc groups that cut across the reporting line. § Core EntitiesUsergains the manager self-reference and M:N Role. F5’s functional-surface paragraph grows: role editor, templates (copy-on-create), manager assignment, per-permission scope, superadmin + last-admin guard rails, and an effective-permissions viewer. - Why: Empire asked for a fully configurable role-management system (2026-09-07). A fixed matrix with per-role levels cannot express it, and retrofitting configurability onto a level-column grant model after FS-6 ships is exactly the “designed once in the auth Foundation slice” failure AD #15 was written to avoid. The one thing Rev 17 made structural — the AI layer never holds an outward-acting permission — is preserved by keeping the level as catalog metadata, where a role editor can surface it (“this role holds N outward-acting permissions”) rather than bury it in a grant column.
- Supersedes, does not edit: Rev 17’s OQ-53(a) resolution. That entry stands as the record of what was decided on 2026-08-21 and why (a materially simpler FS-6 data model at 10–20 users); this revision reverses it on new client information.
- Unchanged: AD #26 — every check names the permission it needs, never the role;
APPROVE_SUB_FLOOR_QUOTEstays a named permission on the seededmanaging_directortemplate (Rev 19). The six roles as confirmed on 2026-08-03, the pricing floor rule and the AI never-auto-acts invariant all stand. - Signals: OQ-51 widened to role authoring and manager assignment; OQ-53(b) reframed — the Tea Room / Executive columns are now template contents; OQ-56 raised — four client-facing hierarchy policy questions: (a) records with no owner (the shared
info@inbox’s leads) underownscope — visible toteam/allholders, or an explicit Unassigned queue; (b) deactivating a manager who has reports — auto-reparent or block until reassigned; © isteamrecursive or direct-reports only; (d) may a manager edit a report’s records, or only view them. - Commercial: treated as a V1 requirement — no change request (Dinil’s call, 2026-09-07). Proposal §3.2 quotes six named roles with fixed granular levels, and this PRD had deferred structural hierarchy; per L-001 /
.claude/rules/source-trust.mdthe divergence is stated rather than silently absorbed — as a post-issue note onclient/project-proposal.md’s header (the convention for corrections to the issued text). The client-facing pack is restated in the same change:client/client-facing-prd.md’s access paragraph,client/open-questions.mdQ1 (re-asked as “what should each starting template hold, and who reports to whom”, carrying OQ-56’s four questions in plain language with their defaults) andclient/dependencies.mdrow 3.client/roles-and-permissions.mdis rewritten in the same change. - Impact: FS-6 spec Rev 3 owed (FR-U6
core.role_grantdrops the level column and gainsteamscope; FR-P1’s level enum becomes catalog metadata; FR-A5’s/api/megrantsarray becomes a flat permission set with scopes; templates + superadmin + orphan-grant check + last-admin protection enter scope). FS-3 plan Decision 13’sPermissionDecisionunion survives unchanged — it is agnostic to how the backend computes a decision; only FS-6’s half of the seam moves (noted to Biseni, whose FS-3 worktree is in flight).architecture.mdRev 30;features.mdF5 row; added scope recorded against R-ARCH-001;client/roles-and-permissions.mdrewritten for the configurable model (revised 2026-09-07 — A-1…A-33; its §2 puts OQ-56’s four reporting-line questions to Empire with defaults; the Sales Manager template defaults toteamscope, with “whole sales organisation” as the A-13 alternative). Design-system § Permission rendering andentity-map.mdRole/Permission rows annotated. Catalog not enumerated — that is FS-6 Rev 3’s and F5’s work. - Approved by: Tharinda, 2026-09-07 (lead sign-off — required because this reverses Rev 17, which Tharinda approved, and refines DO-NOT-REVERSE AD #15). Drafted by Dinil the same day.
Rev 22 — 2026-09-10
- Resolved: OQ-50. The generic-layer reuse-rights question is closed as moot. It was assumed yes on 2026-08-06 and drove the core/domain separation; AD #29 (2026-09-07) stopped extending the split once the reuse was no longer expected, and AD #31 (2026-09-09, FS-14) collapsed it outright — one Gradle module, one
publicschema, feature-first packages. No reusable generic layer results from this engagement, so the rights question has nothing to attach to. - Why: the row had deferred formal closure (“a follow-up”) pending the three-write procedure; the FS-14 harness cascade discharges it.
- PRD body: no change — OQ-50 was a §Constraints / commercial signal and was never folded into the live contract (zero mentions of it in
project-prd.md), so there is nothing to update there. The resolution lives in the AD ledger (AD #29 / #31) andarchitecture.md § Core / Domain Separation. - Resolves: OQ-50.
Rev 23 — 2026-09-11
- Changed — bulk pricing is one price per standard per week. §Domain Model (Buying Standard row) and §F26: the price no longer varies by customer or destination market; market stays a customer attribute and a quote field, never a price key. Revises the Rev 6 answer to OQ-38 (its history row carries the pointer). §F26 also gains the weekly Excel workbook as the price-list source, taken in as draft prices with duplicate standard numbers flagged (OQ-59); the ~35,000 issued / ~1,450 active scale and the searchable-list rule; the per-account send this week’s prices action; and its pointer moves from
bulk-tea-pricing.mdtoforge-et-be/docs/data-model.md§2, whose §2.2 revision is recorded as owed (Tharinda, AD #31). - Changed — scope exclusions. §Out of Scope gains bulk/group email, newsletters and WhatsApp integration (team decision, Tharinda, 2026-09-11; confirmed by Dinil; stated to the client, not requested by them). §F20 and P-13 drop group email; R-11’s mitigation is restated.
- Changed — F36. §In Scope and §F36: badge QR confirmed as needed, a scan-time existing-client check (needs an account/contact lookup from the app — architecture Rev 33), a same-day generic follow-up email (OQ-60); campaign planning is not a new ask. Recorded against R-ARCH-001.
- Changed — smaller answers from the 9 September meeting. §F1 and §F11: tags/labels on accounts, contacts and leads (Pipedrive labels parity). §F5: a deal or quote must always have an active owner.
- Why: the 9 September 2026 requirements follow-up (
discovery/meeting-notes/2026-09-09-requirements-followup.md) and Empire’s weekly price workbook shared on 2026-09-11 — one row per standard per week; its cross-tab price differences were checked and are data drift and mis-keyed numbers, not a rule. - Signals: OQ-52 and OQ-55 updated; OQ-51 gains the active-owner answer; OQ-57–OQ-60 raised (participants on a lead; owner-initiated sharing; workbook ingestion; same-day show email).
- Revises: OQ-38 (Rev 6 answer).
Rev 24 — 2026-09-11
- Changed — role management simplified (AD #32, refining AD #30). §Domain Model User row: a user holds exactly one role (
N:1 Role; the M:N edge and the most-permissive union are gone). §Per-Role Capability Matrix status note: the matrix is the recommended day-one configuration, not a template set — the system ships with the non-deletable system superadmin only and Empire’s Administrator creates each role by hand from the catalog (Altrium supervising at go-live). §Permission levels: three categories — read / edit / execute; theAI-uselevel is removed. System-run AI (scoring, triage, extraction, briefings) is not a permission — results are visible under the record’sread, controlled by the per-feature switch; user-invoked AI assist is implied by the edit/execute permission it serves. The human-in-the-loop invariant is restated: the AI holds no permission and a person commits every output. § Multi-Tenancy: no cross-role union. §F5: no templates UI, one role per user, a block on deleting a role that still has users, the effective-permissions viewer shows the single role; its prototype citation now points at the 2026-09-07 Users & roles screens as a team proposal with three superseded elements. - Why: team decision to remove unwanted complexity — Tharinda, 2026-09-11, confirmed by Dinil. Templates, the multi-role union and a grantable AI level each added machinery nobody had asked for; removing them shrinks FS-6 and F5. Stated to the client (post-issue note on the proposal), not requested by them; no change request.
- Signals: OQ-51 gains the one-role answer and drops the template questions; OQ-53(b) becomes the recommended Tea Room/Executive role contents; OQ-56 wording restated. Client pack restated:
client-facing-prd.mdaccess paragraph,open-questions.mdQ1/Q2,dependencies.mdrow 3,roles-and-permissions.md(revised 2026-09-11, still DRAFT). - Follow-through: FS-6 spec Rev 5 + Rev 6 + the plan revision landed on
docs/FS-6-rev5-ad32(PR #76); a prototype PR for the role screens is owed.
Rev 25 — 2026-09-13
- Changed — AWS ownership/hosting model (AD-O-10 resolved by SOW ET-002). §Non-Functional > Deployment and Hosting: buyer data now stays inside the project’s AWS account — Altrium-owned and operated as Empire’s processor under SOW ET-002 — rather than “the client’s own account”. The managed-cloud-infrastructure SOW has Altrium own and run the AWS organisation/accounts and bill Empire a fixed USD 650/month (all-inclusive of AWS + AI + managed operations; separate from the USD 360/month application support under SOW ET-001); accounts are not transferred to Empire on exit. Reverses the earlier working assumption (Empire opens/hands over the account).
- Why: Empire is a non-technical business and its cloud billing is normally run by its IT provider (Dialog); Empire asked Altrium to run AWS for this project and invoice monthly. The data-governance substance — in-account processing, no separate AI vendor, logical isolation — is unchanged; only the account owner and the billing shape change.
- Cascade:
CLAUDE.mdActive Context + AD-O-10;architecture.mdRev 36 (AD-O-10 Resolved);client/aws-sow.md(new);client/dependencies.md§1.1;client/sow-cost-annexe.mdbilling model;client/client-facing-prd.mddata-governance framing;ai-provider-strategy.md;design/environments-and-identity.md(new); a proposal post-issue note. - No change request: internal ownership/commercial structure is governed by SOW ET-002; L-001 governs the proposal’s scope/price/providers, which are unaffected.
Rev 26 — 2026-09-13
- Changed — migration scope (F4): fresh start on deals. §Functional Surface F4 now imports leads, contacts, and organizations (accounts) (with their notes/activities) + business-critical custom fields — not deals or deal history. Empire chose a fresh start on deals in the new system (2026-09-13); historical deals stay in Pipedrive. The “lose nothing on migration” promise is unaffected (Empire’s informed election; the open lead/account backlog — the thing they need to work — carries over).
- Recorded — delivery phasing intent (pre-Gate-3). Empire’s stated main pain is following up the large cold-lead backlog, so Phase 1 = login, users & roles, leads, accounts, email + AI draft, lead/account migration, lead/account ranking (cold-lead detection), and weekly prices; Phase 2 (tentative) = capture app, sampling, quotes, deals, pipeline. This shifts the capture app’s timing to Phase 2 (still full-scope in V1 — AD #20 scope stands, only sequencing moves). Delivery-planning intent, captured in
CLAUDE.mdActive Context to drive Gate 3; the tracker is not yet phased (Gate 3 not run). - Recorded — WhatsApp is a later phase, not Phase 1 (some customers are WhatsApp-only). Already outside the V1 core (Rev 23); this records the rationale and the phase placement.
- Also recorded (detail lives in FS-6 spec Rev 5 / the
forge-et-bedata model): the permission model is simplified to coarseview/manageper entity + a short sensitive-actions list (within AD #30 catalog latitude, no AD change); the bulk-price key drops itsmarketaxis (already Rev 23) with a standard globally unique and price(standard, week)append-only (AD #14); Product is kept; a share option is a confirmed V1 intent (OQ-58). - Why: the 2026-09-13 working session with Tharinda (products/permissions/phasing confirmations). No change request — internal scope sequencing + a client-elected migration reduction; L-001 governs proposal scope/price/providers, unaffected.
Rev 27 — 2026-09-15
- Changed — the products catalog is removed; the Buying Standard is the sellable record for bulk and private label alike. §In Scope F26 (retitled Buying standards & weekly pricing (deal line items)), F7, F9, F23, F31, F34; §Reference-System P-19/P-20 (P-20 becomes Redesign); §Phasing P1a; §Domain Model — the Product (Catalog item) row is struck and its attributes (tea grade, origin, brand, held certifications, cost, price history) move onto Buying Standard, whose relationships become
N:1 Account,N:N Deals,N:N Certifications,1:N weekly prices(theN:1 Productedge is gone); the Deal, Quote, Certification and Reference-lists rows, the glossary, the Administrator and Tea Room role rows, the capability matrix (Buying Standard replaces Product / Catalog), §F2/§F7/§F9/§F26/§F31/§F34, the two journeys and the success criteria all read standards instead of catalog products. Revises OQ-37 (the bulk-only pricing model — private-label items are standards too) and OQ-20 (the brand attribute moves to the standard). - Added — the standards lifecycle. §F26: standards are created, edited and archived; archiving keeps the price history and removes the standard from quoting and from the weekly round; numbers are never reused, so create always mints a fresh one. A customer receives their standards and this week’s prices through the existing per-account send this week’s prices action (Rev 23) — no new email surface.
- Added — a price-lookup page is intended, not sequenced. §F26: find this week’s price for a standard number, for any role holding the price-view permission; not a Phase-1 priority — sequenced at Gate 3; the prototype’s Pricing screen in its read-only rendering is the shape.
- Why: the team’s requirement change stated 2026-09-15 (Biseni), already applied to the client-approved CRM prototype in harness PR #88 (merged 2026-09-15 — the catalog’s nav entry, screen and modals are gone; the quote picker, the F9 check, global search and the deal type-ahead read standards; the seed carries private-label standards). Rev 26 had recorded Product is kept for now (revisit later) from the 2026-09-13 working session; this revision is that revisit. Lead ack: Tharinda, 2026-09-16 — given as the approval of the FS-15 spec (harness PR #99), whose premise is this revision; it reverses a Rev 26 recording and reaches the shipped backend, and FS-15 is the clean-up.
- Hand-offs (recorded, not done here):
forge-et-be— theproducttable (V3__product.sql), thecatalogpackage (Product,ProductController— FS-10’s worked API example),ListPriceModel, theproduct.*permission keys (ProductPermissions; FS-6 spec Appendix A.2) and theproduct_idFKs onbuying_standardandline_itemindocs/data-model.md§2 all assume the catalog. The data-model revision is Tharinda’s (AD #31); the FS-6 spec Rev and the FS-10 example follow their slice owners (Dinil / Tharinda); the code retirement is scheduled with F26’s spec or earlier. OQ-9’s “product-held certifications” become the standard’s held certifications. - Signals: OQ-61 raised (private-label standards — weekly-priced or quoted per deal?); OQ-27 reworded (the standards list, not a product/SKU list, is the seed and migration source); OQ-9 reworded to the standard.
- Cascade:
CLAUDE.mdActive Context;architecture.mdRev 37;entity-map.mdRev 7;design/prototypes/feature-detail.md(F26/F7/F9/F31 + the drift tables);features.mdF26/F9;design/ui/design-system.mdRev 17;design/prototypes/README.md; the client pack (client-facing-prd.md,open-questions.mdQ5,roles-and-permissions.md§4.3/§5.4/§8,dependencies.md); a proposal post-issue note (the proposal quotes a products catalog with bulk pricing — L-001). - No change request: the proposal’s products catalog with bulk pricing is delivered as the standards book — the same capability as one list instead of two; scope, price and timeline are unaffected (L-001: stated on the proposal header, not silently absorbed).
- Revises: OQ-37 (Rev 2 answer), OQ-20 (Rev 4 — the attribute’s home).
Rev 28 — 2026-09-16
- Changed — the bulk-email exclusion is lifted to a V1 stretch candidate; WhatsApp stays out. §Out of Scope: the combined “bulk/group email, newsletters and WhatsApp integration” bullet (Rev 23) is split. WhatsApp remains excluded, unchanged and for the same reasons. Bulk email is no longer a won’t-build — it becomes F37, recorded in §Deferred (Post-V1) as a stretch candidate: taken into V1 only if the schedule allows, otherwise the first post-V1 item. A second §Out of Scope bullet states the committed position plainly, because the distinction is the load-bearing part: nothing downstream may assume F37 ships, and the committed V1 behaviour — one human-triggered send at a time, weekly prices going out from the per-account send this week’s prices action — is unchanged.
- Added — F37 — Bulk email (weekly price broadcast + newsletters), §Deferred. Two separable capabilities, ranked: (a) weekly price broadcast — one action sending this week’s prices to every client at once, as a batch of per-account emails (each client sees only their own standards, so it is a mail-merge run rather than one message to a list); it repeats F26/F20’s existing per-account action under one confirmation, which makes it the cheaper, lower-risk half and the one to build first. (b) newsletters — one common message to many contacts. Scope discipline if it ships: the minimal send only — no segmentation builder, no campaign analytics beyond F20’s open/link tracking, no template designer beyond F20’s templates.
- Why: Empire confirmed bulk sending as an important requirement and a genuinely useful feature (2026-09-16, recorded by Dinil), naming the two use cases above. Rev 23’s exclusion was a team decision taken on 2026-09-11 without this confirmation, so the ask is recorded rather than declined a second time — while the low priority the client themselves attached to it is recorded just as explicitly, which is why it lands as a candidate and not as scope.
- Trigger: the mid-November checkpoint already held under R-ARCH-001 — the same checkpoint that owns the F36 descope lever. No new review ceremony is created.
- Commercial position — no change request now, one owed if it is committed. The issued proposal does not cover bulk sending (§4 quotes two-way sync, templates and tracking), so taking F37 into V1 is added scope: it needs a change request or a compensating descope under R-ARCH-001. Recording it as a candidate commits nothing and needs neither. Stated on the proposal header as a post-issue note (L-001), which also amends the 11 September note that recorded the exclusion.
- Corrected — a Rev 23 contradiction. §Functional Surface’s email narrative said “No bulk or group email in V1” and then, three sentences later, that “open and link tracking, scheduled send, and group email are included” — the tail was never updated at Rev 23. Rewritten.
- Signals: OQ-62 raised — the four things only Empire (and the sending-domain owner) can settle before F37(b) could be built: per-contact consent/opt-in state and unsubscribe handling, suppression of bounces and complaints, sender identity (a rep’s own mailbox via Graph does not suit a several-hundred-recipient send; a shared address via SES does, and changes deliverability, reply routing and SPF/DKIM), and the audience a newsletter is addressed to. F37(a) is materially clear of all four, which is why it is separable and ranked first. OQ-60 (the same-day capture-app email) is adjacent and stays inside the one-email-per-lead rule either way.
- Cascade:
CLAUDE.mdActive Context;features.md(F37 row + the F20 row’s no-bulk note);design/prototypes/feature-detail.md(F20);discovery/inventories/feature-inventory.mdrow 13;discovery/source-schemas/weekly-price-workbook-schema.md(the sharing line);tracker.yaml(the Rev 23 note); the client pack (client-facing-prd.md,open-questions.md); a proposal post-issue note; and the CRM prototype — a bulk-send affordance on the Pricing screen and a newsletter composer on the Inbox screen, both marked as the stretch they are, plus a new Send a bulk email permission kept separate from Send email from a record (AD #26). - Reverses: the bulk-email half of Rev 23’s §Out of Scope addition. Does not reverse: Rev 23’s WhatsApp exclusion, or Rev 23’s per-account weekly-price action, which stays the committed path and the fallback.
Rev 29 — 2026-09-16
- Changed — record sharing becomes the third visibility path, resolving OQ-58. § Multi-Tenancy / Org Hierarchy previously described access as role + record ownership + the reporting hierarchy. It now carries a third path: the owner of a lead, deal or account, or a manager above them, may share that one record with a named colleague. Empire asked for this on 2026-09-09 and it was flagged a confirmed V1 intent on 2026-09-13; only the one-sentence request existed until now.
- The load-bearing rule — a share widens reach, never authority. The share carries no access level of its own: what the recipient may do is decided entirely by their own role. This was the one recommendation the team overturned during the decision walk (an
accesscolumn on the share was proposed and dropped), and it is stronger for it — AD #26’s guarantee that what a person may do is their role’s configuration survives intact, rather than becoming something a colleague can hand out record by record. It also removes a column, a picker control, and the question of what happens when share-level and role-level disagree, and it covers “cover for me while I’m on leave” with no extra mechanism. - Scope of a share: the record and everything attached — notes, activities, samples, quotes, files — plus the mail linked to that record, on the same footing a
team/allviewer already gets. It never widensmailbox.read. This half touches a privacy promise already made to Empire in writing (client/roles-and-permissions.md§7), which is why it is stated explicitly rather than left to implementation. - Also decided: not transitive (a recipient cannot pass a share on); shares survive owner reassignment and are revocable by the owner, the sharer, or a covering manager; sharing rides on the entity’s
managepermission rather than a new catalog key (managealready contains reassign the owner, which is strictly more powerful), with “only the owner or a manager above them” enforced as a domain rule; shares are displayed on the record, because a per-user effective-permissions view cannot enumerate per-record facts; Pipedrive’sLead - Visible toexport column does not migrate into shares (it is visibility-group semantics — re-verify against the export at spec time). - Distinct from what stays deferred: granular visibility groups (arbitrary named groups cutting across the org chart) remain out of V1, unchanged. Also distinct from OQ-57 (several parties on one lead), which is a domain relationship, not access control.
- Backend position: no change to anything shipped. FS-6 left the ownership context deliberately unwired (
OwnershipContext’s own javadoc hands it to F1/F2), nolead/deal/accounttable exists yet, and theScopeenum,role_grantand thePermissionCheckerinterface are all untouched — sharing is one additional resolver and oneOR. The real cost is the list-query predicate, which F1/F2 owe regardless of sharing:can()is per-record, and without a SQL-level visibility predicate a shared record never appears in the recipient’s list. - Found while prototyping, folded into OQ-53(b): a share onto a role holding no permission for that record type grants nothing — correct under the invariant, but surprising. The recommended Tea Room role is pricing-only, so a lead shared with them stays invisible unless Empire ticks lead-view on that role. The picker marks this rather than letting someone share into silence.
- Signals: OQ-58 closed. OQ-53(b) gains the Tea Room record-visibility note. OQ-56 (hierarchy policy) is unaffected — sharing works alongside it, not instead of it.
- Cascade:
client/roles-and-permissions.md§2 (new sharing subsection + A-34–A-37), §6 (sharing is an action, not a screen), §7 (the mailbox bullet now says sharing does not widen it) and §1 Rule 1 (the stale “roughly a hundred and fifty permissions” corrected to the coarse view/manage model FS-6 Rev 6 shipped);design/data/entity-map.md(arecord_sharerow);design/data/record-sharing-model.md(new — the full decision record); and the CRM prototype — owner + shared-with on lead detail, a share picker that marks who already has access and whose role could not use one, and a SHARED WITH YOU marker in the queue.
Rev 30 — 2026-09-16
- Changed — F11 researches every lead before it ranks it. The F11 summary and detail previously described scoring as classification of the inbound inquiry: AI priority P1–P4 + confidence, an AI summary of the email, a “why this priority” reason, a recommended next action and a drafted reply. They now carry a research pass that precedes scoring on every lead — public-source lookup for the company and the person, synthesised into a structured snapshot (industry, size, location, role, recent signals, certifications, origin/sourcing posture, import-export signals) with a source URL per field — and the score, confidence and “why this priority” are produced from that snapshot plus the inquiry.
- Why the change: research was a confirmed requirement (Dinil, 2026-09-16) that no document had captured. It is not in the PRD body, not in the §A.3 cost table, and not in the client-approved prototype’s Lead Triage screen — the corpus consistently described F11 as classification and attached research only to the capture app (F36). The requirement is what governs; the omission was ours.
- Why it is load-bearing, not a refinement. Empire’s stated main pain is the 200+ unreviewed lead backlog (proposal §57) and Phase 1 is aimed at it. A priority derived from inquiry text alone cannot answer what the ranking actually turns on — whether the company imports tea at all, at what scale, into which market, against which certifications — so without research the P1–P4 band and its stated reason are close to unfalsifiable. This is the same argument that put research on F36; it applies to F11 for the same reason.
- Grounding rules, carried from F12 and the HITL invariant: a field the search cannot ground is left empty, never guessed; a lead whose research returns nothing is still scored, from the inquiry alone, at lower confidence; research never auto-actions a lead — it informs a ranking a human acts on.
- Distinct from the deferred F29 — stated explicitly in two places (§F11 detail and the §Integration Points rows), because the two are easy to collapse and the collapse would be wrong in either direction. F29 is buying firmographic records from a data vendor per company and stays out of V1; this is the AI reading public sources, on the same
SearchProviderseam F36 uses. F29’s deferral does not restrict F11/F36 research, and this change does not pull F29 in. - NFR consequence: the research pass is background work, not a synchronous call. A search round-trip plus synthesis plus scoring does not fit the <10 s budget, so a lead is persisted on arrival and researched/ranked asynchronously, shown as pending-research in the queue; the backlog runs the same job in batch. § Non-Functional Requirements (AI latency) updated.
- Integration Points: a new web-search provider row (behind a
SearchProviderseam, one search per lead); the existing data-enrichment provider row annotated to hold the F29 boundary above. - Commercial position — two divergences from the issued proposal, stated not absorbed (L-001). §A.3’s costed line reads “Triage and score one lead · Haiku 4.5 · Classification against fixed criteria · 0.15 US cents”, which now understates F11; and §A.6’s data-minimisation statement reads “the lead’s own fields (triage)”, which no longer describes what triage sends. The second is a data-handling promise, so it is the one that matters. Both are recorded as post-issue notes on the proposal header. Lead sign-off: Tharinda, 2026-09-16 (relayed by Dinil) — required because the §A.6 half restates a data-handling commitment already made to the client in writing. Cost impact is immaterial (~$20/year all-in at F11 volumes, inside the fixed ET-002 fee); the scope impact is recorded against R-ARCH-001.
- Cascade:
design/ai/ai-provider-strategy.md§2/§3 (the F11 row moves from Haiku-classification to research + synthesis);design/architecture.md(the research/SearchProviderseam promoted out of the capture-app section into the AI layer);client/sow-cost-annexe.md§6/§7 (the enrichment line leaves “capture app once live” for the base, with a figure);client/client-facing-prd.md(the F11-research / F29-add-on distinction in plain language);design/ai/lead-research-source-options.md(new — provider options, verified pricing, recommendation, the open bake-off);features.md+tracker.yaml(F11 scope note, R-ARCH-001). - Provider not chosen here. The seam is the commitment; the provider behind it is an implementation choice pending a quality bake-off, recorded in the new design doc. Resolves: nothing in signals — this was a confirmed requirement we had failed to record, not an open question.
Rev 31 — 2026-09-17
- Changed — a weekly price is a FOUR-FIGURE BUILD-UP, not a single number. §Functional Surface > F26 previously described a weekly price as one USD figure per standard per week. It now records the derivation Empire actually works in: the price is set in LKR, an overhead is added in LKR, the total is converted to USD at the week’s rate, and a margin is added to reach the final USD price given to the customer. All four figures are stored on the price row; only the final USD price is required, and the three working figures are recorded when the Tea Room has them and left empty when a price is set directly.
- Why the change: team decision — Tharinda, 2026-09-17; recorded by Dinil. The structure was visible in Empire’s weekly workbook — the derivation was already noted qualitatively in
discovery/source-schemas/weekly-price-workbook-schema.mdand flagged as OQ-59’s open half — but every modelled artifact carried only the final figure, so the working was being discarded. Keeping it is what makes a past week’s price reconstructable and the margin auditable at the figure the Tea Room actually set, rather than only at quote time. - Refines OQ-43, does not reverse it. OQ-43’s answer was “USD base; other currencies are display conversions only. Currency is not a price-key axis” (Rev 4). USD remains the only currency a price is published in and the currency the customer is quoted, so the key is unchanged and no second price row per currency appears. What is corrected is the implied claim that the USD figure is the origin of the price — it is the end of an LKR chain, and that chain is now kept beside it.
§RisksR-17’s currency clause is restated accordingly. - Not an AD change. AD #14 (effective-dated rows, never updated in place) is untouched and is what the four figures are stored under — they are columns on an append-only price row, not a second mutable record. No DO-NOT-REVERSE row is reversed, so no lead sign-off is required on that ground.
- Data-model consequence — owed, not done here.
forge-et-be/docs/data-model.md§2.2/§2.5 carries a singleunit_price_usd NUMERIC(12,4) NOT NULLonbuying_standard_price, and §2.10 R2 justifies dropping currency from the key on OQ-43’s authority. It needs three added nullable columns (price_lkr,price_lkr_with_overhead,price_usd_pre_margin), a restatement of R2, and the matching change todocs/proofs/pricing/01-schema.sqlso the bidirectional drift check passes. Owner: Tharinda (AD #31). The final USD column staysNOT NULL— only it is required. The immutability trigger’s writable-column set is unchanged, so the three new columns are immutable once written, which is correct for an append-only price. - Open, and narrowed: whether the exchange rate and the overhead are one figure per week or per standard, and who produces the sheet, stay in OQ-59 — they change where those two live (a per-week row such as
auction_sale, versus a column on every price), not whether the build-up is stored. - Cascade:
discovery/source-schemas/weekly-price-workbook-schema.md(the derivation bullet);design/architecture.md§Weekly Bulk Pricing Subsystem (the canonical-price design row);design/prototypes/feature-detail.md§F26;client/client-facing-prd.md(plain-language);.claude/CLAUDE.md§ Active Context; the CRM prototype (both price-entry modals, shipped 2026-09-16);project-prd-signals.mdOQ-59 narrowed. - Resolves: the derivation half of OQ-59 (OQ-59 stays open on rate/overhead grain, template and duplicates).
Rev 32 — 2026-09-17
- Changed: §Functional Surface > F4’s opening sentence no longer lists “deal history” among the migrated data. It now reads that deals and deal history are not migrated, pointing at §Scope > F4, which had recorded Empire’s fresh-start election on deals since 2026-09-13.
- Why: Gate 3 (
/forge-decompose, Run 1) found the two sections disagreeing — §Scope newer and authoritative, §Functional Surface stale — and the F4 sibling specs (F4-a/F4-d/F4-b/F4-c) would otherwise have inherited the stale sentence. Team decision at the dry-run review (Q12-A, 2026-09-17). No requirement changes: this aligns the body with a decision already in the contract. - Also this revision — the Gate 3 pass itself:
## Gate Statusrow 3 flips to ✅ passed (2026-09-17, Run 1). The feature list the PRD’s §Scope enumerates (F1–F26, F31–F34, F36) is now decomposed into 89 sibling features infeatures.md/tracker.yaml; the PRD’s F-numbers survive as each sibling’s parent. §Scope and §Functional Surface are not rewritten to the sibling ids — they stay the Gate-1 snapshot;features.mdis the live index (features.mdheader). - Cascade:
project-prd-signals.md— everyBlockscell re-mapped from parent F-numbers to sibling ids, and OQ-63 (past orders on the account — no V1 feed) raised against F1-d;client/project-proposal.mdheader — post-issue note on the Phase 1 / Phase 2 divergence;client/project-timeline.md§2;.claude/CLAUDE.md § Active Context. - Resolves: — (structural; no OQ resolved).
Rev 33 — 2026-09-21
- Changed: §Non-Functional Requirements > Accessibility gains an explicit light-theme-only line — both frontends (the CRM and the F36 capture PWA) ship the single light (Empire) theme; dark mode is out of scope for all of V1 (neither Phase 1 nor Phase 2), a future-release item only on client request.
- Why: team decision (lead, 2026-09-21). The contract was silent on theming while design-system §9 already stated “Dark mode is not in V1”; the lead asked to make the scope explicit and unambiguous across both frontends and both phases, so nothing reads it as CRM-only or V1-implicit. No requirement removed — this states an existing design decision in the contract. Consistent with F5-b spec Rev 1 (AC-14/NFR-1 light-only) and design-system §9 (Rev 20).
- Cascade:
.forge/design/ui/design-system.md§9 (Rev 20 — the same clarification at source). - Resolves: — (clarification; no OQ resolved).
Rev 34 — 2026-09-23
- Changed: §Risks R-12’s “Graph push subscriptions expire in ~3 days” becomes within seven days — Microsoft’s documented ceiling is 10,080 minutes for Outlook message subscriptions, pinned in code as
MailboxSubscriptions.MAXIMUM_EXPIRY. The row’s mitigation wording is unchanged (F20 stays P0; scheduled renewal plus a lapsed-subscription alarm); only the figure was wrong. - Why: F20-a established the real figure from Microsoft’s documentation on 2026-09-22 and pinned it in code, but never wrote the correction back to R-12 or to
architecture.md. F20-b’s spec then reproduced the stale number in good faith and rested a scope decision on it — whether to build a Graph-side subscription-reconciliation pass, where an orphan window of three days versus seven is the entire argument. The spec’s third review pass caught it by checking the claim against the backend repo rather than against the harness documents, which is the only way it was findable: four copies existed, all four agreed with each other, and all four were wrong — R-12,architecture.md§Mailbox integration,discovery/research/m365-mailbox-integration.md(which namesarchitecture.mdas its authority and then restates the figure) and the F20-btracker.yamlnote. This entry first said there were two. The scoped fourth review pass found the other two and corrected this sentence — which had reproduced the very copied-claim error it was written to fix. - Cascade:
.forge/design/architecture.md§Mailbox integration (Rev 39 — the same correction at the other copy, which now points at the constant rather than restating a number)..forge/specs/F20-b-inbound-sync-inbox-spec.mdD-3 re-argues its deferral on seven days and carries W-7 recording this write-back as owed. - Resolves: — (correction; no OQ resolved).
Rev 35 — 2026-09-23
- Changed: §Scope and Boundaries > Phasing / Sequencing Intent, row Capture stream (parallel): “weeks 8–13, landing within the Phase-2 window” becomes alongside the Phase 2 work (weeks 7–13, 26 Oct – 27 Nov), placed there at Gate 3 with its scope unchanged, and the descope lever is now dated — decided at the mid-November checkpoint, Fri 13 Nov. The same correction in §F36 — Trade-Show Lead Capture App, whose “runs as a parallel work-stream, weeks 8–13” becomes “alongside the Phase 2 work, weeks 7–13 (placed there at Gate 3)”. The F36 scope, the parallel-stream shape and the lever itself are unchanged.
- Why: Gate 3 (2026-09-17) placed F36 in Phase 2 (
tracker.yaml: all six F36 features, F36-a to F36-f, havedelivery_phase: 2), andclient/project-timeline.md§2 schedules the stream “alongside the Phase 2 work” with its checkpoint on Fri 13 Nov. The row still carried the pre-Gate-3 “weeks 8–13”, and an independent review of the CLAUDE.md compression (harness PR #184, where AD #20 now reads “originally weeks 8–13; Gate 3 placed it in Phase 2”) flagged it. The issued proposal names no week numbers for the stream, and its 17 September post-issue note already records the capture application in the second release, so no new post-issue note is owed (L-001). - Cascade: the same correction in
.forge/features.md(parallel-lanes table, Capture stream row),.forge/design/architecture.md§Trade-Show Capture App “Scheduling” line (architecture Rev 40),.forge/design/prototypes/feature-detail.md(F36 “Delivery” line),.forge/design/ui/design-system.md§10 open item 5’s due point (design-system Rev 23),.claude/CLAUDE.mdReference Map capture-repo row, and.forge/team-guide.md(repo-tree comment). Architecture revision entries and approved foundation specs that say “weeks 8–13” or “week 8” are point-in-time records and stay as written. - Resolves: — (correction; no OQ resolved).
Rev 36 — 2026-09-24
- Changed: §Users and Access > Per-Role Capability Matrix — the Account row’s Sales Rep column
gains “archive (own)” and “reassign (own)”: a rep may archive, and hand to a colleague, an account
they own; the Contact row’s Sales Rep column gains “archive (own account)”. The Account and
Contact rows’
Dbecomes archive for Sales Managers and Administrators: V1 has archive and restore, no hard delete. The Reference data row and §Users and Access > Roles, Administrator — “dropdown lists” is qualified: in V1 the account dropdown lists are fixed in code and changed by a small developer release, with no administration screen; the certification list is a table. - Why: the lead’s answers on 2026-09-24 to the questions F1-a and F1-b left open. On reassign: the
approved CRM prototype’s permission list already describes “Manage accounts and contacts” as including
reassign, and F1-a ships it that way (F1-a spec Q-10, Rev 4); the matrix listed reassign for managers
only. On delete: F1-a ships archive only, and archive keeps every note, timeline row and attachment
referenceable — which is also why reps may archive their own records, as F1-a already allows
(
account.manageatown) and F1-b’s screen shows. On dropdown lists: no feature owns an administration screen, and a list change is a small release; the certification list is already a table. - Cascade: F1-a spec Rev 4 (FR-A7, FR-O1, FR-A9, Q-4, Q-10); F1-b spec Rev 2 (Q-1, Q-4);
client/roles-and-permissions.md(managers archive rather than delete; reps archive their own accounts and contacts and may hand over an account they own — footnote ⁷, A-12, A-38; the account dropdowns in the Reference data row are fixed by us in V1);client/open-questions.md(the lists and the dormancy threshold go to Empire, OQ-66 and OQ-67); the F32-b spec stub (the bulk action is archive). Sharing an account without handing it over is unchanged — record sharing, F5-g (Rev 29). - Resolves: OQ-64, OQ-65.
Rev 37 — 2026-09-24
- Changed — the access model is realigned to the lead’s locked decision set (AD #34 access, AD #35 ownership and sharing, AD #36 audit trail).
- §Users and Access > Roles: the Managing Director is a recommended role created by Empire’s administrator, not a seeded profile; it holds
quote.approve_sub_floorby recommended configuration, not by a code rule. The V1-roles note says OQ-47’s levels are met byview/manageper entity plus special keys, states that an outward action on a record is no longer grantable apart from managing that record (except where it has its own key), and says a fresh installation ships with the Flyway-seeded IT Admin role and one user only. The Administrator row namesaccess.manageandsettings.manage. - §Per-Role Capability Matrix: the status note says the system ships with the editable IT Admin role (no superadmin) and that every cell is a recommended set of keys at a scope, not a code rule. The legend maps the actions onto the catalog (R =
view; C, U, archive, reassign, share, A =manage; Cfg =settings.manage) and restatesown/team/allwith shares. Cells restated: R&D “R (linked)” becomesviewatown(the records shared with the taster); Contact is governed byaccount.*; a task follows its record’s key; Email / Message becomes own mailbox only (email.manage) plus the mail linked to records the user reaches, and the Sales Manager’s “send (all)” is gone; the certification cells split acrosssettings.manage,buying_standard.manageandaccount.manage; import needssettings.manageand export followsviewand is audited; the Buying Standard row splitsbuying_standard.view(the whole book with costs) fromaccount.view(an account’s standards with the final USD price); Users & roles has no Delete, the administrator’s cell isaccess.manage, and the Sales Manager has only the minimal name-and-role directory. A new note lists the distinctions the minimal catalog cannot express (see Lead decisions owed below). A Rev 37 note follows the three older matrix notes. - §Permission levels → §Permission keys: the category table (read / edit / execute), the “outward-acting” warning and the Composition paragraph are removed; the D8 catalog table (Phase 1 and Phase 2 keys), “
manageimpliesview”, the retired keys and the areas that have no key are added; FS-6 Appendix A is named as the authoritative key list. Where the AI sits and the human-in-the-loop invariant are restated onview/manage/ special keys. - §Named permissions → §Special keys, and the Managing Director:
APPROVE_SUB_FLOOR_QUOTEbecomes the catalog keyquote.approve_sub_floor; the MD arrangement is the recommended configuration, and the proposal’s “approved by the Managing Director” is stated as met by configuration; the role is created by Empire’s administrator (nothing but IT Admin is seeded); the separation-of-duties note is replaced — separation of duties is configuration only, self-escalation is unlimited, and the audit trail is the control. The OQ-53(b) note is restated in keys. - §Multi-Tenancy / Org Hierarchy: rewritten to AD #35 —
own/team/allincluding shares;teamrecursive;allincludes unowned records; edit versus view atteamis configuration; no key reaches another person’s mailbox; sharing transitive, with sharing, removal and reassignment undermanage; the eligible-only share picker; reassignment rules; children follow the parent; one visibility rule for lists and single records, built by F5-f; unowned leads visible only atall, with an Unassigned filter; deactivation refused while the user has reports; deactivated users stay in the tree. - §Functional Surface > F5: rewritten — exact-match sign-in for active pre-created users; no auto-provisioning, invitation or Entra sync; all administration under
access.manage; the role mandatory at creation; users never deleted; an email change clears the identity link; deactivation ends sessions and is refused for oneself, for the lastaccess.manageholder, while the user has reports, and while they own records; the Flyway-seeded IT Admin with a per-environment email; the last-admin invariant; unlimited self-escalation; changes apply on the next request; the share surface, reassign action, minimal user directory and the effective-permissions viewer behindaccess.manage. - §Domain Model: User (pre-created, one mandatory role, deactivated never deleted, the migration-seeded first user), MailboxConnection (per user only; no unowned connection), Lead (owner nullable only for import edge cases), Account (owner required), Quote (owner; approval key); new rows Record share and Audit event.
- §F7, §F11, §F20, §F23, §F26: F7’s sub-floor approval names the key; F11 gains the Unassigned filter (no queue); F20 handles
info@by forwarding, may addinfo@as an ordinary user, and drops the shared inbox and “team inboxes”; F23 states that imports needsettings.manageand exports are audited; F26 namesbuying_standard.manage,buying_standard.view/account.viewfor the price-lookup page, andprice.publishfor publishing. - §Integration Points (inbound email), §Deferred (the inbound-trigger row; the visibility-groups row gains sharing), §Reference-System Classification P-34, and §In Scope F5.
- §Non-Functional Requirements: Security — AuthN/AuthZ restated on the grant model, with the MD and Administrator restrictions as recommended configuration; mailbox security says no role reads another person’s mailbox. Observability — the audit-trail line is replaced by AD #36’s definition (every write and security event, same transaction, not reads,
SYSTEMactor, kept indefinitely, no viewer in V1, timestamps on every table). - §Out of Scope gains: self-service user creation (auto-provisioning and invitation email); synchronising users from Entra; a shared inbox inside the CRM (F20-e); an audit-viewer screen; hard delete of users.
- The header’s Last updated line, which had stayed at Rev 28, now reads Rev 37.
- §Users and Access > Roles: the Managing Director is a recommended role created by Empire’s administrator, not a seeded profile; it holds
- Resolves: OQ-51, OQ-52, OQ-53, OQ-54, OQ-56. (OQ-53 closed as not required after the independent review of 2026-09-24: the lead had said role contents need no definition because every role is Empire’s configuration.)
- Revises: the folds of OQ-14 (
info@stays an inbound channel, reached in V1 by forwarding, not a shared connection), OQ-47 (the levels are met byview/manageplus special keys, with no category axis) and OQ-58 (sharing transitive; removal by anymanageholder; the eligible-only picker; the roll-up throughteam; therecord_sharecolumns; built as the F5-f substrate). Each history row carries the pointer. - Supersedes, does not edit: the relevant parts of Rev 19 (the Managing Director as a seeded role; separation of duties as a property of the default), Rev 21 (the permission category metadata; OQ-56 as an open question), Rev 24 (the non-deletable system superadmin; the read / edit / execute categories) and Rev 29 (non-transitive shares; revocation by the owner, sharer or a covering manager; “only the owner or a manager above them” may share; a picker that marks users whose role could not act on the record). Those entries stand as the record of what was decided and why.
- Signals: OQ-53 is closed as not required (moved to Resolved) — part (a) is superseded by AD #34, and part (b), which keys the recommended Tea Room and Executive roles hold, needs no answer because every role is Empire’s configuration (the lead, 2026-09-22; applied after the independent review of 2026-09-24). OQ-60 and OQ-62 gain a note on sender identity (no Microsoft 365 shared-mailbox sender through Graph; an SES sender would need a key the catalog does not have); neither is resolved. No remaining signal row blocks F20-e (dropped from V1) or F5-f (re-scoped to the ownership and sharing substrate).
- Why: the lead’s design conversation of 2026-09-22 to 2026-09-24 — roles built from scratch on the fewest keys that describe the business, a migration-seeded administrator so the first person can get in, one owner plus explicit shares for reach, and an audit trail as the control that makes unlimited self-escalation and transitive sharing acceptable. Lead sign-off: Tharinda, 2026-09-24 (AD #34, AD #35, AD #36) — required because this reverses Rev 24 and Rev 29 positions Tharinda approved and refines DO-NOT-REVERSE AD #30 and AD #32.
- Delivery: F5-k (access-model realignment, new), F5-f (re-scoped to the ownership and sharing substrate), F39 (audit trail substrate, new); F5-c, F5-g and F5-h re-scoped (F5-d by its own session); F11-b’s unassigned queue becomes a filter; F20-e dropped from V1.
- Commercial position (L-001) — the documents were restated in the same change (the proposal’s 24 September post-issue note;
client/roles-and-permissions.md,open-questions.md,dependencies.mdandclient-facing-prd.md); what remains owed is the lead’s call on the change-request wording and the line to Empire (drafted inengagement-log.md§ AD #34–#36):- The issued proposal §2 describes access as role plus “granular permission levels (read-only, edit, AI-use, and execute)”, and its 11 September post-issue note restated that as read / edit / execute with one built-in Administrator role. Under AD #34 an execute-type action can no longer be granted apart from
manageon the entity, and there is no system Administrator role. Done in the same change — the proposal’s 24 September post-issue note; whether it may also say “no change request arises” is the lead’s call, and the line to Empire remains owed. - The proposal’s Managing Director approval is delivered only as configuration, and the holder of
access.managecan grant approval to their own role, with an audit trail. Client question Q3 andclient/roles-and-permissions.mdpromised that Administrators cannot approve and that no single account can lower the floor and approve. Both are restated and the post-issue note covers it; the line to Empire says so outright. - The
info@decision agrees with the proposal, so the proposal needs no note on it; the issued dependency register and Q6 told Empire the shared address would be connected; both are corrected, and the line to Empire says so. - The unowned-records default put to Empire (OQ-56(a)) changed, and the Q2 default (invitation email; records stay put) changed;
client/open-questions.mdandclient/roles-and-permissions.mdare re-revised; Empire is told in the line. - The production administrator’s email is a new client input (Empire IT); a production database cannot start until it is supplied.
- The issued proposal §2 describes access as role plus “granular permission levels (read-only, edit, AI-use, and execute)”, and its 11 September post-issue note restated that as read / edit / execute with one built-in Administrator role. Under AD #34 an execute-type action can no longer be granted apart from
- Lead decisions owed (recorded in the capability-matrix note):
- Delete:
<entity>.managecovers every action, so the matrix’sDfor managers and not for reps cannot be expressed; whether leads, deals, quotes and the other records have a hard delete at all, or are archive-only like accounts. - The default that
account.viewshows an account’s standards with the final USD price, while the costs needbuying_standard.view. - (R&D reach and whether a Sales Manager imports were listed here at first; they are recommended-role contents, which the lead said need no definition — see OQ-53 — so they are advice in
client/roles-and-permissions.md, not decisions owed.)
- Delete:
- Cascade (the harness documents the access-model realignment touches):
.claude/CLAUDE.md(AD #34–#36 and the rows they refine); the feature specs of F5-k, F5-f, F39, F5-c, F5-e, F5-g, F5-h, F11-b and F20-e (F5-d is re-scoped by its own session);features.mdandtracker.yaml;design/architecture.md,design/data/entity-map.md,design/data/record-sharing-model.md,design/ui/design-system.md,design/environments-and-identity.md,design/coding-standards.md,design/prototypes/feature-detail.md,test-strategy.mdandcloud-gated-validation.md; and the client pack (client/roles-and-permissions.md,client/open-questions.md,client/dependencies.md,client/client-facing-prd.md, and a post-issue note onclient/project-proposal.md).
Rev 38 — 2026-09-25
- Changed: §Domain Model > Key Entities gains a Tag row: a label kept per entity type (account, contact, lead, deal), its name unique within its type regardless of case, with an optional colour; the name list readable by any signed-in user, tagging a record part of managing that record, renaming or deleting a name a configuration action; N:N to Accounts and Contacts (F1-a), Leads (F11-a) and Deals (F2-a). The header’s Last updated line moves to Rev 38.
- Why: tags are required functionally by §F1 (Rev 23 — grouping and filtering, consumed by F32’s saved filters) but appeared in neither the Key Entities table nor the entity map. F1-a ships the
tag,account_tagandcontact_tagtables (F1-a spec FR-S5, FR-D5), and its work item WI-3.1 decided the gating (plan D16). This is a one-row addition that records an entity already in scope; it is not a scope change, and the issued proposal is unaffected (L-001). - Cascade:
.forge/design/data/entity-map.mdRev 12 (theTagrow with its four joins and owners, in the same change);forge-et-be/docs/data-model.md§3 (the shippedCREATEs, in the F1-a WI-3.1 backend PR). - Resolves: — (addition; no OQ resolved).
Rev 39 — 2026-09-25
- Changed: §Functional Surface > F26’s closing sentence no longer says the
data-model.md§2.2 revision is owed; it names §2 as revised for Rev 23 (nomarketprice axis), Rev 27 (no product) and Rev 31 (the four-figure build-up), owner Tharinda (AD #31). The header’s Last updated line moves to Rev 39. - Why: the revision has landed on
forge-et-bemain(§2.2 and the §2.5 build-up columns), and the stale “owed” marker was found while specifying F26-a (F26-a spec FR-D2, review finding I6). A correction of a stale status note, not a scope change; the issued proposal is unaffected (L-001). - Cascade:
.forge/design/data/entity-map.mdRev 13 (the cluster-4 Price row’s “revision owed” cell) and.forge/design/architecture.md§Weekly Bulk Pricing (two matching markers), in the same change;.forge/project-prd-signals.mdOQ-59’sBlockscell drops F26-a with a dated waiver note (F26-a spec W-1). - Resolves: — (no OQ resolved; OQ-59 stays partial and passes to F26-c).
Rev 40 — 2026-09-26
- Changed: §Domain Model > Key Entities › Lead’s relationships gain “N:1 matched Account / Contact (optional, before conversion)”: a lead may record which existing customer and contact the inquiry comes from, before and separately from any conversion. The header’s Last updated line moves to Rev 40. (Drafted as Rev 39; renumbered when
mainmerged in, because F26-a landed its Rev 39 first.) - Why: F11-a (spec approved 2026-09-25, FR-O6, FR-D2) ships
lead.account_idandlead.contact_idso that F4-b, F11-d and F11-f read one column for “which customer is this” (developer decision, 2026-09-25). The match is a link, not ownership: it changes nobody’s reach (AD #35), and it is not the conversion link, which stays F11-e’s. It records a relationship inside the lead substrate already in scope under §F11; it is not a scope change, and the issued proposal is unaffected (L-001). - Cascade:
.forge/design/data/entity-map.mdRev 15 (the Lead → matched Account / Contact key-edges row, theLeadandLead sourcerows, D-4 closed, the D-5 note — in the same change);forge-et-be/docs/data-model.md§3 (theleadCREATEwith the two columns, in the F11-a backend PR). - Resolves: — (addition; no OQ resolved).
Rev 41 — 2026-09-27
- Changed: §Scope and Boundaries › Phasing gains a note that the Gate 3 re-grouping (2026-09-17) supersedes the 2026-08-18 consolidation for which phase each item ships in, pointing at
tracker.yamldelivery_phaseandfeatures.md § Delivery Plan. §Integration Points › Pipedrive no longer lists deals for migration: deals are a fresh start (2026-09-13). §Integration Points › Amazon Bedrock, §Security › AI data handling and §Non-Functional › AI provider now say “the project’s AWS account” instead of “the client’s own AWS account”, without naming an owner — account ownership and billing are commercial terms that live in the SOW and the proposal, not in the requirements. - Why: the second harness review (2026-09-27) found these statements lagging decisions already recorded elsewhere — Gate 3 Run 1 and the proposal’s 17 September note (phasing) and the 13 September migration election; the AWS wording is made owner-neutral so the PRD does not restate a commercial term (lead decision, 2026-09-27). Corrections, not new decisions.
- Cascade:
.forge/client/client-facing-prd.mdand.forge/client/project-timeline.mdcarry the phasing and migration corrections in client terms (revised 2026-09-27);.forge/design/ai/ai-provider-strategy.md(owner-neutral AWS wording) and.forge/design/architecture.mdRev 44 in the same change. - Resolves: — (correction; no OQ resolved).
Rev 42 — 2026-09-27
- Changed: §Non-Functional › Accessibility gains a Supported sizes bullet: the CRM on desktop and tablet (768px floor) with every page reflowing at 320 CSS px (WCAG 2.1 SC 1.4.10); the capture app on phone first and on tablet and desktop too, in both orientations; both tested in Chromium and WebKit; the browser accessibility and responsive checks gate every pull request.
- Why: lead decisions, 2026-09-27, after an audit of the built screens in both apps at phone, tablet and desktop sizes found four defects (BUG-028 to BUG-031) and showed that no CI lane ran the browser checks. The WCAG 2.1 AA target already in this section includes SC 1.4.10, so reflow at 320px makes the existing target explicit rather than adding one. The proposal commits to “a responsive web application”, so no commercial scope changes.
- Cascade: design-system Rev 28 (§ Viewport and breakpoints, § The capture app);
.forge/test-strategy.mdRev 10 (§ Browsers & Viewports); the checks and the CI gate are built as F40; the defects are BUG-028 to BUG-031. - Resolves: — (addition; no OQ resolved).
Rev 43 — 2026-09-27
- Changed: a grant loses its scope (AD #37). §Permission keys: the grant is
(role, permission_key); each entity that has an owner gains a third key,<entity>.all(lead.all,account.all;deal.all,quote.all,sample.allin Phase 2), which widens the holder’sviewormanageon that entity to every record and means nothing without one of them; the key table’s “Scoped?” column becomes “allkey?”. §Multi-Tenancy / Org Hierarchy: the three per-permission scopes are replaced by one fixed rule — a key reaches records owned by, or shared with, the user or anyone below them in the reporting tree, and theallkey reaches every record, unowned ones included; the section states the two things given up (a role cannot reach further withviewthan withmanageon one entity; a user with people below them always reaches their records) and that setting a manager is an access decision. The capability-matrix legend keeps “own”, “team” and “all” as shorthand, redefined by the new rule; footnote ¹ (R&D), theDfootnote and the OQ-53 note are restated in its terms; a Rev 43 note joins the matrix notes. The User row, the role-editor sentence and the effective-permissions sentence of §F5, the unowned-records rule (nowlead.allholders) and the Deferred visibility-groups row follow. The header’s Last updated line moves to Rev 43. - Why: the lead’s design (Tharinda, 2026-09-27). For a user with nobody below them the
teamscope already gave exactly whatowngave, so the choice between them only mattered for managers, and the reporting line already says who manages whom; only “every record” needs a setting. Every recommended role in the matrix and inclient/roles-and-permissions.mdholds one reach per entity forviewandmanagealike, so each is expressible without scopes — the lead’s worked example (seven users, five roles) gives identical results under both models. The role editor loses a scope choice on every key, and the grant table loses a column. - Cascade:
.claude/CLAUDE.mdAD #37 (and pointers on AD #15, AD #30, AD #35) with its full text inengagement-log.md § AD #37;architecture.mdRev 45;design/data/entity-map.mdRev 16;design/data/record-sharing-model.md(revised 2026-09-27);design/ui/design-system.mdRev 29 (§8 role editor);test-strategy.mdRev 11;client/roles-and-permissions.md(revised 2026-09-27 — A-6’s alternative withdrawn, A-9 restated),client/client-facing-prd.mdandclient/open-questions.md(Q1) in client terms, and a post-issue note onclient/project-proposal.md(27 September);design/data/roles-and-permissions-key-map.md(the “All records” row);design/coding-standards.md,design/prototypes/feature-detail.md(a drift row: the prototype sets a reach per permission),guides/tech-stack-reference.mdandcloud-gated-validation.md§B F5-f (wording); the specs of F5-c, F5-d, F5-e, F5-f, F11-b, F1-b, F2-a, F4-a, F4-b and F4-d, and revision notes on the shipped F1-a and F11-a specs;features.mdandtracker.yaml(F5-l added; F5-d, F5-e and F5-f blocked by it). The code change is the new feature F5-l (after F5-k; spec stubspecs/F5-l-reach-from-reporting-line-spec.md), which also narrows unowned records to theallkey’s holders, adds theallkeys to FS-6 Appendix A and revisesforge-et-be/docs/auth/anddocs/data-model.mdwith the code. - Resolves: — (refinement; no OQ resolved).
Rev 44 — 2026-09-28
- Changed: the Pipedrive migration is narrowed to leads, organisations and people. §Scope › F4, §Functional Surface › F4, §User Journeys › Journey D and §Integration Points › Pipedrive no longer list notes, activities or custom fields; §F4 now says the migration runs more than once (a first load, then a final run after the data freeze that adds what Pipedrive gained in between and never overwrites a record changed in the CRM), records the counts from the exports (692 leads, 4,235 organisations, 4,977 people), and states that every imported lead and account is owned at first by one named Empire user — Kavindu Anthony in production — who reassigns them, with the Pipedrive owner’s name kept on each record. §Scope › F34, §Reference-System Classification › P-09 and §Functional Surface › F34 no longer say Pipedrive custom fields migrate through F34. A custom field’s value may still be read to fill a typed field (
Person - Trade Fairsfills a lead’s source). Journey D is aligned with OQ-21 (everything migrates with its Pipedrive ID; AI triage runs after import) and gains the final run; §NFR › Scale points at the F4 counts instead of “migration volume TBD”; §NFR › Migration describes the first load and the final run as batch writes and lets the dry-run preview, which writes nothing, answer in the request. The header’s Last updated line moves to Rev 44. - Why: lead decisions (Tharinda, 2026-09-28): migrate leads and accounts only, where accounts means Pipedrive organisations and people; and give every imported lead and account one owner at first, who reassigns them later, so no Pipedrive owner is matched to a CRM user and no Pipedrive user list is needed. Empire had not asked for notes, activities or custom fields to be migrated; the delivery team records the change without a client request. It narrows the issued proposal’s §4 migration line and the 17 September post-issue note’s “and their notes”, so a post-issue note on
client/project-proposal.md(28 September) states it (L-001). The final run after the freeze answers the replay half of architecture AD-O-7 for the migration’s read side (F4-a spec). - Cascade:
client/project-proposal.md(post-issue note, 28 September);client/client-facing-prd.md(revised 2026-09-28, in client terms);design/architecture.md(AD-O-7 row);discovery/source-schemas/(organisation and people export schemas added; the leads schema’s stale notes corrected); the F4-a spec (drafted against this revision) and the F4-b, F4-d stubs (notes no longer migrated; one import owner). OQ-26 no longer blocks F4-a or F4-c, and OQ-41 no longer blocks F4-a (project-prd-signals.md). Owed, not done here: F4-c (Migrate activities and custom fields, Phase 2) loses its content; whether it is dropped, or kept for the final run and the client migration report, is a tracker decision for the lead. - Resolves: OQ-6.
Rev 45 — 2026-09-28
- Changed: §Non-Functional › Accessibility › Supported sizes: the accessibility and responsive browser checks still gate every pull request, but they run on the developer’s machine before the PR is marked ready, not in CI. The header’s Last updated line moves to Rev 45. (Drafted as Rev 44; renumbered when
mainmerged in, because F4-a landed its Rev 44 first.) - Why: F40’s gap check (2026-09-28) found that Rev 42’s “gate every pull request” contradicted
architecture.md § Browser E2E runs in two CI tiers(AD-O-11), which keeps browser runs out of PR CI because Actions minutes are one shared org budget. A CI browser job for the CRM would also build and start the backend on every run and need a token to readforge-et-be. The developer decided (Dinil, 2026-09-28, with Tharinda’s agreement) to run the full matrix locally through one command per app repo, enforced by a harness hook and a seconds-long last step of theSlow lanethat reads the result from the PR description. The proposal is unaffected. - Cascade:
.forge/test-strategy.mdRev 12 (§ Browsers & Viewports); F40’s spec (Context, FR-7 to FR-13) carries the gate, the hook and theSlow lanestep, and its implementation updates.claude/CLAUDE.md § Quality Gates,coding-standards.md §7.1,architecture.md § CI/CD Pipelinesand both app repos’CLAUDE.md. AD-O-11 is unchanged. - Resolves: — (refinement; no OQ resolved).
Rev 46 — 2026-09-30
- Changed: §Multi-Tenancy / Org Hierarchy › Record sharing now says who may share, remove a share and reassign the owner: anyone whose role holds the entity’s
manageand who reaches the record — its owner, a manager above the owner, a holder of the entity’sallkey, or a colleague it was shared with. › Unowned records adds that since Rev 44 no planned V1 path leaves a lead unowned, so the rule is a safeguard, and F11-b keeps the Unassigned filter as one (lead, 2026-09-30). The capability matrix’s Sales Rep cells for Lead and Account name reassignment of a record shared with the rep as well as one they own. The header’s Last updated line moves to Rev 46. - Why: lead confirmation (Tharinda, 2026-09-30) of AD #35 as written, after the delivery team found the client roles document saying a representative “cannot reassign records they do not own”, which contradicted AD #35 for a record shared with them. The lead chose the simple rule — any
manageholder in reach may reassign, share and remove shares — over limiting reassignment to the owner, the owner’s managers andall-key holders. The unowned-lead sentence follows Rev 44’s single import owner, which the §Multi-Tenancy paragraph had not been updated for. No AD changes: AD #35 D-1 and D-3 already said this. - Cascade:
client/roles-and-permissions.md(revised 30 September: a “What changed on 30 September” paragraph, the representative profile’s “Can” and “Cannot” lines, A-38, A-7 and “Who sees records nobody owns?”);client/open-questions.md(revised 30 September);design/data/record-sharing-model.mdD-1;design/data/roles-and-permissions-key-map.md; the F5-f stub (the worked example gains the case of a sharee reassigning) and the F11-b stub (the Unassigned filter kept as a safeguard);test-strategy.mdRev 14 (the sharing row);.claude/CLAUDE.md(the live-decisions line);tracker.yaml(the F5-f and F11-b notes) andfeatures.md(the F5-f row);engagement-log.md(the dated lead decision). - Resolves: — (refinement; no OQ resolved).
Rev 47 — 2026-09-30
- Changed: §Functional Surface › F4 says both migration runs stay (the first load and the final run after the data freeze); the import owner of a run is the signed-in user who runs it unless the run names another, and holds
lead.manage,account.manage,lead.allandaccount.all; they hand the records out with the bulk “change owner” action, and a later run never takes a record back from its new owner; the kept Pipedrive owner name is never shown on a record or in an everyday list, only in the hand-out; the parallel run begins on a date agreed with Empire once the migration is ready. › F32 says the “reassign owner” bulk action on the lead and account lists arrives in Phase 1 (new feature F32-c), and every other bulk action and deal reassignment stays in Phase 2 (F32-b). The header’s Last updated line moves to Rev 47. - Why: lead decisions (Tharinda, 2026-09-30). After an import every record belongs to the import owner, and a rep reaches none of theirs until it is handed to them (AD #37), so a bulk hand-out has to exist by the first load; handing out thousands of records one at a time is not workable. Only the hand-out is brought forward; the rest of F32-b stays where Gate 3 put it. Both runs stay as a safeguard. The 9 October start of the parallel run was the schedule replaced on 29 September; the first load needs F4-d and F4-b, late in Phase 1.
- Cascade: new stub
specs/F32-c-bulk-reassign-spec.md; the F32-b, F4-d, F4-b and F5-h stubs; the F4-a spec Rev 4;tracker.yaml(F32-c added to Phase 1; F32-b, F4-d, F4-b notes);features.md(F32-c row, phase counts, the import chain);team-guide.md(the Import chain);client/client-facing-prd.md(revised 30 September: the parallel-run date and the Phase 1 hand-out);.claude/CLAUDE.md(Phase scope);engagement-log.md(the dated decisions). - Resolves: — (refinement; no OQ resolved).
Rev 48 — 2026-09-30
- Changed: §Non-Functional › Security, the AI data-handling bullet: Claude is reached “via Amazon Bedrock, called from inside the project’s AWS account” rather than “inside the project’s AWS account and region”; it states which Bedrock route each model runs on (Haiku 4.5 US-only, Sonnet 5.5 for now Global only), and that where requests are processed is OQ-15, with the default that processing outside the Region and the US is acceptable unless Empire asks for a US-only route. The header’s Last updated line moves to Rev 48.
- Why: lead decisions (Tharinda, 2026-09-30). Sonnet 5.5 replaced Sonnet 5 (AD #19 revised), and AWS’s model pages, read the same day, offer it on Bedrock only through the Global route, which may process a request in any commercial AWS Region; neither pinned model runs in one Region on Bedrock’s standard endpoint. That contradicts the “account and region” wording of the OQ-15 resolution (Rev 12), so the lead reopened OQ-15 for its location part only. The cloud-AI acceptance of 2026-07-28 stands. The lead expects Empire not to mind where the models run, and a US-only route costs 10% more, so the default is to accept processing outside the Region.
- Cascade:
project-prd-signals.md(OQ-15 back as ⏳ open, location only; the 2026-08-19 row in## Resolved Open Questionsabove is kept as the record of the cloud-AI acceptance);client/open-questions.md(Q18, revised 30 September);client/project-proposal.md(the 30 September post-issue note);design/ai/ai-provider-strategy.md§4b and §5 row 9;.claude/CLAUDE.mdAD #19;engagement-log.md(§ AD #19 — models revised 2026-09-30). - Resolves: — (reopens OQ-15; no OQ resolved).
Rev 49 — 2026-10-02
- Changed: §Multi-Tenancy / Org Hierarchy: the default reach counts a share only while its recipient is active; “Deactivation and the hierarchy” says a share to a deactivated user is inert — it reaches nobody, the recipient’s managers included, and counts again on reactivation — and that records shared with a user do not block their deactivation; “Unowned records” says an unowned record cannot be shared. The header’s Last updated line moves to Rev 49.
- Why: the F5-f spec (approved 2026-10-02). The inert-share rule is the lead’s decision at spec authoring (Tharinda, 2026-10-02), closing the question AD #35 and the sharing memo left open. The no-sharing-of-unowned rule follows this PRD’s existing “visible only to holders of
lead.all”, which the memo’s SQL had not respected; since Rev 44 no V1 path writes an unowned lead, so it is a safeguard. - Cascade:
specs/F5-f-team-visibility-scope-spec.md(FR-R1, FR-R3, FR-O1, AC-3, AC-4);design/data/record-sharing-model.md(§2.5, revision 2026-10-02);design/data/entity-map.md(Rev 19, theRecord sharerow);test-strategy.md(Rev 15);client/roles-and-permissions.md(the 2 October note and the sharing section). - Resolves: — (refinement; no OQ resolved).
Rev 50 — 2026-10-03
- Changed: §Non-Functional Requirements, “File uploads & schema extensibility”: uploads stay virus-scanned, and scanning can be switched off only by a deliberate, acknowledged deploy setting; while it is off each upload is recorded as not scanned (never as clean), the switch is logged, audited and counted, and Empire is told before production is switched off. §Risks R-16’s mitigation says the same. The header’s Last updated line moves to Rev 50.
- Why: the lead confirmed GuardDuty Malware Protection on 2026-10-02 and asked that it can be switched off later without breaking file handling (F38-j). The PRD said every upload is scanned with no exception, so the F38-j spec review (Pass 1, finding I7) raised the divergence; the lead chose a PRD revision over an unrecorded operational exception (Tharinda, 2026-10-03). The issued proposal commits only account-level GuardDuty, not file scanning (§A.4), so there is no commercial change; the client-facing PRD’s “virus-scanned” wording stands, and the F38-j operator procedure tells Empire before production is switched off.
- Cascade:
specs/F38-j-scan-off-switch-spec.md(decision 4, FR-9, FR-12). - Resolves: — (refinement; no OQ resolved).
Rev 51 — 2026-10-05
- Changed: §Domain Model › Key Entities, the Lead row: Empire uses both leads and deals in Pipedrive today; a lead is an inquiry not yet qualified, kept in the inbox, a deal is one a rep pursues on a pipeline, and a lead becomes a deal only when a person converts it. The header’s Last updated line moves to Rev 51.
- Why: OQ-5 answered by Dinil (team, from the Pipedrive audit) on 2026-10-05, when
forge-gap-checkreported it as the one Blocker on the F11-b spec. The answer confirms the boundary the PRD already modelled (the Lead row’s “lives in the triage inbox, not a pipeline stage”, and §Reference-System Classification P-05’s “manual add + convert-to-deal preserved”), so no feature’s scope changes. - Cascade:
project-prd-signals.md(OQ-5 row removed); the F11-b, F11-e and F4-b specs no longer list OQ-5 as live. F11-e and F2-c, which OQ-5 also blocked, are unblocked on it. - Resolves: OQ-5.
Rev 52 — 2026-10-05
- Changed (corrections, no new decision): § Phasing / Sequencing Intent, the capture-stream row, and the F36 section (twice): the checkpoint is held during Phase 2, date to be set, and the “weeks 7–13” timing is dropped. The F37 trigger names the Phase 2 scope conversation at the Phase 1 hand-over (Fri 27 Nov), which R-ARCH-001’s 2026-09-29 restatement put in place of the mid-November checkpoint. § Non-Functional › Deployment and Hosting, and § Risks, R-02, as below. The capture-stream row: the checkpoint is held during Phase 2, date to be set, as the Statement of Work of 2026-09-29 says (it read weeks 7–13 and Fri 13 Nov). § Deployment and Hosting: the AWS account is Empire Teas’ own since 2026-09-29 (it read Altrium-owned under SOW ET-002),
us-east-1was confirmed againstus-west-2on 2026-10-01, and OQ-15’s processing-location half is reopened (Rev 48). § Risks, R-02: the same reopening. - Why: the second project audit (2026-10-05) found these sentences describing the state before 2026-09-29 and 2026-09-30.
- Cascade:
design/architecture.md(the same three facts),client/project-proposal.md(post-issue note of 5 October on AD #38),design/prototypes/feature-detail.md(the capture checkpoint). - Resolves: — (correction; no OQ resolved).
Rev 53 — 2026-10-06
- Changed: §Functional Surface > F26. (1) The exchange rate and the overhead are kept per price row, not per week: the LKR-per-USD rate the conversion used is stored on the price row beside the four build-up figures, recorded when the Tea Room has it; the overhead is the difference between the two LKR figures already on the row. (2) The weekly workbook is taken as it is today — no fixed template is required of the Tea Room — and the upload reports every row or cell it cannot read instead of guessing. The sentence “converted to USD at the week’s rate” now reads “at an exchange rate”. The header’s Last updated line moves to Rev 53.
- Why: team decision — Tharinda, 2026-10-06; recorded by Dinil when
forge-gap-checkreported OQ-59 as the one Blocker on the F26-c spec. A per-price rate holds whichever way Empire works (one rate a week or several), and the week-35 workbook already carries two rates in one week (327.5 on one tab, 327 on another), so storing it once per week would already be wrong for that file. With the rate stored, the importer can check the build-up figures against each other, whichdata-model.md§2.5 says the schema cannot do today. Accepting today’s workbook avoids depending on a change to how the Tea Room works each week; the importer handles the quirks recorded indiscovery/source-schemas/weekly-price-workbook-schema.mdand reports anything else. - Not an AD change. AD #14 is untouched: the rate is one more immutable column on the append-only price row.
- Data-model consequence — F26-c’s.
forge-et-be/docs/data-model.md§2.5 gains the rate column and §2.14 item 6 is closed, in the backend PR that carries F26-c’s migration (F26-c spec). - Cascade:
project-prd-signals.mdOQ-59 (update of 2026-10-06; stays partial; F26-c leaves itsBlockscell);specs/F26-f-price-review-publish-spec.md(the rate column hand-off);discovery/source-schemas/weekly-price-workbook-schema.md;design/prototypes/feature-detail.md§F26; the CRM prototype’s seed comment;client/open-questions.mdQ13;.claude/CLAUDE.md§ Active Context;specs/F26-c-weekly-workbook-upload-spec.md. - Resolves: the rate/overhead-grain and template halves of OQ-59. OQ-59 stays partial on the unit of the LKR figures (F26-c builds on LKR per kg, shown on screen) and who produces the sheet.
Rev 54 — 2026-10-06
- Changed: §Lifecycle States, the Lead line:
archived → new(restore) is added besidenew → approved | converted | archived, and Archive / nurture is named as thearchivedstatus — there is no separate nurture status. Archive is allowed fromnewandapproved, never fromconverted. The header’s Last updated line moves to Rev 54. - Why: F11-f spec decisions 2 and 3 (Dinil, 2026-10-06): the PRD’s per-lead actions name “Archive/nurture”, the prototype labels the state “Archived / nurture”, and ongoing nurturing is F25 sequences, so one action and one status suffice; a mistaken archive can be undone, as an account’s can (Rev 36). Built by F11-f (
POST /api/v1/leads/{id}/archiveand…/restore). - Cascade: none beyond the entity map’s Lead row, which the F11-f spec PR already changed.
- Resolves: — (refinement; no OQ resolved).
Rev 55 — 2026-10-08
- Changed: §Functional Surface > F11 (the feature line and the triage section) and the two flows that score leads. (1) The AI priority has three levels, P1–P3, not four. (2) Hot / cold is a second indicator, separate from the priority, and is not AI: a lead is hot while the buyer’s last message (the inquiry that created the lead, or a later email from the buyer linked to it) is no more than 30 days old, and cold after that. Only the buyer’s own messages count; Empire’s emails and follow-ups do not keep a lead hot. The 30 days is a setting. The inbox filters on it, so the team can find the cold leads and follow them up (F11-h).
- Why: lead decision — Tharinda, 2026-10-08, while settling the Phase 1 acceptance criteria. Gate 3’s answer to Q6 (2026-09-17) had made hot/cold a relabelling of the priority (hot = P1–P2, cold = P3–P4), which told the team nothing the priority did not. Measured from the buyer’s replies, it finds the leads that have gone quiet, which is the follow-up problem Empire named on 2026-09-09. Three priority levels are enough to order the work. Before this revision the PRD body did not define hot/cold at all (F11-b spec W-3).
- Not client-facing until the Phase 1 acceptance criteria are issued. The proposal and
client/client-facing-prd.mdsay only “scored for priority with a stated reason”, and neither names P1–P4 or hot/cold; the criteria (E2) state both. The client-approved prototype shows four priority pills, a prototype detail this revision overrides. - Cascade:
features.md(F11-g row and graph label);tracker.yaml(F11-g title);specs/F11-g-ai-lead-ranking-spec.md(Context, In Scope, one open question);design/ui/design-system.mdRev 40 (§5 Priority row, a provisional Hot / cold row, the vocabulary note and the confidence row);design/prototypes/feature-detail.md§F11;client/phase-1-acceptance-criteria.mdE2 (on its own branch, PR #516). Left as written:engagement-gate-runs.mdGate 3 Run 1 Q6 (a point-in-time record). - Resolves: — (lead decision; no OQ resolved).
Rev 56 — 2026-10-08
- Changed: §Functional Surface > F11 (the feature line and the triage section) and the two flows that draft for leads; §Functional Surface > F1 (the feature line and the accounts section). (1) AI drafts for leads are made only when a person asks: a button on any lead generates the reply or re-engagement and shows it for review before anything is sent. Nothing is drafted in advance — not for each new inquiry, and not for the migrated backlog, which is researched, ranked and summarised on day one, with the cold leads found through the hot/cold filter (Rev 55). (2) A contacts page (new feature F1-f, Phase 1) lists every contact the user can reach, with a search, including contacts with no account — Pipedrive people with no organisation that F4-d migrates, and contacts synced later — and lets a user open, edit and attach a contact to an account.
- Unchanged, confirmed: research covers the company and the person (Rev 30). The lead chose full person research over company-only on 2026-10-08, so F11-c’s spec designs the person lookup with a privacy position for buyer personal data, some of it about EU persons (data minimisation, per-field source, a deletion path), before it is built.
- Why: lead decisions — Tharinda, 2026-10-08, settling the open points of the Phase 1 acceptance criteria. Drafting on request avoids AI output nobody reads and keeps every draft current at the moment it is used. Account-less contacts (F1-d spec Q-4) otherwise had no screen in any feature.
- Narrows what Empire was told — stated, not absorbed: the proposal (§3, The value case: “puts a drafted, human-approved re-engagement in front of the team”) and
client/client-facing-prd.md(§3.3: every inquiry “given a drafted reply for review”; the backlog’s “drafted re-engagement list”; §7’s target row) read as drafts made in advance. The proposal gains a post-issue note of 8 October; the client PRD is revised; the Phase 1 acceptance criteria state it (E5, E8). - Cascade:
client/project-proposal.md(post-issue note, 8 October);client/client-facing-prd.md(revision of 8 October: §3.1, §3.3, §7);client/phase-1-acceptance-criteria.md(E1, E5, E8, new C7; open points settled);features.mdandtracker.yaml(new F1-f; F11-h and F11-c notes);specs/F1-f-contacts-list-spec.md(stub);specs/F11-h-ai-drafted-follow-up-spec.mdandspecs/F11-c-lead-research-job-spec.md(Context);owed.md(the account-less contacts row);design/prototypes/feature-detail.md§F11 (the draft is on request; the prototype banner no longer holds). - Resolves: — (lead decisions; no OQ resolved). Answers F1-d spec Q-4.
Rev 57 — 2026-10-09
- Changed: F37 (bulk email) moves from §Deferred (Post-V1), where it was a stretch candidate (Rev 28), to §In Scope (V1) as a Phase 2 feature in two parts: F37-a, the weekly price broadcast to every customer, and F37-b, one message to the contacts of a saved list (newsletters, product promotions, trade-fair invitations) with a summary of sends, deliveries, opens and clicks for each send. The per-send summary is new; Rev 28 limited F37 to F20’s per-email tracking. §Out of Scope loses the “bulk email is not a committed V1 deliverable” bullet; §Functional Surface > F20, the P-13 parity row, the deferred-entities row, the marketing-consent NFR and R-15 are restated to match. F28 (the campaign product: segmentation builder, reports across sends, editor, subscriber management) stays deferred.
- Why: on 2026-10-09 Kavindu Anthony asked whether the CRM will have email campaigns (newsletters with engagement tracking, product promotion, trade-fair invitations and follow-ups, contacts grouped by market and customer type, open and click figures). Lead decision — Tharinda, 2026-10-09: the lead replied that Phase 2 includes bulk email for these uses with opens and clicks tracked per send, and that Phase 2’s features will be prioritised with Empire at or before the Phase 1 hand-over. Record:
discovery/meeting-notes/2026-10-09-kavindu-bulk-email.md. - Adds to what Empire was told — stated, not absorbed: the issued proposal places email campaigns among the later-phase Pipedrive add-ons and excludes a full marketing suite; bulk sending is added scope (L-001). The relief is a compensating descope of other Phase 2 work, agreed with Empire in that prioritisation meeting (R-ARCH-001, whose Phase 2 shortfall grows by two features). The proposal gains a post-issue note of 9 October.
- Cascade:
client/project-proposal.md(post-issue note, 9 October);client/client-facing-prd.md(§ out of scope, revision of 9 October);client/phase-1-acceptance-criteria.md(the not-in-Phase-1 list);client/open-questions.md(Q10);project-prd-signals.md(OQ-62);features.mdandtracker.yaml(F37-a, F37-b in Phase 2, the stretch section removed, R-ARCH-001 item 8, F20-c and F26-d notes);specs/F37-a-weekly-price-broadcast-spec.mdandspecs/F37-b-bulk-email-to-a-list-spec.md(stubs); the specs that pointed at the F37 stretch;design/prototypes/feature-detail.md;owed.md;.claude/CLAUDE.md § Active Context. - Resolves: — (lead decision; no OQ resolved). OQ-62 still blocks F37-b.
Rev 58 — 2026-10-09
- Changed: §Non-Functional › File uploads & schema extensibility, and risk R-16’s mitigation. Virus scanning of uploads (GuardDuty Malware Protection, behind the FS-11 seam) is built but switched off in every environment by F38-j’s acknowledged setting, so every upload is recorded as not scanned; type and size validation still apply. GuardDuty’s account-wide threat detection is also off.
- Why: Altrium management’s decision (Thilina, with Malinda), 2026-10-09, relayed to Shyamin Ayesh by Tharinda: go without GuardDuty and add it later only if it is needed or Empire asks. The scanner code is kept, so turning it on needs AWS setup and two settings, not a code change.
- Narrows what Empire was told — stated, not absorbed: the proposal (§A.2.2 “Threat detection (GuardDuty)”, §A.4) and
client/client-facing-prd.md(§3 “uploads are validated and virus-scanned”, §5 “continuous threat detection”). The proposal gains a post-issue note of 9 October; the client PRD is revised; the cost annexe marks the GuardDuty line not enabled. Empire is told before production (owed.md). - Cascade:
client/project-proposal.md(post-issue note, 9 October);client/client-facing-prd.md(revision of 9 October: §3, §5);client/sow-cost-annexe.md(GuardDuty row);design/architecture.mdRev 52 (security baseline, the virus-scanning row, AD-O-4);specs/F38/cloud-bring-up-spec.mdRev 7, with notes in the F38-a, F38-b, F38-e and F33 specs;cloud-gated-validation.md(§B FS-11 rows withdrawn or narrowed; A-1, §C-8);design/environments-and-identity.md(§4 item 8, §5);owed.md;tracker.yaml(F38-b note). - Resolves: — (management decision; no OQ resolved).
Rev 59 — 2026-10-10
- Changed: §Domain Model › Key Entities gains a Note row: free text on an account, contact or lead, with its author and its last editor; any user who may manage the record may add, edit or delete any note on it; a deleted note is marked deleted and kept, never removed, and disappears from every screen with its timeline entry; only an account’s notes appear on its timeline. The Tag row reads renaming or retiring a name is configuration (it read renaming or deleting): a tag is retired, never deleted, stays on the records that carry it, and its colour is one of the design system’s badge tones. §Per-Role Capability Matrix’s
Dparagraph names notes as deleted by marking. The header’s Last updated line moves to Rev 59. - Why: the F1-e spec (notes and tags on accounts and contacts, drafted and reviewed 2026-10-10 by Biseni Perera) settles note editing and deletion for every record type, which F11-f left to it (F11-f FR-N4), and tag configuration under
settings.manage. The PRD had no Note entity — notes appeared only as a timeline kind — so the spec’s record needed one. Marking rather than removing follows the matrix’s rule that a hard delete is a lead decision; retiring rather than deleting a tag keeps the audit events F4-a’s final migration run reads pointing at a tag that exists. - Does not change what Empire was told:
client/client-facing-prd.mdand the Phase 1 acceptance criteria (C4, C1) say notes and tags are added to accounts and contacts and that a note counts as contact; neither says how a note is deleted or a tag removed from use. - Cascade:
specs/F1-e-notes-and-tags-spec.md(Decisions 2 and 6, FR-N4, FR-G4);design/data/entity-map.md(Note and Tag rows);discovery/inventories/feature-inventory.md(labels row). - Resolves: — (spec decision; no OQ resolved).
Rev 60 — 2026-10-11
- Changed: global search (F31) leaves V1. §In Scope loses the F31 bullet and §Out of Scope gains a Global search exclusion with its reason. §Problem Statement’s Pipedrive feature list, the §Reference-System Classification row P-30 (now Won’t build), the §Phasing / Sequencing Intent P0 row, the §Per-Role Capability Matrix’s Global search row (removed), §Functional Surface › F31 (now a dropped stub) and the §Success Criteria parity line follow it. The header’s Last updated line moves to Rev 60.
- Why: lead decision, 2026-10-11. Each list screen carries its own search over the records the user reaches (accounts F1-b, leads F11-b, users F5-c, every later list through F32), and the revised app frame drops the top bar that held the inert search box (forge-et-fe#201). A search across every record type adds an index design (AD-O-2) and a results screen for little the lists do not already give, while Phase 2 is already short of capacity (R-ARCH-001).
- What Empire was told: the issued proposal names global search (scope summary, §2, §6/§7, the parity line), and so do the issued
client/client-facing-prd.md(Section 3, Phase 2) andclient/project-timeline.md(§2, Phase 2). The lead decided Empire is not told separately, so those two issued documents are left as issued; whoever next revises and sends either one removes the line. The proposal carries a post-issue note. The two drafts not yet issued —client/phase-1-acceptance-criteria.mdandclient/roles-and-permissions.md— no longer mention it. - Cascade:
tracker.yamlandfeatures.md(F31dropped, recorded in Phase 1);client/project-proposal.md(post-issue note, 11 October);design/architecture.md(§ Search, AD-O-2 closed, Rev 53);design/data/entity-map.md(Search row, Rev 25);design/ui/design-system.md(§6 Search, Rev 46);design/prototypes/feature-detail.md(F31 section and the drift row);specs/F31-global-search-spec.md(dropped note);owed.md(removing the inert top-bar search box from forge-et-fe). Done specs and foundation specs that point at F31 (F1-b, F11-b, F20-b, F5-b; FS-2, FS-3, FS-10) are records and stay as written. - Resolves: — (lead decision; no OQ resolved).
Rev 61 — 2026-10-11
- Changed: three Phase 2 features leave V1: the AI product-brief assistant (F16), AI similar past quotes (F7-e) and the price-lookup page (F26-e). §Out of Scope gains an exclusion for each. §In Scope loses the F16 bullet, and the F7 bullet loses similar-past-quote matching. §Phasing / Sequencing Intent’s P1b row drops F16. §Users and Access’s AI a user invokes list drops the product-brief assistant. §Functional Surface: the F7 section no longer surfaces similar past quotes, F14 no longer surfaces comparable past quotes, F16 becomes a dropped stub, and the F26 paragraph’s price-lookup sentence is replaced by where each reader finds a price. The quote workflow sketch drops shows similar past quotes. The header’s Last updated line moves to Rev 61.
- Why: lead decision, 2026-10-11, to reduce Phase 2, which does not fit its build window (R-ARCH-001). F16 was already the first descope lever, and a rep fills the brief by hand. F7-e is a convenience; earlier quotes are on the account and in the quote list. F26-e mostly repeated what F26-b’s Standards book, the account page and the Tea Room’s round (F26-f) already show; past prices stay in the effective-dated rows (AD #14), with no history screen.
- What Empire was told: the issued proposal names the product-brief assistant in its AI layer (§2), so it carries a post-issue note (11 October, second note). It does not name similar-quote matching or a price-lookup page. The issued
client/client-facing-prd.mdsays similar past quotes are surfaced (Section 3, quotes); the lead decided Empire is not told separately, so it is left as issued, and whoever next revises and sends it removes that clause. The draftclient/phase-1-acceptance-criteria.mdno longer lists the price-lookup page. - Cascade:
tracker.yamlandfeatures.md(F7-e, F16, F26-edropped, recorded in Phase 1; graph and lane table);client/project-proposal.md(post-issue note);design/ai/ai-provider-strategy.md(the F16 cost row);design/architecture.mdanddesign/ui/design-system.md(the AI-surface lists);design/prototypes/feature-detail.md(F16 section, the similar-quotes panel);cloud-gated-validation.md(§C AI-features row and## Log); the stubs of F16, F7-e, F26-e, F14, F7-a and F7-c. Approved or done specs that point at them (F12’s Blocks line, F26-a, F26-b, F26-f) are left as written; F12 and F26-f are in progress with their owners. - Resolves: — (lead decision; no OQ resolved).
Rev 62 — 2026-10-11
- Changed: §Functional Surface › F26’s bulk-pricing paragraph states that a published price whose week starts today or later can be withdrawn — marked withdrawn with who, when and why, never changed or deleted, no earlier price reopened — so a corrected price can be published for that week. The header’s Last updated line moves to Rev 62.
- Why: AD #14’s second exception (lead sign-off 2026-10-11), built by F26-f (spec FR-1, FR-10). It was recorded in the AD table and the spec but not yet in the PRD body.
- What Empire was told: the lead asked on 2026-10-11 for the client documents to be revised with this and the Rev 60 and Rev 61 drops. This supersedes the “left as issued” lines in Rev 60 and Rev 61:
client/client-facing-prd.md(header line of 11 October; Sections 3.1, 3.2, 3.3 and 4),client/project-timeline.md(header line; the Phase 2 rows of Sections 1 and 2, which also now list bulk email, added to Phase 2 by Rev 57) and the draftclient/roles-and-permissions.md(header; the product-brief and standards-book rows) are revised. Revising them in the repository sends nothing; when they reach Empire is the lead’s call. - Cascade:
design/data/entity-map.md(the overview’s search line). - Resolves: — (records an AD; no OQ resolved).
Rev 63 — 2026-10-11
- Changed: two Phase 2 features leave V1: creating custom fields during an import (F23-c) and the capture app’s same-day follow-up email (F36-f). §Out of Scope gains an exclusion for each. §In Scope’s F36 bullet and §Functional Surface › F36 record the same-day email as dropped; §Functional Surface › F23 maps columns only to fields that exist. The header’s Last updated line moves to Rev 63.
- Why: lead decision, 2026-10-11, to reduce Phase 2 (R-ARCH-001). F23-c is a convenience: an administrator creates the field first. F36-f needed its own sender and approval design (OQ-60), still unanswered by Empire; the same message can go to a show’s contacts through bulk email to a saved contact list (F37-b), or a rep emails a lead from the CRM (F20).
- What Empire was told: F36-f was Empire’s own ask (2026-09-09), and question Q15 (OQ-60) went to them on 2026-09-27. Q15 is marked withdrawn in
client/open-questions.md, andclient/client-facing-prd.mdgains a revision line and a Section 4 entry. The proposal predates both asks and names neither, so it needs no note. - Cascade:
tracker.yamlandfeatures.md(F23-c and F36-fdropped, recorded in Phase 1; graph and lane table);project-prd-signals.md→project-prd-history.md(OQ-60);cloud-gated-validation.md(the F36-f row and## Log); the stubs of F23-c, F36-f, F23-b, F34-a, F36-c, F36-d and F20-c;design/prototypes/feature-detail.md(F36). - Resolves: OQ-60 (closed as not needed).
Rev 64 — 2026-10-11
- Changed: the capture app’s offline queue (F36-e) leaves V1, and F36-a gains a keep-and-retry safety net. A capture whose upload fails is kept on the phone, shown as not sent, and sent again when the app is open with a connection; there is no background sync or full offline mode. §In Scope’s F36 bullet and §Functional Surface › F36 say so; §Out of Scope gains the exclusion. The header’s Last updated line moves to Rev 64.
- Why: lead decision, 2026-10-11, to reduce Phase 2 (R-ARCH-001). Offline capture was our own design, not a promise: neither the proposal nor the client-facing PRD mentions it, and Empire never answered Q23 (offline behaviour), which was itself framed as a descope candidate. iPhone browsers do not support background sync, so a full queue would still send only when the app is open. The retry keeps the one real risk covered — a card lost at a busy stand — for a fraction of the work.
- What Empire was told: the proposal does not name offline capture.
client/open-questions.mdQ9 did tell Empire that working with no signal was agreed for this version; Rev 65 records the correction and Q9 gains a dated note (no capture is lost — it is kept and resent). - Cascade:
tracker.yamlandfeatures.md(F36-edropped, recorded in Phase 1; F36-a’s note);project-prd-signals.md(OQ-55: the offline gap and its Blocks column);design/architecture.md§ Trade-Show Capture App (Rev 54);design/ui/design-system.md§9’s capture-app offline states (Rev 47);design/coding-standards.md(the capture-app cache note);design/data/entity-map.md(the capture-app line);design/prototypes/feature-detail.md(F36); the stubs of F36-a and F36-e. - Resolves: — (lead decision; OQ-55 stays partly open on the badge QR encoding).
Rev 65 — 2026-10-11
- Changed (corrections, no new decision): an independent consistency review of Revs 60–64 found places they missed or stated wrongly. (1) Price history: Rev 61 said there is no price-history screen, but F26-f (done) shows each standard’s published prices under History in the Tea Room’s weekly round (FR-22), and Phase 1 criterion G5 relies on it; §F26 and the F26-e exclusion now say so — only the separate price-lookup page is dropped. (2) §In Scope’s F14 bullet no longer surfaces comparable past quotes (missed by Rev 61). (3) §In Scope’s F23 bullet, row P-09 and the F23 reconciliation note no longer create custom fields on import (missed by Rev 63). (4) Row P-30 is classified Redesign (to per-list search), one of the four classes the legend allows, instead of “Won’t build”. (5) The phasing note no longer lists search in Phase 2. (6) §F26 no longer says “there is no bulk circular” — F37-a (Rev 57) sends the week’s prices to every customer. (7) The F36-e exclusion cites Q23 of the 2026-09-03 requirements questions and records that
client/open-questions.mdQ9 had told Empire offline capture was agreed. (8) The AI-feature ranges read F11–F15, and the Functional Surface intro names its range. - Correction to Revs 60 and 61: they cite the proposal’s “scope summary, §2, §6/§7, the parity line”; the passages are §3 Success criteria (the parity line), §4 Scope (the core CRM list and the AI layer) and §6/§7 (the phase plan). The proposal’s 11 October notes are corrected by a third note of that day, since earlier notes are not edited.
- Cascade:
client/open-questions.md(Q9 note);client/project-proposal.md(third 11 October note);client/client-facing-prd.md(search no longer listed in Phase 2);client/phase-1-acceptance-criteria.md(a check for price withdrawal);client/roles-and-permissions.md(price history only for the Tea Room); and the internal documents the review named —features.md,tracker.yaml,engagement-log.md,owed.md,cloud-gated-validation.md,environments-and-identity.md,design/ui/design-system.md,design/prototypes/feature-detail.md,design/data/entity-map.md,design/architecture.md,design/ai/ai-provider-strategy.md,client/project-timeline.md,project-prd-signals.md(OQ-67 also blocks F18-a),.claude/CLAUDE.md(Active Context), the F15-a, F17-a, F18-a, F26-e, F36-e, F38-a and F38-d specs. - Resolves: — (corrections).
Rev 66 — 2026-10-11
- Changed: the five Pipedrive add-on equivalents are dropped, not just deferred. F27 (website chatbot and live chat), F28 (the email-campaign product), F29 (purchased data enrichment), F30 (LinkedIn integration) and F35 (sales goals and targets) had been out of V1 since 2026-08-20 and listed under §Deferred (Post-V1) as intended later. They leave §Deferred and become one §Out of Scope bullet: not built in V1 and not planned after it. Rows P-23 to P-26 and P-33 move from Defer to Discard; the phasing table’s Post-V1 row, the deferred-entities row, the LinkedIn and data-enrichment integration rows, the marketing-consent and third-party-data security notes, the compliance note, the marketplace and marketing-suite exclusions, and risks R-13 and R-15 say so; the add-on row leaves §Per-Role Capability Matrix. The header’s Last updated line moves to Rev 66.
- Not affected: F37 (the weekly price broadcast and the bulk send to a saved list, Phase 2, Rev 57) and the AI lead research that reads public sources (F11-c and F36, Rev 30). Both were always distinct from F28 and F29.
- Why: decision of 2026-10-11 that none of the five is expected to add enough for Empire to justify building it. The lead signs off by approving the PR that carries this revision. The same change gives each of the five a
droppedrow in Phase 1 oftracker.yaml, so the dashboard lists every PRD feature. - What Empire was told: the issued proposal lists the add-on equivalents as intended for a later phase pending the Pipedrive audit, and
client/client-facing-prd.mdcalled them post-V1 items. The proposal gains a post-issue note; the client-facing PRD now says they are not planned. Whether Empire is told separately is the lead’s call (owed.md). If Empire turns out to rely on one of them today, it goes through the change process (R-13). - Cascade:
tracker.yamlandfeatures.md(F27, F28, F29, F30 and F35 as dropped Phase 1 rows);project-prd-signals.md(OQ-23 and OQ-29 now serve only as early warning, and block nothing);client/project-proposal.md(post-issue note);client/client-facing-prd.md(Section 4 and a dated header line); the F17-b record stub;design/prototypes/feature-detail.md;design/ai/lead-research-source-options.md;owed.md;engagement-log.md; the F11-c, F17-a and F37-b stubs. Left as they are: the discovery inventories, research and meeting notes, which are dated records, and the comments inside the CRM prototype. - Resolves: — (decision; OQ-23 and OQ-29 stay open as early warning).
Rev 67 — 2026-10-11
- Changed: §Non-Functional › File uploads & schema extensibility. “Empire is told before production” (Rev 58) becomes “the client-facing PRD states it”: the lead decided on 2026-10-11 that Empire is not told separately that GuardDuty is off, because the client-facing PRD (revised 9 October) and the proposal’s post-issue notes of 9 and 11 October already say so.
- Why: the decision closed the
owed.mditem on 2026-10-11 (engagement-log.md, 2026-10-11), and this sentence still described the closed step. - Cascade:
specs/F33-file-attachments-spec.md(the 9 October note). The register,environments-and-identity.mdand the proposal were corrected on 2026-10-11. - Resolves: — (lead decision; no OQ resolved).
Rev 68 — 2026-10-12
- Changed: §Functional Surface › F20 (summary line and detail section). Reps send four kinds of email from the CRM — new, reply, reply-all and forward (a forward keeps the original’s attachments) — with formatted text (bold, italic, lists, links). Attaching a new file to an outgoing email is a separate Phase 1 feature, F20-h (Email attachments). Two deliberate divergences from Pipedrive (§Reference-System Classification P-12, Replicate): no signature is added (Microsoft Graph does not add or expose the Outlook signature — OQ-68 asks Empire whether they want a CRM signature setting), and one send goes to at most 50 recipients, so it is not a bulk send (F37).
- Why: the F20-c spec (2026-10-12) answered Q16 of the 2026-09-03 question list (reply, reply-all and forward, never answered), and its review found that leaving out new attachments, formatting and the signature under a Replicate classification needed the lead’s sign-off. Tharinda decided on 2026-10-12 to keep formatting in F20-c, because every feature that fills a body (F11-h, F20-d, F26-d) depends on what a body is, and to build attachments as F20-h, so the features waiting on F20-c do not wait for attachments; and confirmed the signature and recipient-cap divergences.
- Cascade:
specs/F20-c-compose-and-send-spec.md(decisions 2 and 7, Context “Parity with Pipedrive”); newspecs/F20-h-email-attachments-spec.mdstub;tracker.yamlandfeatures.md(F20-h);cloud-gated-validation.md(backlog-table row for F20-h);project-prd-signals.md(OQ-68 raised; OQ-69 raised and withdrawn);owed.md. - Resolves: OQ-69 (withdrawn).
Rev 69 — 2026-10-12
- Changed: nothing in the PRD body.
## Resolved Open Questionsgains rows for OQ-32, OQ-34, OQ-35 and OQ-36, each marked withdrawn. They were raised on 2026-07-17 and removed from the signals file (OQ-34, OQ-35 and OQ-36 on 2026-07-17, OQ-32 on 2026-07-22) without a history row, so their numbers appeared nowhere. Three other numbers were reused at the same time: OQ-38 first asked about volume and quality price tiers, OQ-39 about CC/BCC on CRM email, and OQ-40 about lead-source tracking; those three were removed on 2026-07-17 and their numbers given to new questions on 2026-07-22 (destination markets, weekly price ownership, margin approval), so this entry is their only record..claude/rules/prd.md § Conventionsnow says an OQ dropped without an answer moves here as withdrawn and is never deleted. - Why: the lead decided on 2026-10-12 that no hand-allocated number is skipped, reused or deleted (
.claude/rules/id-numbering.md); an audit of every numbered id found these four OQ numbers unused and the three reused. - Cascade:
.claude/rules/prd.md(Conventions). No other file names these OQ numbers. - Resolves: OQ-32, OQ-34, OQ-35, OQ-36 (withdrawn).
Rev 70 — 2026-10-13
- Changed: § Non-Functional › Migration — the Pipedrive migration’s write runs may answer in the request, each in one transaction, while they fit the load balancer’s idle timeout, and become a background job otherwise; “writes as a batch job” becomes “writes as one batch”. § Non-Functional › Email/sync throughput — “imports run as background jobs” gains the same exception for the Pipedrive migration.
- Why: the F4-d spec (Decision 2, developer decision 2026-10-13; approved the same day). A write run is made twice in the project’s life, over about 9,900 rows; one transaction makes a half-written run impossible, and a background job would have to keep Empire’s export files in S3 between the request and the job and needs a job framework and a status endpoint the backend does not have. The dry run (F4-a) and the
proposalDigestthe write run must send back are the validation the PRD asks for. Tharinda (lead) is asked to confirm on the F4-d spec PR (F4-d Q-2). - Cascade:
.forge/design/architecture.md(§ Components › Background Worker and Job Queue, § Data Flow item 3, the AD-O-7 row’s status); a post-issue note on.forge/client/project-proposal.md(its §5 diagram and prose put the migration on the queue); F4-d spec FR-H1. - Resolves: none.
Rev 71 — 2026-10-14
- Changed: §Role-based access, the V1 key table:
price.publishis held only besidebuying_standard.vieworbuying_standard.manage. It was an independent key. - Why: BUG-167 (the 2026-10-08 test pass). A role could hold “Publish weekly prices” with Pricing set to None, and the holder could neither reach the Pricing page nor read a price round, which F26-f allows only with
price.publishtogether withbuying_standard.view. Tharinda decided on 2026-10-14 that the catalog refuses the combination: the API refuses it with422 REQUIRED_PERMISSION_MISSING, the role editor clears and blocks the tick while Pricing is None, a migration removes the grants of that shape already stored, each audited, and the startup grant-consistency check refuses one. - Cascade:
specs/foundation/006-auth-scaffolding-spec.md(Rev 16, Appendix A.1);specs/F5-d-roles-editor-spec.md(Rev 3, FR-1, FR-7, FR-15);design/ui/design-system.md(Rev 57, the §9 model note);owed.md(the change request settled);bugs.mdandtracker.yaml(BUG-167).
Rev 72 — 2026-10-15
- Changed: §Role-based access, the capability matrix’s Email Template row, restated in the catalog’s terms: every user who can send (
email.manage) uses the shared templates and creates, edits and deletes their own personal ones; the shared templates are created, edited and deleted undersettings.manage; a role withoutemail.manageuses none. The R&D column, which read “R”, now reads as the Sales Rep’s, matchingclient/roles-and-permissions.md§4.4. The row had read as separate actions per role (“C, U (own)”, “C, U (all)”), which the minimal catalog cannot express. The note under the matrix now says email templates are deleted by marking, as notes are, that the weekly-prices template cannot be deleted, and that a personal template is its owner’s alone. - Why: the F20-d spec (approved 2026-10-15) settles both, as the note asked of the feature’s spec (its decisions 2 and 5, FR-P1 to FR-P4, FR-T5).
- Cascade:
specs/foundation/006-auth-scaffolding-spec.md(Rev 17, Appendix A.3);design/data/entity-map.md(Rev 30);project-prd-signals.md(F20-d leaves OQ-62’sBlockscell);owed.md.client/roles-and-permissions.md§4.4 already reads this way and does not change. - Resolves: — (no OQ resolved; OQ-25 and OQ-62 were waived for F20-d’s spec, not answered).
Rev 73 — 2026-10-16
- Changed: §Domain Model, the Lead row: the draft reply is generated on request and not kept — “draft reply (generated on request and not kept, Rev 56, Rev 73)”. It had read “draft reply (generated on request, Rev 56)”, which left open whether a draft is stored on the lead.
- Why: the F11-h spec (approved 2026-10-16), decision 4: a draft is a proposal for one email; what was sent, and whether it was sent as drafted or edited, is already on F20-c’s
email_sendrow (ai_drafted,edited_by_human), and the call itself on F12’sai_invocationrow. Keeping drafts would add a column holding buyer details and AI text, an audit trail per draft, and drafts that go stale as the thread moves, while drafting again costs about a US cent. The panel keeps the draft in the browser until it is sent, discarded or the page is left (F11-h FR-V1, FR-D8). Built by F11-h-WI-1.1 (POST /api/v1/email/ai-draftswrites only theai_invocationrow and itsai.invokedevent). - Cascade: none beyond the Lead row —
git grep "generated on request"finds no other statement of it outside dated records. - Resolves: — (no OQ resolved).
Rev 74 — 2026-10-17
- Changed: §F20, one sentence: only email linked to a record is stored, in full with its attachments; the rest of the mailbox is read live from Microsoft; linking is automatic (address or thread) and manual; a record or mailbox added later gets its past mail linked; users and the administrator can block addresses and domains; the mailbox connects itself at sign-in; “each user connects their own mailbox” in §F20 and in §Domain Model’s MailboxConnection row now reads “each user’s own mailbox is connected”, and the User row’s “connects their own mailbox/calendar” now reads that they are connected at sign-in when the role holds
email.manage. F20-f is resliced (/forge-decompose --reslice --apply, capability-cluster cut) into F20-i (the live inbox and connecting at sign-in), F20-f (linking and the timeline, keeping its id), F20-j (past-mail linking) and F20-k (block lists), all Phase 1, priority high. The audit tripped S1 (cross-layer spread), S3 (four capability clusters) and S4 (capability density); S2 did not trip. The lead accepted the cut on 2026-10-17. - Why: Kavindu Anthony asked on 2026-10-16 why whole mailboxes are synced (
discovery/meeting-notes/2026-10-16-kavindu-progress-update.md§5); the lead’s decisions of 2026-10-17 are indesign/architecture.md§ Mailbox integration (Rev 62, Rev 63). The new ids followF<n>-<letter>(.claude/rules/id-numbering.md), so F20-f keeps its id and the authored specs that cite it for linking (F1-a, F1-c, F1-d, F11-f, F11-g, F20-c, F26-d) stay correct. - Cascade:
tracker.yaml(F20-i, F20-j, F20-k added; F20-f’sblocked_bygains F20-i;setup.decomposition.reslice_runs);features.md(rows, graph, lane C); the F20-f stub narrowed and three stubs created;engagement-gate-runs.md(Gate 3 Run 2 — Reslice);owed.md. The client-facing PRD and the Phase 1 acceptance criteria are updated with the F20 specs (owed.md). - Resolves: — (structural, no OQ resolved).
Rev 75 — 2026-10-21
- Changed: §F11’s research paragraph: the system searches public sources for the company, and the role of the person named on the lead is taken from what those sources state; the person’s name is not sent to the search service, and a search for the person is added only once Empire Teas agrees. §Integration Points’ web-search row: one search per lead for the company, a second for the person only when switched on after Empire agrees; the provider is Exa on its free credit, provisional (SP-ARCH-004), replacing the sentence preferring an in-account AWS option.
- Why: the F11-c spec (approved 2026-10-21) chose Option 1 for the person lookup (Dinil, 2026-10-21; spec decision 5): Exa’s terms give it a perpetual licence over search text (
design/ai/lead-research-source-options.md§12, §13), so a person’s name, which is personal data, is kept off it until Empire, as controller, agrees. The provider was chosen provisionally on 2026-10-20 (that doc’s §11, Revs 8–9). - Cascade: the F11-c spec FR-H2:
owed.md(raising with Empire that the proposal’s post-issue note of 16 September, the client-facing PRD §3.3 and §4 and the draft criterion E1 say the person is looked up — the issued and client-facing documents are not edited until Empire answers);project-prd-signals.md(OQ-15’sBlockscell); the entity map;ai-provider-strategy.md§2; the cloud register; the F11-d, F11-h, F11-g, F36-c and F4-b specs. - Resolves: — (no OQ resolved; OQ-15 waived for F11-c, not answered).
Rev 76 — 2026-10-21
- Changed: §F13 and its line in §Functional Surface: the account briefing is made when a person whose role holds
account.manageasks for it, reads the account’s profile and communication timeline, reads no order history in V1, and offers a re-engagement prompt on a dormant account in place of the reorder/upsell trigger; nothing is stored. “Where the AI sits”: “account briefing generation” moves from AI the system runs on its own to AI a user invokes on a record, implied byaccount.manage. - Why: the F13 spec (
specs/F13-ai-account-briefing-spec.md), decisions 1–3. OQ-63 (past orders) was waived at the F13 gap check (Dinil, 2026-10-21) to its default (b), so the trigger that read order history has nothing to read; the dormancy rule of F1-c replaces it. The Phase 1 acceptance criterion E6 has a person ask for a briefing, and a view-only role must not start AI work, so the lead moved the briefing within AD #32 (Tharinda, 2026-10-21;engagement-log.md§ Architecture Decisions — full rationale archive). - Cascade:
.claude/CLAUDE.mdAD #32;design/architecture.md§ AI data-handling boundary;client/client-facing-prd.md§3.3 and its revision line;client/roles-and-permissions.md(Where the AI fits); the proposal header (post-issue note of 21 October 2026);specs/foundation/006-auth-scaffolding-spec.mdRev 18;project-prd-signals.mdOQ-63 (waiver note,Blockscleared);owed.md. - Resolves: — (OQ-63 stays open with Empire; F13 builds its default).
Rev 77 — 2026-10-26
- Changed: §F20: a failed mailbox connection never stops the sign-in; a contact may hold several email addresses, and mail to or from any of them links (the Pipedrive migration keeps every address); a lead keeps one address; a converted lead’s email is also linked to the account, contact and deal it becomes. §Domain Model’s Contact row reads “a main email address and any others” for “email”. Reslice: F20-l (Several email addresses per contact, Phase 1, high) is split from F20-f, which keeps the matching (substrate-cut;
engagement-gate-runs.md § Gate 3 Run 3 — Reslice). - Why: the lead’s decisions of 2026-10-26 on the three open email questions raised on 2026-10-19 (
design/architecture.md§ Mailbox integration, Rev 74). 246 people in the Pipedrive people export have several work addresses in one cell, and keeping only the first would leave mail from the others unlinked. The address work has no email dependency and must land before the first production run of the migration, while F20-f waits on F20-b, F1-c, F11-f and F20-i, so it is its own feature (the lead, 2026-10-26). - Cascade:
tracker.yaml(F20-l added; F20-f’sblocked_bygains F20-l;setup.decomposition.reslice_runs);features.md(rows, graph, lane C); the F20-l stub;engagement-gate-runs.md(Gate 3 Run 3 — Reslice);design/architecture.md(Rev 74);design/data/entity-map.md(Rev 34); the F1-a, F4-a, F4-d, F11-d and F20-a revisions; the F20-f, F20-i, F20-j, F20-k, F11-e, F4-b, F2-c, F22, F23-b and F36-a stubs;discovery/source-schemas/pipedrive-people-export-schema.md;guides/email-in-the-crm-explained.md;owed.md. The client-facing documents still read correctly (they speak of a buyer’s known address and a connection with nothing to press) and are not changed. - Resolves: — (structural, no OQ resolved).
Rev 78 — 2026-10-26
- Changed: statements that earlier revisions left behind, corrected to match decisions already recorded; no requirement changes. (1) §Domain Model’s File Attachment row and §F33: uploads are validated and, while virus scanning is switched off, recorded as not scanned (Rev 58 had changed only the file-uploads NFR and R-16). (2) §Domain Model › Lifecycle States, Lead: the AI priority is
P1|P2|P3(Rev 55 had changed §F11 only). (3) §F32 and its line in §Scope and Boundaries › In Scope: the bulk actions list archive, not delete, matching the archive-not-delete default beside the capability matrix and the F32-b stub. (4) §Users and Access, AI a user invokes: the example of the dropped product-brief assistant (Rev 61) is replaced by the quote assistant. (5) §Users and Access › Multi-Tenancy / Org Hierarchy: Empire was told of the 2026-09-24 access-model change in the update of 6 October 2026, so it is no longer owed. (6) §Business Specifics (current tooling), §F4 and R-04: no Pipedrive audit is coming (2026-10-06); the record volumes are counted (OQ-6, Rev 44), and what the exports cannot answer is asked of Empire directly (OQ-22, OQ-23, OQ-26); R-04’s mitigation states the import approach of §F4 (OQ-21). (7) R-16: the row repeated its id and item text, which shifted its Owner and Status cells; the repeat is removed. - Why: a review of the harness documents on 2026-10-26 found each statement contradicting the PRD’s own later revisions or the signals file.
- Cascade:
project-prd-signals.md(the assumptions paragraph on OQ-15; theBlockscells of OQ-17, OQ-25 and OQ-67); the client documents corrected in the same change (client/open-questions.mdPart 2,client/dependencies.md,client/roles-and-permissions.md,client/client-facing-prd.mdSection 8,client/sow-cost-annexe.md,client/project-timeline.md). - Resolves: — (no OQ resolved).
Rev 79 — 2026-10-26
- Changed: the permission key
access.manageis folded intosettings.manage, which is now the one key for all configuration and all user, role and reporting-line administration, and opens every page of the CRM’s Settings page. §Users and Access › Roles (the Administrator row), the Per-Role Capability Matrix’s status note and its Users & roles row, §Permission keys (onesettings.managerow, labelled “Manage settings, users and roles”, in place of the two; the retired administration keys fold intoaccess.manageand then intosettings.manage; the full user list behindsettings.manage), §Special keys (the MD note and the separation-of-duties note), §Multi-Tenancy / Org Hierarchy (who sets the manager), the user-administration, roles and ownership paragraphs (the seeded IT Admin holdssettings.manage; the last-admin invariant and the deactivation refusal name it; the user list and the effective-permissions viewer sit behind it), §Scope’s won’t-build reason, §Domain Model’s User row and §Security namesettings.manage. The capability matrix’s Email Template row: a role holdingemail.managewithoutsettings.manageuses templates in the compose dialog and creates or edits none, personal templates included; a personal template is written by a holder of both keys. - Why: design-system Rev 70 put Users, Roles and the email templates on the Settings page, and Nanndish decided on 2026-10-26 that the Settings permission alone opens the whole page; the seeded IT Admin held only
access.manage, so the two keys were folded into one, and one key per Settings page was more than Empire has asked for. He decided the same day that writing email templates, personal ones included, needssettings.manage. Full text:.claude/CLAUDE.mdAD #34 andengagement-log.md § Architecture Decisions — full rationale archive, “AD #34 —access.managefolded intosettings.manage, 2026-10-26”. The code change isforge-et-bebranchfeature/settings-manage-absorbs-access(with the migrationV54, which moves everyaccess.managegrant, audited; forge-et-be#319) and forge-et-fe#337. - Cascade:
.claude/CLAUDE.md(AD #32, AD #34, AD #37, Active Context);engagement-log.md;owed.md;design/ui/design-system.md(Rev 71);design/architecture.md(Rev 79);design/data/entity-map.md(Rev 36);design/data/record-sharing-model.md;design/data/roles-and-permissions-key-map.md;design/prototypes/feature-detail.md;design/ui/shadcn-backfill.md;test-strategy.md(Rev 29);cloud-gated-validation.md;guides/tech-stack-reference.md;guides/aws-setup-explained.md;.claude/agents/backend-reviewer.md;client/roles-and-permissions.md(its 26 October revision line; the issuedclient/client-facing-prd.mdis anowed.mdrow); the FS-6 spec (Rev 19) and the F1-e (Rev 7), F5-b (Rev 4), F5-c (Rev 10), F5-d (Rev 6), F5-e (Rev 1), F5-h (Rev 2), F5-k (Rev 3), F20-d (Rev 2), F38 (Rev 30), F38-d (Rev 5), F38-f (Rev 5) and F38-h (Rev 5) specs; dated notes on the draft F7-d and F32-c stubs. - Approved by: Nanndish, 2026-10-26; the AD #34 refinement signed off by Tharinda, 2026-10-27.
- Resolves: — (no OQ resolved).
Rev 80 — 2026-10-27
- Changed: §Non-Functional Requirements › Accessibility, Supported sizes: the sentence “Both are tested in Chromium and WebKit … and the accessibility and responsive browser checks gate every pull request” gains how the CRM’s checks are split between a pull request and the full run. An ordinary CRM pull request checks one size per layout band (320 reflow, tablet 768, laptop 1024, desktop 1280), and on a Windows machine Chromium only; every size and both engines are still checked at each wave pull request, whose merge is the one that ships, at each pull request into
release/*, and every night onmainonce the nightly full run is running (until its first run, the wave pull request’s gate is the only such run). The capture app keeps every size and both engines on every pull request. The supported sizes and browsers do not change. - Why: the lead’s answers to Q-3 and Q-11 of the test-run cost design (
.forge/design/harness/test-run-cost.md, P-1 and P-9, approved 2026-10-27): a full gate took about 50 minutes on a Mac and 2.6 to 3 hours on Windows, and Playwright’s WebKit on Windows is an engine no user runs, where most of the open Windows flaky-test bugs are. - Cascade:
.claude/CLAUDE.md § Quality Gates › Browser QA,.forge/test-strategy.md § Browsers & Viewports, the F40 spec (Rev entry for FR-7, FR-8, FR-10, FR-11 and NFR-1), and the browser gate itself (forge-et-fe’s PR profile). - Resolves: — (no OQ resolved).
Rev 81 — 2026-11-02
- Changed: §Non-Functional Requirements › Accessibility, Supported sizes: the accessibility and responsive browser checks no longer gate every pull request. They run at every size and in both engines every night on
main, in the nightly full run on the lead’s Mac, and a failure found there is fixed before new work starts. A pull request intomain, a wave pull request included, runs none of them; a pull request intorelease/*, which the nightly run does not test, still runs them in full before it is marked ready. Rev 80’s per-pull-request split (one size per layout band, Chromium only on Windows) no longer applies, because no pull request intomainruns the checks. The supported sizes and browsers do not change. - Why: the lead’s decision of 2026-11-02 (
.forge/design/harness/test-run-cost.mdS-4, revision line of 2026-11-02): a browser gate took about an hour on a Mac and two to three hours on Windows for each pull request, and the nightly full run, running since 2026-10-27, checks everything onmainevery night, so a defect is found by the next morning and fixed then, with the others found that night. - Cascade:
.claude/CLAUDE.md § Quality Gates › Browser QAand § Common Commands,.claude/rules/git-conventions.md § Pull Requests,.claude/rules/review-record.md,.forge/test-strategy.md § Browsers & Viewportsand § When Tests Run,.forge/design/coding-standards.md§7, the F40 spec (a revision entry), the browser-gate hook (unregistered), each frontend’sFast lanelast step (only on a pull request intorelease/*),/forge-deliver,/forge-pr-open,/forge-pre-pr-review,/forge-review-pr,frontend-reviewer, each frontend’sCLAUDE.mdand pull-request template, and.forge/guides/test-runs-explained.md. - Resolves: — (no OQ resolved).
- Approved by: Tharinda, 2026-11-02.
Rev 82 — 2026-11-04
- Changed: §Non-Functional Requirements › Security and Compliance, AI data handling: Haiku 4.5’s background work — lead triage (F11-d) and lead research (F11-c) — is called through Haiku 4.5’s Global inference profile, which may process a request in any commercial AWS Region, where the environment’s task definition sets it (dev from F43’s wave 4; UAT and production from the cloud-gated register’s §D, production staying on the US-only route if its Global quota is 0). The calls a person waits for (inquiry extraction, email drafts, the account briefing) stay on Haiku 4.5’s US-only route. The paragraph no longer says Haiku 4.5 runs on a US-only route. OQ-15 is resolved, by the lead’s decision: processing outside the project’s Region and outside the US is accepted, the paragraph says so and gives the grounds, and a US-only route stays a configuration change if Empire ever asks. §Risks R-02 records the same.
- Why: the lead’s direction of 2026-11-02 (
.forge/specs/F43-ai-quota-backlog-and-busy-answer-spec.mdpoint 4, FR-R3): AWS keeps the account at 10 Haiku 4.5 requests a minute on each of its two routes, and F42’s dev proof showed background work filling the minute that people’s calls share; on its own route it has its own 10 a minute, and the Global route is billed at list, about 10% below the US-only route. OQ-15 is closed by the lead on 2026-11-04. Empire never answered Q18 (asked 30 September), and the AWS Bedrock documentation, read the same day, settles the security side. Every call is made from Empire’s own account, and data between Regions stays on the AWS network, encrypted in transit. Model providers have no access to prompts or answers, and neither AWS nor the providers train on them. All of this holds on every route, so the route changes only which Region processes a request. - Cascade: OQ-15’s row moved from
project-prd-signals.mdto this file’s## Resolved Open Questions(its 2026-08-19 row, annotated);.forge/client/open-questions.mdQ18 (answered and closed) and.forge/client/client-facing-prd.md(§ 3.3, § 5); a post-issue note on.forge/client/project-proposal.md, an addendum to the 30 September note (L-001);.forge/design/architecture.md§ Provider and model routing;.forge/design/ai/ai-provider-strategy.md§3, §4b, §4d, §5 row 9;.forge/design/environments-and-identity.md§4;.forge/cloud-gated-validation.md§D and §B F43; revision entries in the F11-d and F11-c specs. Empire Teas is not told separately: the updated Q18, the client-facing PRD and the proposal’s post-issue note count as telling them (decided 2026-11-04, F43 spec Rev 2). Also, for OQ-15’s resolution: - this PRD’s § Constraints › Deployment line on the AI-data-privacy question;
.forge/client/phase-1-acceptance-criteria.md§ E and.forge/client/dependencies.md(the AWS account paragraph);.forge/discovery/meeting-notes/2026-10-16-kavindu-progress-update.md§6, a dated note;.forge/owed.md: the row to ask Empire OQ-15 / Q18 is closed, and the 16 October corrections row is updated;.forge/cloud-gated-validation.md: §B F11-c item 7, F11-h item 5 and §D;.forge/test-strategy.md;.claude/CLAUDE.md§ Stack;tracker.yamlR-PRD-002’s note;.forge/design/architecture.md, andai-provider-strategy.md§4b and §5 row 9;- revision entries on the specs that relied on OQ-15 being open: F43 Rev 2, F11-d Rev 8, F11-c Rev 6, F11-h Rev 7, F13 Rev 3, F12 Rev 8 and F38-f Rev 6.
The older specs’ dated gap-check and W lines stay as records. The same PR records two rules in ai-provider-strategy.md §4d, “Which model to try first” and “Which route”, and marks model-choice-evaluation-plan.md as the agreed approach; neither changes this PRD’s body.
- Resolves: OQ-15 (the location question reopened at Rev 48), moved to
## Resolved Open Questionsabove. - Approved by: Tharinda (lead), 2026-11-02, with the F43 spec (point 4, FR-R3, FR-H1); OQ-15’s resolution, Tharinda (lead), 2026-11-04.
Rev 83 — 2026-11-05
- Changed: §Functional Surface › F11, its summary line in §Scope and the web-search row in §Integrations: Rev 30’s “every lead is researched before it is ranked” gains a fallback. When the research cannot run — the search service is down, paused or switched off, or the lead has had no research for 24 hours since it entered the CRM or was last edited — the lead is ranked from the inquiry alone, at lower confidence, and ranked again once its research completes. Research still runs on every lead, and the final priority rests on it; no lead is left unranked because the search service is unavailable.
- Why: F11-g’s spec (
.forge/specs/F11-g-ai-lead-ranking-spec.md, decision 3, chosen by Dinil on 2026-11-05). Lead research runs on Exa’s free credit with no second search provider until Empire answers on Exa (SP-ARCH-004), so a search outage, a paused provider or a missing key would otherwise leave every new lead, and the migrated backlog that Phase 1 criterion E8 asks to see ranked, without a priority. Ranking from the inquiry alone is already the PRD’s rule for a lead whose research finds nothing, so the fallback reuses it. A lead ranked by the fallback costs a second ranking call (F11-g NFR-6). - Cascade:
.forge/active-context.md(the live-decisions sentence);.forge/design/ai/ai-provider-strategy.md§2 (the lead research + scoring row);.forge/client/client-facing-prd.md§3.3 and §4, with a revision line;.forge/client/phase-1-acceptance-criteria.mdE2 (a draft, not yet issued);.forge/owed.md(Tharinda’s sign-off). The issued proposal’s 16 September post-issue note (“lead triage researches each lead before ranking it”) is unaffected: research still runs on every lead and the final priority rests on it. The F11-c spec’s Context quotes Rev 30 as it stood when F11-c was written and is left as a dated record. - Resolves: — (no OQ resolved).
- Approved by: Tharinda (lead), 2026-11-05 (sign-off reported by Dinil in the F11-g session), before the F11-g spec was approved (F11-g Q-2).
Rev 84 — 2026-11-06
- Changed: the Pipedrive comparison’s P-30 row names F32-d beside F32-a; §Functional Surface › F23, F32, F33 and F34 each gain one sentence: the capability is built first on accounts, contacts and leads, in a feature started early in Phase 1 (F23-d, F32-d, F33-a, F34-c), and the Phase 2 records (deals, quotes, samples, buying standards, product briefs) join it in Phase 2 (F23-a, F32-a, F33-b, F34-a). One addition: §F33 listed deals, accounts, contacts, quotes, samples and product briefs, and now names leads too, because the split’s Phase 1 part covers the account, contact and lead pages (the lead agreed that split on 2026-11-06); nothing else in a requirement changes.
- Why: Gate 3 Run 4 (reslice,
.forge/engagement-gate-runs.md), the lead’s decision of 2026-11-06. Phase 2 does not fit its date on its own (R-ARCH-001), so the Phase 2 work that needs neither deals nor the capture app is started in Phase 1’s time: F20-g, F22 and F37-a move to Phase 1 as they are, and F32-a, F23-a, F33 and F34-a are cut by record type, the account, contact and lead parts becoming F32-d, F23-d, F33-a and F34-c (cut pattern: cluster-cut, by record type — signals: S2, the deal half of each waits on F2-c and the other half on nothing open; the developer’s verdict: accepted). F33’s own row is removed, as F38’s was on 2026-10-11, and its stub became F33-a’s. Phase 1’s own work, fixes and release come first; each of the seven moves back to Phase 2 before Phase 1 seals unless done (as F38-i). What Empire was told is unchanged: these remain Phase 2 commitments (the proposal’s 17 September post-issue note and PRD Rev 57 for bulk email), and none enters the Phase 1 acceptance criteria unless it is done and the lead adds it. §In Scope, the Gate-1 snapshot, is not edited;delivery_phaseintracker.yamlis the live placement. - Cascade:
tracker.yaml(the seven features, F33-b, the narrowed F32-a, F23-a, F34-a, F37-b’s blocker,setup.decomposition.reslice_runsandlast_gate_run, R-ARCH-001’s note);features.md(Delivery Plan counts, graph, lanes, rows); the spec stubs (F32-d, F23-d, F33-b and F34-c created; F33’s renamed to F33-a; F32-a, F23-a and F34-a narrowed; F20-g, F22 and F37-a moved to Phase 1; F37-b, F32-b, F34-b, F10 and F23-b repointed); revision entries in the approved specs that named the old owners (F1-a Rev 13, F1-d Rev 11, F1-e Rev 12, F1-f Rev 9, F11-b Rev 9, F11-f Rev 8, F20-b Rev 6, F20-c Rev 10, F20-h Rev 3, F26-d Rev 3, F38-c Rev 4, F38-j Rev 2, FS-6 Rev 20);.forge/active-context.md(phase scope);.forge/owed.md;.forge/checklists/phase-boundary-checklist.md§3 (a check that moves an early-started feature on at the seal);.forge/design/prototypes/README.md(the Phase 2 labels on F37-a and F20-g);.forge/cloud-gated-validation.md§B rows and its log;.forge/design/data/entity-map.md;.forge/design/environments-and-identity.md§2.4 and the Empire IT agenda (F22 needs onlyContacts.ReadWrite, which is granted). - Resolves: — (structural, no OQ resolved).
- Approved by: Tharinda (lead), 2026-11-06.
Rev 85 — 2026-11-07
- Changed: §Domain Model, the File Attachment row: its N:1 list of records adds Lead, as §F33 has said since Rev 84. The §In-scope features F33 line adds leads too.
- Why: F33-a spec FR-H2 (approved 2026-11-06): Rev 84 added leads to §F33, and F33-a attaches files to leads, but the domain row was not updated with it.
- Cascade: entity-map D-6 and the File row list the owner types (F33-a’s harness PR).
client/client-facing-prd.mdnames no list of records for attachments, so it is unchanged. - Resolves: — (no OQ resolved).
- Approved by: Dinil Gamage, 2026-11-06, with the F33-a spec; it carries out Rev 84’s addition of leads (Tharinda, the lead, 2026-11-06).
Rev 86 — 2026-11-07
- Changed: five lines, so that follow-ups read as assigned, not owned:
- §Lifecycle States › Activity reads
open (overdue | today | upcoming) → done,done → open(reopen), in place ofoverdue | today | upcoming → done; - §Domain Model › Key Entities, the Activity (Task) row: “N:1 owner” becomes “N:1 assignee (an active user who holds the record’s
managekey and reaches it; the assignee gives no reach of its own)”; - the User row: “1:N owns Leads, Accounts, Deals, Quotes, Samples, Activities” becomes “1:N owns Leads, Accounts, Deals, Quotes, Samples; 1:N is assigned Activities”;
- the Sales Rep role row, both columns: the second column’s “Owns leads, accounts, deals, quotes, activities, and their email for their book of business” becomes “Owns leads, accounts, deals, quotes, and their email for their book of business; is assigned activities”, and the third column’s “Create/manage own leads/accounts/deals/quotes/activities” becomes “Create/manage own leads/accounts/deals/quotes, and the activities on the records they own or that are shared with them” (a Sales Rep with no reports reaches the records they own or that are shared with them, and may manage a follow-up on any of them — spec FR-P1);
- §F3 — Activities: each row’s “type · account · owner” becomes “type · record (account, deal or lead) · assignee” (spec FR-V2).
- §Lifecycle States › Activity reads
- Why: F3-a spec decision 8 and FR-H1 (
.forge/specs/F3-a-follow-up-tasks-spec.md): a follow-up has an assignee, not an owner in the AD #35 sense. It gives the assignee no reach of its own, a follow-up is seen through its record’s reach, and a follow-up left with a deactivated assignee is shown to the record’s owner to reassign. F3-a shipped the reopen (done → open) in wave 2. The Sales Rep row is reworded under.claude/rules/prd.mdstep 4, as the last other statement that activities are owned (F3-a WI-4.1 plan## Notesitem 3). - Cascade: none in this file beyond the five lines (the §F3 row line was found by a fresh-context review, since the
ownssearch does not match “owner”; the capability matrix’s Activity row and the F3-a spec’s FR-A3, which quotes the old lifecycle, are left as they are and recorded in.forge/owed.md).git grep -n -i -E "(^|[^[:alpha:]])owns[^;|]*activit" -- .forge/project-prd.md .forge/client/prints only.forge/client/client-facing-prd.md’s Sales Representative line, which is issued to Empire and is left for the lead’s decision (.forge/owed.md, “Decide whether the client-facing PRD’s Sales Rep line changes”). The entity map’s Activity rows and the other specs’ records were updated by F3-a’s spec PR (FR-H1); F1-c’s spec records theactivitytimeline rows in its Rev 4. - Resolves: — (no OQ resolved).
- Approved by: Nanndish Satgunarajah, 2026-11-08, at F3-a’s close (Reflect)
Rev 87 — 2026-11-08
- Changed: §Users and Access › Per-Role Capability Matrix, the Saved view / list view row: every role may create, change, delete and share its own views and read the views others share; only the owner changes or deletes a view; an Administrator may also delete a shared view whose owner is deactivated. The row had made sharing a Sales Manager and Administrator capability (“Cfg shared”) and given R&D read only.
- Why: the F32-d spec (2026-11-08). Saved views, personal and shared, are promised in the issued proposal (“list views, saved filters, and bulk actions”) and the client-facing PRD (“personal or shared saved filters”), and Kavindu asked on 2026-10-09 for saved contact lists the team sends to. The managers-only sharing rule was the delivery team’s own proposal in
client/roles-and-permissions.md, never confirmed by Empire, and a key to express it would add a permission Empire must configure for little gain. Sharing follows the list’sviewkey, so no new key ships (FS-6 Appendix A, decided by F32-d FR-V5). - Cascade:
client/roles-and-permissions.md(the List views & saved filters row and the Sales Manager’s Can list, still a draft for Empire),design/data/roles-and-permissions-key-map.md(saved views need no key), FS-6 Appendix A (the Saved views row: decided),.forge/design/data/entity-map.md(thesaved_viewrow settled, alist_preferencerow added).client/client-facing-prd.mdsays “personal or shared saved filters” without naming who shares, so it is unchanged. - Resolves: — (no OQ; the F32-d spec’s Q-2).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-08.
Rev 88 — 2026-11-08
- Changed: §Functional Surface > F34 — “text, number, date, single/multi-select, and similar types” is pinned to five types (text, number, date, single select, multi select), and a field is archived and restored, never deleted. §Users and Access capability matrix, the Custom field definition row — the administrator’s “D” becomes “archive and restore (never delete)”.
- Why: F34-c spec (approved 2026-11-08), Q-1 (a) and FR-1, FR-6: a closed type set bounds the build, a further type stays an additive change, and archiving keeps the values records already hold; deleting a field would lose them.
- Cascade: none in the client pack —
client/client-facing-prd.mdand the proposal name no field types and no delete. The key table’ssettings.managerow still mentions custom fields “in Phase 2”; the AD #41 rewrite of that table is the.forge/owed.mdrow “Write the spec revisions for the per-page Settings keys” (Nanndish), and FS-6 Appendix A Rev 22 records the new keysettings.custom_fields.manage. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha, 2026-11-08, with the F34-c spec.
Rev 89 — 2026-11-09
- Changed: §Lifecycle States › Email: the tracking states are
sent → opened → clicked, for an email the rep chose to track, with nodeliveredstate for a single email; a scheduled email’s statesscheduled → sent | cancelled | failedare added. §Functional Surface › F20’s sentence on tracking and scheduled send now says tracking is chosen per email (on by default), that opens and clicks are estimates, that only the fact, time and link of an open or click are kept (no IP address or device), and that an email can be scheduled up to 30 days ahead and cancelled from a Scheduled folder. - Why: the F20-g spec (Dinil, 2026-11-09, decisions 1–4 and 10). Microsoft Graph reports no delivery for a sent message, and a delivery receipt is honoured only inside one Exchange organisation, so a
deliveredstate for a single email could not be observed; reading bounce reports belongs to F37-b, which needs them for its suppression list (OQ-62). The issued proposal lists email “tracking” (§4) without naming states, and the client-facing PRD promises delivered only in F37-b’s bulk summary, so neither changes. Keeping the minimum about each open and click follows the PRD’s data-protection note for buyer contacts. - Cascade: the F20-g spec’s FR-D1:
.forge/design/data/entity-map.md(email_tracking,email_tracked_link,scheduled_email;email_send.tracked),design/architecture.md§ Scheduling, revision entries on the F20-c, F20-h, F1-c, F20-i and F20-a specs, and lines on the F37-b, F25-b and F20-f stubs. F37’s own line in §Functional Surface keeps delivered in F37-b’s summary, which F37-b’s bounce detection makes observable. - Resolves: — (no OQ; the F20-g spec’s gap-check W-1 and W-2).
- Approved by: Dinil Gamage, 2026-11-09, with the F20-g spec.
Rev 90 — 2026-11-11
- Changed: §Functional Surface > F34 — custom fields on accounts, contacts and leads (F34-c) are Phase 1 scope, accepted with Phase 1; F34-a (deals, quotes, buying standards, samples) and F34-b (list columns and filters) stay in Phase 2.
- Why: the lead’s decision of 2026-11-11. F34-c, started early in Phase 1 under Rev 84, closed on 2026-11-11 with both waves merged, and the lead chose to have it accepted with Phase 1 rather than moved back to Phase 2 (
.forge/owed.md, the row on the Phase 2 features started early). - Cascade:
client/phase-1-acceptance-criteria.md(Section K, a dated header line, the “Not part of Phase 1” list);client/client-facing-prd.md(a dated header line, the Section 3 summary and the Section 3.4 custom-fields line);client/project-timeline.md(a dated header line, the Phase 1 and Phase 2 rows of Sections 1 and 2);.forge/active-context.md;.forge/owed.md.tracker.yamlalready holds F34-c atdelivery_phase: 1. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-11.
Rev 91 — 2026-11-12
- Changed: §Functional Surface > F36 and §Delivery phasing (the capture stream row and the Gate 3 re-grouping note) — the capture app, F36-a to F36-d, is delivered in Phase 1, scope unchanged; Gate 3 had placed it in Phase 2. The capture-app checkpoint, which the Statement of Work holds during Phase 2, is proposed for Fri 2026-11-20, Phase 1’s build target, to be confirmed with Empire (the lead, 2026-11-12).
- Why: at the progress meeting of 10 November 2026, Empire asked for the capture app to be usable for its January 2027 trade fairs (Sahampathy Dissanayake) and in Phase 1 (Kavindu Anthony), and Rukshan Salgado, Thilina Abeyrathna and Malinda Wellalage (Altrium) asked for it in Phase 1 too. The delivery team and Empire agreed it there; the lead recorded it in the tracker on 2026-11-11 (
features.md § Delivery Plan). - Cascade:
.claude/CLAUDE.mdAD #20 (with its archive entry in.forge/engagement-log.md);features.mdandtracker.yaml(the phase titles and themes);client/client-facing-prd.md,client/project-timeline.mdandclient/phase-1-acceptance-criteria.md(a dated header line each, and Section L of the acceptance criteria); the proposal (a post-issue note);.claude/rules/client-docs.md;.forge/team-guide.md;.forge/guides/tech-stack-reference.md;.forge/design/ai/lead-research-source-options.md; the meeting note of 10 November;.forge/owed.md.architecture.md § Trade-Show Capture App. Left for their owners (.forge/owed.md, the row “Carry the capture app’s move to Phase 1 into every document”): the UAT and production work that Section L needs for Phase 1’s round (who builds it; F38-h decision 6, F38-i FR-18,architecture.md § CI/CD Pipelines, the release rules), the F36 spec bodies, and Empire’s confirmation of the checkpoint’s date. The lead decided on 2026-11-12 that F38-h builds the capture app’s UAT hosting for Phase 1’s round, with F38-i doing the same for production. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-11 (the move); recorded 2026-11-12.
Rev 92 — 2026-11-12
- Changed: §Functional Surface > F20 — F44 is added: an email from a customer or a lead that F20-f has linked to its record, with no reply after one working day (an administrator can change the wait on the Settings page), creates a follow-up in Activities for the person who received it, with Reply (compose, from a template or an AI draft) and No Reply Needed. Phase 1, after F20-f. §F33 — F33-a (attachments on accounts, contacts and leads) is done and accepted with Phase 1; F33-b stays in Phase 2.
- Why: at the progress meeting of 10 November 2026 Roshan asked whether the CRM can identify emails nobody has replied to, and Kavindu Anthony said it was one of Empire’s main reasons for the CRM and believed it was built; nothing in the proposal, the PRD or
features.mddid it (checked 2026-11-12). The lead chose each part of the design on 2026-11-12 (specs/F44-unreplied-email-follow-up-spec.md§ Context). F33-a was finished, started early in Phase 1 under Rev 84, and shown in that demo; the lead chose to have it accepted with Phase 1, as Rev 90 did for F34-c. Two answers from the same meeting were settled the same day without a PRD change: nothing AI-drafted is ever sent without a person approving it (as the PRD already says; confirmed by the lead with Rukshan Salgado, Thilina Abeyrathna and Malinda Wellalage), and the ERP link stays out of scope until Empire names the ERP that replaces IFS (project-prd-signals.mdOQ-2). - Also changed (OQ-2): §Integration Points (the ERP row), the domain section’s ERP line and R-01 — Empire is replacing IFS with an ERP not yet chosen, so the automated link waits until Empire names it and gives its details; V1 keeps the handoff document.
- Cascade:
tracker.yaml(F44, Phase 1; F33-a’s phase comment);features.md(F44’s row, the graph and the Phase 1 count; F33-a’s row, the Delivery Plan list and the started-early lane);client/open-questions.mdQ8 and Q17 (IFS restated); a post-issue note on the proposal (IFS being replaced);specs/F44-unreplied-email-follow-up-spec.md(a stub);client/phase-1-acceptance-criteria.md(F7, Section M, a dated header line, the “Not part of Phase 1” list);client/client-facing-prd.mdandclient/project-timeline.md(a dated header line each, and the Phase 1 lines);project-prd-signals.mdOQ-2;.forge/active-context.md;.forge/owed.md. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-12.
Rev 93 — 2026-11-12
- Changed: §Non-Functional Requirements › Accessibility, Supported sizes: a frontend pull request into
mainstill runs no full browser gate, but now runs the browser tests its change selects (pnpm e2e:gate --selected) before it is marked ready, and again whenmaingains a merge in the same feature area before it merges. The nightly full run still runs every test at every size. - Why: the lead’s decision of 2026-11-12 on
.forge/design/harness/fewer-nightly-reds.md(Draft 2, recommendations 1–3 as R-1 to R-3): most red nights a pre-merge step could have caught since Rev 81 came from a change merged without running the browser tests it affects (forge-et-fe#442) or from two pull requests breaking each other (#470 and #472); a selected run costs minutes. - Cascade:
.claude/CLAUDE.md § Quality Gates › Browser QA;.claude/rules/git-conventions.md§ Branching and § Pull Requests;.claude/rules/test-record.md;.claude/agents/frontend-reviewer.md(dimension 32);.claude/commands/forge-review-pr.md(step 5, step 9);.claude/commands/forge-deliver.md,.claude/lib/forge-deliver/agent-prompt-template.md,.claude/skills/forge-pr-open/SKILL.md,.claude/skills/forge-pre-pr-review/SKILL.md;.forge/test-strategy.md;.forge/checklists/quality-checklist.md;.forge/guides/test-runs-explained.mdand the guides index;.forge/design/harness/test-run-cost.md(a pointer from S-4);.claude/rules/review-record.md;.claude/rules/release-and-branch-protection.md,.forge/team-guide.md,.forge/design/coding-standards.mdand.claude/lib/nightly/run-nightly.py(no full browser gate); each frontend repo’sCLAUDE.mdand PR template, in forge-et-fe#514 and forge-et-pwa#121. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-12.
Rev 94 — 2026-11-12
- Changed: §Functional Surface › F36 and its full description: the capture app’s lead notifications are in-app only (a notifications bell and an on-screen notice); it sends no email, where both said a notification email via SES (email-only). Scope stays Option C, with notifications in the app.
- Why: AD #42 (refines AD #20): the in-app notifications drawn in the capture-app prototype on 2026-11-12 tell a rep what an email would, and dropping the email removes F36-c’s sender, recipients and template. Proposed by Nanndish with the capture-app redesign; accepted: a rep with the app closed learns nothing until they open it.
- Cascade:
.claude/CLAUDE.md(§ Stack, the Infra and capture-app lines; AD #20’s row; the new AD #42 row);.forge/design/architecture-decisions-why.md(row 42);.forge/engagement-log.md(§ AD #42 in the rationale archive);.forge/design/architecture.mdRev 87 (diagram, components, § Trade-Show Capture App, AD-O-3);.forge/client/project-proposal.md(post-issue note of 12 November, §4 and §A.7);.forge/client/client-facing-prd.md(header, Section 3.5, Section 8 and its cost paragraph);.forge/client/project-timeline.md;.forge/client/phase-1-acceptance-criteria.md(L5 and its release note);.forge/client/dependencies.md(the SES row);.forge/client/roles-and-permissions.md(the capture app’s table);.forge/client/sow-cost-annexe.md(the lead-notification line);.forge/specs/F36-c-capture-research-notify-spec.md(draft stub: title, context, scope, the withdrawn SES-sender question);.forge/features.mdand.forge/tracker.yaml(F36-c’s title and notes);.forge/design/prototypes/feature-detail.mdandREADME.md;.forge/cloud-gated-validation.md(§B FS-5 SES row, the F36-c backlog row,## Log);.forge/guides/tech-stack-reference.md;.forge/active-context.md(Open risks). Left for the lead in.forge/owed.md(“The backend’s SES sender has no consumer after AD #42”): the approved F38, F38-e, F38-h and F38-i specs, which still name F36-c as the sender’s first consumer and a capture notification to prove. - Resolves: — (no OQ resolved;
.forge/owed.md’s capture-app decisions row, item 7). - Approved by: Tharinda Rajapaksha (the lead), 2026-11-12 (confirmation reported by Nanndish).
Rev 95 — 2026-11-12
- Changed: §Functional Surface › F36, its full description, the capture stream’s row in the phase table, and the V1 permission-key table (a
capture.userow): a captured lead reaches the CRM only after a person reviews it in the capture app, where the PRD said captured leads land in the F11 triage intake. Converting creates the CRM lead through the F11 intake with lead-source tracking, owned by the person who converted it, with the capturing rep kept as captured by (AD #35’s one owner); rejecting it, with an optional reason from a set list and an optional note, keeps it in the capture app only, where it can be moved back to review. Cards and badges are captured one at a time (the POC’s Scan a pile and choosing several photos are not built) and sent with Submit for Research. Using the app needs its own permission,capture.use(Use the capture app), a special key likeprice.publishwith full use of the app and nothing finer, wherelead.managewas the default. - Why: the capture-app decisions of 2026-11-12 (
.forge/owed.md, the row “Fold the capture-app decisions of 2026-11-12”, items 1 to 4), made by the delivery team (Nanndish, with the UI/UX designer) and accepted by the lead with the screens drawn in.forge/design/prototypes/pwa-current/(prototypes/README.md § The capture-app screens and flows agreed on 2026-11-12).- Cascade:.forge/specs/foundation/006-auth-scaffolding-spec.mdRev 23 (Appendix A.1, A.3, A.4);.forge/design/architecture.mdRev 88 (§ Components, § Trade-Show Capture App);.forge/client/project-proposal.md(post-issue note of 12 November);.forge/client/client-facing-prd.md(header, Section 3.5);.forge/client/phase-1-acceptance-criteria.md(header, L1–L3);.forge/client/roles-and-permissions.md(header, Rule 3’s named permissions, the unowned-records answer, §4.7, A-41);.forge/design/data/roles-and-permissions-key-map.md;.forge/design/prototypes/feature-detail.md(F36);.forge/owed.md(the capture-app decisions row). Left for the specs, in that row: F36-a states where a held capture is stored (with the entity map’s capture paragraph) and what converting needs besidecapture.use; F36-c states how research runs before conversion; F36-b states one capture at a time for card reading. The approved F11-a spec’s rationale quotes the PRD’s earlier wording and is left as it was written; its intake contract is unchanged, since converting goes through it. Not settled here and still open on OQ-55: what the badge QR encodes, and whether exhibitor planning is in V1. - Resolves: — (no OQ resolved).
- Approved by: Nanndish (F36’s owner), 2026-11-12, folding decisions Tharinda accepted with the screens that day.
Rev 96 — 2026-11-12
- Changed: §Functional Surface › F36 and its full description: planning before a show is in Phase 1 — the capture app’s Plan tab (a campaign per show with its first and last day; the exhibitor list loaded from a CSV or Excel file read without AI, or exhibitors added by hand, edited and archived; visit notes and visited marks), with the AI researching and ranking every exhibitor. The screens and data are built in F36-a and the research and ranking in F36-c, as later waves of each; no new feature. The reduced capture-and-feed shape now keeps the review in the app. Converting a captured lead also needs
lead.manage, besidecapture.use(the F36 description and the V1 permission-key table). OQ-55’s working position, which kept the POC’s pre-show exhibitor targeting out of V1, is reversed for that one extra; pile-splitting and camera-roll import stay out. - Why: the lead’s decision of 2026-11-12, reported by Nanndish, after Empire saw the plan screens at the progress meeting of 10 November; it answers item 8 of the
.forge/owed.mdrow “Fold the capture-app decisions of 2026-11-12”. Kept inside F36-a and F36-c rather than a feature of its own, and split between them (Nanndish, the same day): the screens and data sit with F36-a’s capture-app shell, and the research with F36-c’s research of captured leads, the reason given when the split was chosen. Also recorded the same day, outside the PRD body: the states drawn after the review are accepted as drawn (the lead, reported by Nanndish); the converting rule above is Nanndish’s; AD #37 is left as it is; and F36-a’s spec is drafted reading the common QR formats until the badge photo arrives (Nanndish’s waiver, F36-a’s stub). - Cascade:
.forge/project-prd-signals.md(OQ-55, still partial: the badge QR encoding stays open);.forge/specs/foundation/006-auth-scaffolding-spec.mdRev 24 (Appendix A.1’scapture.userow);.forge/design/architecture.mdRev 89 (§ Trade-Show Capture App);.forge/client/project-proposal.md(post-issue note of 12 November);.forge/client/client-facing-prd.md(header, §3.5);.forge/client/phase-1-acceptance-criteria.md(header, L8, L9 and the reduced-release paragraph);.forge/client/project-timeline.md(header, “The capture application”);.forge/client/roles-and-permissions.md(header, Rule 3’s named permissions, §4.7, A-41);.forge/client/open-questions.md(header, Q9 and its summary row);.forge/specs/F36-a-capture-and-upload-spec.mdand.forge/specs/F36-c-capture-research-notify-spec.md(draft stubs: scope, the QR waiver, the converting key, research before conversion);.forge/tracker.yaml(F36-a and F36-c notes);.forge/design/prototypes/feature-detail.md(F36);.forge/design/ai/lead-research-source-options.md(the exhibitor-list import note);.forge/owed.md(the capture-app decisions row; the row asking Tharinda to accept the new states closes, replaced by one asking him whether the addition changes the price, and to cost the exhibitor research inclient/sow-cost-annexe.md). - Resolves: — (OQ-55 stays partial: the badge QR encoding).
- Approved by: Nanndish (F36’s owner), 2026-11-12, folding the lead’s decision as Nanndish reported it.
Rev 97 — 2026-11-12
- Changed: §Domain Model › Key Entities gains Captured lead, Capture campaign and Exhibitor (the capture app’s entities, which Revs 95 and 96 describe in F36’s prose), and the dropped marketing Campaign row says it is not the capture campaign; §Lifecycle States gains the captured lead’s states (
to_review → converted | rejected,rejected → to_review). - Why: the F36-a spec’s review (Pass 1, B2, 2026-11-12): the spec adds three tables and a state set that the PRD’s entity table did not name; F11-a’s FR-D2 set the precedent of revising Key Entities in the spec’s PR.
- Cascade:
.forge/specs/F36-a-capture-and-upload-spec.md(FR-D1);.forge/project-prd-signals.md(OQ-55: F36-a leaves its Blocks column under Nanndish’s drafting waiver of 2026-11-12, as F11-c and F11-f left theirs). The entity map follows in F36-a’s harness PR (FR-D1). - Resolves: — (no OQ resolved).
- Approved by: Nanndish (F36’s owner), 2026-11-12, under his overnight delegation (drafted by Claude; for his review).
Rev 98 — 2026-11-13
- Changed: §Functional Surface › F36’s full description: the capture app shows its ranking as a tier, Hot (P1), Warm (P2) or Cold (P3), with a fit score from 0 to 100; the tier is a fit ranking, separate from the CRM’s hot / cold engagement status (Rev 55: how recently the buyer wrote), and is not carried into the CRM, where a converted lead shows the CRM’s own engagement status and its own P1–P3 priority.
- Why: the F36-c spec’s review (Pass 1, finding I4, 2026-11-13): the capture app reuses the words Hot and Cold, which the CRM uses for engagement, so a lead ranked Cold in the app arrives in the CRM as hot (it was just received). The tier names stay, because Empire saw them in the POC and the accepted screens (
.forge/design/prototypes/pwa-current/) draw them; the PRD now says what each word measures. Decided by the main session under Nanndish’s overnight delegation of 2026-11-12, for his review (F36-c decision D-20). - Cascade:
.forge/specs/F36-c-capture-research-notify-spec.md(FR-CV1, FR-LD5, decision D-20; its W-7 closed). No other document states the capture app’s tiers (git grepof the tier words across.forge/client, the PRD,architecture.md,feature-detail.mdanddesign-system.mdfound only the Pipedrive-label note indesign-system.md, which is about migrated leads and is unchanged). - Resolves: — (no OQ resolved).
- Approved by: Nanndish (F36’s owner), 2026-11-13, under his overnight delegation (drafted by Claude; for his review).
Rev 99 — 2026-11-17
- Changed: §Functional Surface › F11’s research paragraph and §Integration Points’ web-search row: a second search, for the person named on a lead (their name with their company), is made while the person-search switch on the CRM’s Settings page is on; an administrator turns it on or off, and it is on by default. With it off, the person’s name is not sent and their role is taken from the sources about the company, as before. This replaces Rev 75’s “the person’s name is not sent to the search service; a search for the person is added only once Empire Teas agrees”. The web-search row’s provider note also drops “provisional until Empire Teas answers” (Exa was chosen on 10 November 2026; SP-ARCH-004 closed).
- Why: Empire Teas answered the F11-c person-lookup question at the progress meeting of 10 November 2026 (Kavindu Anthony;
.forge/discovery/meeting-notes/2026-11-10-kavindu-crm-progress-update-demo.md§8): a person search may go to Exa, behind a Settings switch that is on by default. F11-i builds the switch (.forge/specs/F11-i-person-search-switch-spec.md). Empire’s written agreement and Exa’s data processing agreement stay owed (.forge/owed.md); until they cover an environment’s Exa account, its administrator turns the switch off before Empire’s real leads enter it (F11-i decision 8). - Cascade:
.forge/specs/F11-c-lead-research-job-spec.md(Rev 9);.forge/specs/foundation/006-auth-scaffolding-spec.md(Rev 25);.forge/design/ai/lead-research-source-options.md(its revision);.forge/owed.md;.claude/CLAUDE.md(the lead-research Reference Map row);.forge/features.md(the F11-c row);.forge/plans/F38-h/uat-data-checklist.md;.forge/client/sow-cost-annexe.mdchecked (F11-i FR-H2). The issued proposal, the client-facing PRD and criterion E1 already say the person is looked up, so they are unchanged. The F36-c spec (being built in another session) is revised after its wave 2 merges (owed.md). - Resolves: — (no OQ resolved; F11-c’s Q-1 was closed in that spec).
- Approved by: Dinil (F11’s research owner), 2026-11-17, under his overnight delegation (drafted by Claude; for his review).
Rev 100 — 2026-11-18
- Changed: §Functional Surface — three Phase 2 features started early in Phase 1 are done and accepted with Phase 1: F20-g (open and link tracking, and scheduled send; done 2026-11-16) in §F20; F32-d (list views and saved filters on accounts, leads and contacts; done 2026-11-13) in §F32; F37-a (the weekly price broadcast; done 2026-11-18) in the F37 summary, the F20 summary and §F20. F32-a, F32-b and F37-b stay in Phase 2.
- Why: the lead’s decision of 2026-11-18: every Phase 2 feature started early and finished in Phase 1 goes into Phase 1’s acceptance criteria. Rev 90 (F34-c) and Rev 92 (F33-a) applied the same choice one feature at a time; this applies it to the three finished since, and to any still started early (F23-d) when it is done (
.forge/owed.md, the row on the Phase 2 features started early). - Cascade:
client/phase-1-acceptance-criteria.md(F8, F9, G9, Section N, a dated header line, the “Not part of Phase 1” list, the oldest header line moved to Change history);client/client-facing-prd.md(a dated header line, the Phase 1 list in Section 3, Sections 3.1, 3.2 and 3.4, Section 4’s bulk-email item);client/project-timeline.md(a dated header line, the Phase 1 and Phase 2 lists of Sections 1 and 2, the oldest header line moved to Change history);client/open-questions.md(Q10’s price side);client/user-guide-weekly-prices.md(the bulk send’s phase);features.md(the three rows, the Delivery Plan’s started-early row and the Phase 1 summary);checklists/phase-boundary-checklist.md(the started-early item);design/prototypes/README.md(F37-a’s note);.forge/active-context.md;.forge/owed.md.tracker.yamlalready holds F20-g, F32-d and F37-a atdelivery_phase: 1. The issued proposal gains no post-issue note: bulk email was added scope from Rev 57, and Rev 90 and Rev 92 added none. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-18.
Rev 101 — 2026-11-18
- Changed: §Functional Surface › F36’s summary line and full description, and F11’s research paragraph (a lead converted from the capture app arrives researched and ranked, and the CRM does not research or rank it again at conversion). (1) A converted captured lead keeps the capture app’s ranking: at conversion, Hot, Warm or Cold becomes the CRM lead’s P1, P2 or P3, with the ranking’s confidence, reason and next action, and the CRM does not rank the lead again at conversion; the 0–100 fit score stays in the app. This reverses Rev 98’s “is not carried into the CRM”; Rev 98’s point that the tier is a fit ranking, separate from the CRM’s hot / cold engagement status, stands. (2) A campaign can carry an optional focus: a short note the AI ranks the campaign’s exhibitors and captured leads for, on top of the standard rules, never replacing them, passed to the AI as data. Option A of the design discussion: the focus can change the tier and so the priority. A ranking made with a focus says so (the AI’s reason, a fixed line with the focus text and campaign name, a marker in lead lists with the text on hover, tap or keyboard focus, and a CRM lead-list filter); the focus text is kept with the ranking and used by a later ranking in the CRM; editing the focus ranks the campaign’s exhibitors and its leads still to be reviewed again (the ranking only, not the research), and leaves converted leads as they are; the marker is on lead lists, not the exhibitor list, where every exhibitor of a campaign with a focus would carry it. OQ-55 listed the POC’s campaign “lens” among what the POC built beyond F36; this takes the idea up.
- Why: the lead’s decision of 2026-11-18, reported by Dinil (F36-c’s owner), after a design discussion: ranking a captured lead twice, once in the app and again in the CRM, could give the same lead two different priorities, and a show often has a purpose of its own (a buyer type or a market) that the standard rules cannot know. Option A was chosen over showing the focus as a separate “focus match” badge because, to the sales team, P1 means “chase this first”, which is what a show’s focus is for; the marker and the filter keep visible which priorities used a focus. Not raised with Empire Teas: the delivery team decides it inside F36’s agreed scope. Built now, in Phase 1: as a further wave of F36-c after its second, with the campaign form’s field in F36-a’s screens and F11-g’s ranking reading the kept focus; the F36-c and F11-g spec revisions and F36-c’s new wave are owed (
.forge/owed.md). - Cascade:
.forge/project-prd-signals.md(OQ-55’s note on the campaign lens);.forge/engagement-log.md(the decision);.forge/owed.md(the F36-c wave, the spec revisions, and telling Nanndish about the campaign form’s field). Owed, not in this change:.forge/specs/F36-c-capture-research-notify-spec.md(FR-CV1, FR-RK3, D-20 and its Scope Boundaries’ “lens” line), whose wave 2 is being built in another session onfeature/F36-c-wave-2, and.forge/specs/F11-g-ai-lead-ranking-spec.md(a converted captured lead’s ranking, the kept focus, the CRM list’s marker and filter).git grepof “not carried into the CRM”, “lens” and “campaign-aware” across.forgeand.claudefound no other current statement, and the fresh-context review added F11’s “an AI research pass precedes scoring on every lead”, now qualified for a converted captured lead; the rest are dated records (meeting notes, plans of merged work items, Rev 98), the POC’s own command (.claude/commands/et-run-pwa-poc.md) and the POC snapshots. The client-facing pack does not describe the tier’s path into the CRM or a campaign focus, and the issued proposal promises research and scoring without these details, so neither changes and no post-issue note is added; the running cost of ranking again after a focus edit is stated in the F36-c spec revision againstclient/sow-cost-annexe.md. - Resolves: — (no OQ resolved; OQ-55 stays partial: the badge QR encoding).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-18, as reported by Dinil.
Rev 102 — 2026-11-19
- Changed: §In Scope › the F37 summary line. (1) F37-b, bulk email to a saved contact list, is started early in Phase 1 on 2026-11-19; it moves back to Phase 2 if it is not done by the end of Empire’s Phase 1 UAT round. (2) OQ-62’s questions (consent, unsubscribe, suppression, sending address) were to be answered by Empire before F37-b was specified; they are now decided by the delivery team in F37-b’s spec, and Empire is asked to confirm the choices at the Phase 1 hand-over. Nothing else in F37’s requirements changes.
- Why: the lead’s decisions of 2026-11-19. F37-b’s three blockers (F37-a, F32-d, F20-g) are done, so it can start in Phase 1’s time, as F37-a did. The lead’s rule of 2026-11-13 (no questions to Empire before UAT,
.forge/active-context.md) covers OQ-62 once work reaches it. The lead also set when an unfinished started-early feature moves back: at the end of Empire’s Phase 1 UAT round, not before Phase 1 seals as Rev 84 said, so it can go out in UAT drop 1 or drop 2. This applies to F23-d and F37-b; F38-i keeps its own rule. What Empire was told is unchanged: F37-b and F23-d stay Phase 2 commitments in the client pack until each is done, when it joins the Phase 1 acceptance criteria (Rev 100’s rule). - Cascade:
tracker.yaml(F37-b’s move and claim, merged in forge-et-harness#1542; the Phase 1 and Phase 2 themes and Phase 1’s notes; thedelivery_phasecomments of F20-g, F32-d and F37-a, done, and of F23-d and F37-b, with the new rule);features.md(the Delivery Plan rows and paragraph, the started-early lane, the F23-d and F37-b rows);.forge/owed.md(the started-early row, the OQ-62 row, the hand-over list, the Phase 1 boundary-check row, and a new row for the lead to decide how Phase 1 seals and its UAT release opens around the two drops, which this change and the plan of 2026-11-18 leave open);.forge/active-context.md;.forge/checklists/phase-boundary-checklist.md§3;.forge/project-prd-signals.md(OQ-62’s note); the F37-b spec header and constraints; the F23-d spec (Rev 1);.forge/engagement-log.md;.forge/design/prototypes/README.md. Left as written: dated records (gate runs, audits, revision entries, the plans and specs of features already done). Not changed: the client pack and the proposal, because the started-early features are not promised in Phase 1. - Resolves: — (no OQ resolved; OQ-62 stays open until F37-b’s spec records the team’s choices).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-19.
Rev 103 — 2026-11-19
- Changed: §Scope and Boundaries › In Scope (V1) › F37 (the (b) part states the decided sender, audience, unsubscribe and bounce handling instead of Rev 102’s “which the delivery team decides in F37-b’s spec”); §Domain Model’s (Not built) Campaign row (F37-b keeps a list of addresses left out of bulk sends, with no per-contact consent state); §Non-Functional Requirements › Security and Compliance › Marketing consent (the opt-out model with an unsubscribe link that opens a page with one button, Empire’s postal address in every email, no opt-in tick, and opt-in for some contacts as a change made at UAT); §Risks › R-15’s mitigation (the spec decides OQ-62 and Empire confirms at the hand-over, replacing “F37-b’s spec is not approved until Empire answers OQ-62”); §Users and Access, the key table and the capability matrix’s Email row (
email.bulk_send, added at F37-b’s spec review). (1) Sender: each email of a bulk send goes from the mailbox of the person who confirms it, through Microsoft Graph, one email per contact, at most 20 a minute as F37-a does and at most 5,000 recipients in any 24 hours counted with the sender’s price broadcasts, so replies reach that person; a shared address through SES is not built. (2) Consent: every contact on the saved list receives it unless they have unsubscribed; there is no opt-in tick; every bulk email carries Empire’s postal address (a new plain setting, at the spec review) and an unsubscribe link that needs no sign-in and opens a page with one Unsubscribe button, because mail security scanners open every link. (3) Suppression and delivered: bounce reports in the sender’s mailbox are read; a bounced address shows not delivered, delivered is the number sent less the number bounced, and an unsubscribed or bounced address is left out of later bulk sends until someone who can edit the contact clears it at the contact’s request, audited. Complaints are not handled: Microsoft Graph gives a mailbox no complaint reports (F37-b spec, Out of Scope). (4) Audience: a saved contact list (F32-d), the contacts in the sender’s reach. (5) A new key:email.bulk_send(§ Users and Access key table and the capability matrix’s Email row), held by no role until an administrator grants it (FS-6 Rev 26). - Why: F37-b was started early in Phase 1 on 2026-11-19, and under the lead’s rule of 2026-11-13 (No questions to Empire before UAT,
active-context.md) a question waiting on Empire is decided by the delivery team when work reaches it: the simplest option for Empire’s staff, minimal, and flexible. The lead chose each option at F37-b’s gap check on 2026-11-19. The sender’s own mailbox needs nothing new in AWS or in Empire’s DNS and reuses F37-a’s paced send path and F20-g’s tracking; Exchange Online’s limits (30 messages a minute and 10,000 recipients a day per mailbox) allow a few hundred contacts in about a quarter of an hour. No opt-in tick, because Empire’s contacts are business contacts and a tick on each migrated contact would block the first newsletter. Reading bounces keeps the delivered figure Rev 57 promised, which Microsoft Graph does not report for a single email (F20-g decision 2). - Cascade:
.forge/project-prd-signals.md(OQ-62’s row removed; its resolved row added above);.forge/client/open-questions.md(Q10 states the choices for Empire to confirm, its summary row, a dated header line, the oldest header line moved to Change history);.forge/client/client-facing-prd.md(Section 3.4’s bulk-email line, a dated header line, the oldest header line moved to Change history);.forge/owed.md(the Q10 row now asks Empire to confirm; the hand-over meeting’s list);.forge/features.md(F37-b’s row);.forge/active-context.md(Phase scope);.forge/specs/F37-b-bulk-email-to-a-list-spec.md(the detail); FS-6 Rev 26 (email.bulk_sendin Appendix A.1); “changed by F37-b” revisions on F20-b (Rev 8, the bounce step in its ingest), F20-g (Rev 5), F20-c (Rev 17), F1-a (Rev 14) and F1-d (Rev 16);.forge/design/data/entity-map.md(Rev 48);.forge/discovery/inventories/feature-inventory.md(row 13);tracker.yaml(F37-b). A fresh-context reviewer compared the old and new wording on 2026-11-19 (.claude/CLAUDE.md§ Compounding Engineering, rule 6): every passage faithful or narrowed, widened or reversed as intended, none dropped; its two must-fixes (the F20-b home of the ingest change, F20-g Rev 5’s approval) and its wording fixes are applied.git grep OQ-62across.forgeand.claudefound these other current mentions, left as they are: the approved specs of F20-d, F20-g, F26-d and F37-a, which record that the questions were F37-b’s; the prototypes (design/prototypes/crm/is frozen, andcrm-current/with its README entry andfeature-detail.mdare redrawn by F37-b’s plan, as a UI change updates the current prototype in the same delivery);tracker.yaml’s dated R-ARCH-001 updates; and dated records (the 2026-10-09 meeting note, audits, the engagement log). The issued proposal gains no post-issue note: bulk email has been added scope since Rev 57, and its notes of 16 September and 9 October record the questions as they stood then, which this decision does not change. - Resolves: OQ-62.
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-19.
Rev 104 — 2026-11-19
- Changed: §Domain Model › Key Entities, the Email / Message row’s relationships: “N:1 Account/Contact/Deal/Lead” becomes many links per email, one per record — an account, contact or lead, and a deal from Phase 2. The File Attachment row is unchanged: a received email’s files are kept in F20-f’s own table, not as File Attachment rows.
- Why: the F20-f spec (approved 2026-11-19), D-1: one email often concerns several records — a buyer and a colleague of theirs on CC are two contacts, perhaps of two accounts, and a lead may be on the same thread — so each link is a row of its own (
email_link), not a column on the message. D-6 keeps received files out of FS-11’sattachmenttable. - Cascade:
.forge/design/data/entity-map.md(the Email message row; new Email link, Email link refusal and Email attachment rows; its Rev 50);.forge/design/architecture.md§ Mailbox integration (its Rev 93). No client-facing document states the email’s relationships, so none changes. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-19, with the F20-f spec.
Rev 105 — 2026-11-20
- Changed: phase wording only, no requirement. (1) §Phasing / Sequencing Intent, the Re-grouped at Gate 3 note: it adds that parity work needing no deals was built early and accepted with Phase 1 (email tracking and scheduled send, the weekly price broadcast, and list views, attachments and custom fields on accounts, contacts and leads — Revs 90, 92 and 100), and that other Phase 2 work started early joins Phase 1 only when done (Rev 102). (2) §Reference-System Classification, rows P-13 and P-24: F37 is committed to Phase 2 with its price broadcast (F37-a) accepted with Phase 1, instead of “F37 (Phase 2)”. (3) §Non-Functional Requirements › Security and Compliance › Marketing consent, §Risks › R-15, and §Domain Model’s (Not built) Campaign row: F37-b is a Phase 2 commitment started early in Phase 1 (Rev 102), instead of “Phase 2” alone. (4) The Gate 3 note also says that F38-i, the production environment, keeps its own rule.
- Why: the lead asked on 2026-11-20 for every statement of phase scope to match the features moved between the phases. Rev 102 changed F37’s In Scope line but left these four places saying Phase 2 alone, and the Gate 3 note did not mention the parity work accepted with Phase 1.
- Cascade:
tracker.yaml(both phases’ themes and notes);features.md(the Delivery Plan rows);.forge/active-context.md(a pointer to the started-early terms); the F20-f spec (Rev 3, F20-k’s phase);.forge/guides/email-in-the-crm-explained.md(its header);.forge/design/prototypes/feature-detail.md(F37’s phases);.claude/rules/tracker.md§ Delivery Phase Rules (a phase move updates the phase themes and every statement of the feature’s phase in the same change);.forge/engagement-log.md. Unchanged, by the rule of Rev 100 and Rev 102: the client pack, which keeps a started-early feature in Phase 2 until it is done. - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-20.
Rev 106 — 2026-11-20
- Changed: § Non-Functional › Security, AI data handling: the background work on Haiku 4.5’s Global route now names the summary of a migrated lead from its linked email (F11-j) beside lead triage, research and ranking.
- Why: the F11-j spec (approved 2026-11-20), D-9 and D-11: a migrated Pipedrive lead with no inquiry text gets its summary from the buyer’s linked received email, as a background job on the Global profile with the other background features (F43 FR-R3). The paragraph already sends buyer communication text, synced email included, to Bedrock, and § F11 already names an “AI summary of the inbound email”, so the scope is unchanged; only the list of what runs on the Global route grows.
- Cascade:
.forge/design/architecture.md§ Provider and model routing and § AI data-handling boundary (its Rev 94);.forge/design/ai/ai-provider-strategy.md§2 and §4b;.forge/client/phase-1-acceptance-criteria.mdE8 (a draft, not yet issued). The issued proposal needs no note: its post-issue note of 27 September 2026 already says each lead is “summarised as it arrives” by a background job (F11-j D-11). - Resolves: — (no OQ resolved).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-20, with the F11-j spec.
Rev 107 — 2026-11-20
- Changed: §Functional Surface › F22 and §In Scope (V1) › the F22 summary line: contact sync reads each user’s Outlook contacts one way into the CRM (nothing is written to Outlook), offers the ones the CRM does not hold on a per-user From Outlook review list where the user adds or dismisses them, and fills only empty fields on matched contacts; “one- or two-way per user configuration” and “syncs People … into the CRM” are gone. §Scope and Boundaries › Reference-System Classification: P-16 changes from Replicate to Redesign, with the reason. §Integration Points, the Microsoft 365 row: contact sync is one way. §Domain Model › Key Entities, the Contact row: “may be added from a user’s Outlook contacts”. The PRD’s Last updated header line names this revision. The old caveat that one person held by two users can become two records is kept in its true form: it no longer happens when the two copies share an email address.
- Why: the lead’s three decisions of 2026-11-20 in the F22 spec (
.forge/specs/F22-contact-sync-spec.md§ Context): one way, because CRM contacts are shared team records and an address book is one person’s list; a review list, because an address book holds personal contacts and a contact with no account is visible only toaccount.allholders (F1-f, AD #35, AD #37); blanks only, so a colleague’s value is never overwritten. Empire is not asked before UAT (the lead’s rule of 2026-11-13); write-back is a change Empire can ask for at UAT. The issued proposal says only “contact sync” (§4), so no post-issue note is needed. - Cascade:
.forge/discovery/inventories/feature-inventory.md(row 16);.forge/design/prototypes/feature-detail.md(F22);.forge/design/data/entity-map.md(the Contact row and the Contact sync link row; its Rev 54);.forge/design/architecture.md(the overview sentence; its Rev 95);.forge/features.md(the F22 row);client/client-facing-prd.md(the calendar and contact sync bullet split, a dated header line) — all in the F22 spec PR (spec FR-14). - Resolves: — (no OQ resolved; Q37 of the 2026-09-03 question set was never answered and stays a dated record).
- Approved by: Tharinda Rajapaksha (the lead), 2026-11-20, with the F22 spec.