chore: archive v1.5 milestone

Author: mshegolev

.planning/MILESTONES.md

@@ -2,25 +2,39 @@
 
 ## Current
 
+None.
+
+Next step: run `$gsd-new-milestone` when the next milestone goal is ready.
+
+## Completed
+
 ### v1.5 Live Review + Safe Approved Speak
 
 **Started:** 2026-05-21
-**Status:** Active
+**Completed:** 2026-05-21
+**Archived:** 2026-05-21
+**Status:** Complete
 
 Focus: make the live loop reliable after real call recordings by improving
 post-call review, handoff detection, speaker bootstrapping, approved draft
 playback, and regression replay.
 
-Planned:
+Delivered:
 - Prioritized post-call review queue for handoffs, questions, mentions, and
   classifier-critical labels.
-- Addressing v2 to separate direct handoffs from incidental name mentions.
-- Speaker bootstrap to reduce `unknown` coverage without required live
-  diarization.
-- Safe playback of explicitly approved generated drafts.
-- Saved-session regression replay and sanitized reliability reports.
+- Addressing v2 that separates direct handoffs from incidental name mentions
+  and explains why Saymo answers, waits, or skips.
+- Speaker bootstrap workflow to reduce `unknown` coverage without required
+  live diarization.
+- Safe playback of explicitly approved generated drafts through local
+  TTS/playback gates.
+- Saved-session regression replay with sanitized reliability reports.
 
-## Completed
+Archives:
+- `.planning/milestones/v1.5-ROADMAP.md`
+- `.planning/milestones/v1.5-REQUIREMENTS.md`
+- `.planning/milestones/v1.5-MILESTONE-AUDIT.md`
+- `.planning/milestones/v1.5-phases/`
 
 ### v1.4 Live Conversation Memory + Answer Cockpit
 

.planning/PROJECT.md

@@ -18,20 +18,30 @@ enough that the call still feels live.
 
 ## Current State
 
-Saymo has shipped milestone v1.4 Live Conversation Memory + Answer Cockpit. The
+Saymo has shipped milestone v1.5 Live Review + Safe Approved Speak. The
 product can build local transcript ledgers from captured sessions, search and
 ask questions about those sessions with transcript citations, generate pending
-answer drafts grounded in meeting memory and configured sources, and record
-explicit cockpit actions plus sanitized audit trails.
+answer drafts grounded in meeting memory and configured sources, record
+explicit cockpit actions plus sanitized audit trails, review real captured
+sessions, reduce incidental name-handoff false positives, bootstrap speaker
+labels, play only approved generated drafts, and replay saved sessions as local
+regression fixtures.
 
-**Status:** v1.5 Live Review + Safe Approved Speak started on 2026-05-21.
+**Status:** v1.5 Live Review + Safe Approved Speak archived on 2026-05-21.
+No active milestone is defined; the next step is `$gsd-new-milestone`.
 
-The 2026-05-21 recorded session exposed the next reliability gaps: all 507
-windows are still `speaker=unknown`, both `asked_to_speak` samples look like
-false positives on a name mention, and approved answer drafts are not yet
-routed into the existing safe playback path.
+The 2026-05-21 recorded session exposed reliability gaps that v1.5 addressed:
+`speaker=unknown` coverage can now be reviewed and bootstrapped, incidental
+name mentions can be separated from direct handoffs, approved answer drafts can
+use the safe playback path, and saved sessions can be replayed as regression
+fixtures.
 
-## Current Milestone: v1.5 Live Review + Safe Approved Speak
+## Current Milestone: None
+
+No active milestone is currently open. v1.5 is archived under
+`.planning/milestones/`.
+
+## Last Completed Milestone: v1.5 Live Review + Safe Approved Speak
 
 **Goal:** Make the live loop reliable after real call recordings by improving
 post-call review, handoff detection, speaker bootstrapping, approved draft
@@ -110,7 +120,8 @@ playback, and regression replay.
 
 ### Active
 
-No active v1.5 requirements remain.
+No active requirements remain. Define the next requirement set with
+`$gsd-new-milestone`.
 
 ### Out of Scope
 
@@ -182,4 +193,4 @@ No active v1.5 requirements remain.
 | Wire only approved generated drafts into playback | Approval/audit is the safety boundary; generated text may play only after an explicit user action | Implemented in v1.5 |
 
 ---
-*Last updated: 2026-05-21 after starting milestone v1.5*
+*Last updated: 2026-05-21 after archiving milestone v1.5*

.planning/ROADMAP.md

@@ -2,13 +2,18 @@
 
 ## Overview
 
-Milestone v1.5 focuses on Live Review + Safe Approved Speak: using real
-captured-call evidence to reduce false handoffs, bootstrap speaker labels,
-close the approved-draft playback loop, and make saved sessions reusable as
-regression fixtures.
+Saymo has completed milestone v1.5 Live Review + Safe Approved Speak. There is
+no active roadmap phase right now; the next product increment should start with
+`$gsd-new-milestone`.
 
 ## Completed Milestones
 
+- ✅ **v1.5 Live Review + Safe Approved Speak** - Phases 18-22 shipped
+  2026-05-21.
+  Archives: `.planning/milestones/v1.5-ROADMAP.md`,
+  `.planning/milestones/v1.5-REQUIREMENTS.md`,
+  `.planning/milestones/v1.5-MILESTONE-AUDIT.md`,
+  `.planning/milestones/v1.5-phases/`
 - ✅ **v1.4 Live Conversation Memory + Answer Cockpit** - Phases 14-17 shipped
   2026-05-20.
   Archives: `.planning/milestones/v1.4-ROADMAP.md`,
@@ -19,177 +24,23 @@ regression fixtures.
   `.planning/milestones/v1.3-REQUIREMENTS.md`,
   `.planning/milestones/v1.3-MILESTONE-AUDIT.md`
 - ✅ **v1.2 Trigger Training Console** - Phases 8-10 shipped 2026-05-20.
-  Archive: `.planning/milestones/v1.2-ROADMAP.md`
+  Archives: `.planning/milestones/v1.2-ROADMAP.md`,
+  `.planning/milestones/v1.2-REQUIREMENTS.md`,
+  `.planning/milestones/v1.2-MILESTONE-AUDIT.md`
 - ✅ **v1.1 Call Intelligence Loop** - Phases 5-7 shipped 2026-05-18.
-  Archive: `.planning/milestones/v1.1-ROADMAP.md`
+  Archives: `.planning/milestones/v1.1-ROADMAP.md`,
+  `.planning/milestones/v1.1-REQUIREMENTS.md`
 - ✅ **v1.0 Speedly Catcher + Speedly Sayer** - Phases 1-4 shipped
   2026-05-15.
 
 ## Active Milestone
 
-- **v1.5 Live Review + Safe Approved Speak** - Phases 18-22 planned.
+None.
 
 ## Phases
 
-**Phase Numbering:**
-- Integer phases (18, 19, 20, 21, 22): Planned milestone work
-- Decimal phases (19.1, 19.2): Urgent insertions if needed
+No active phases remain.
 
-- [x] **Phase 18: Post-call Review Queue** - Prioritize real-session samples
-  that most affect handoff, speaker, and classifier quality.
-- [x] **Phase 19: Addressing v2** - Separate direct handoffs from incidental
-  name mentions and explain the decision path.
-- [x] **Phase 20: Speaker Bootstrap** - Reduce unknown-speaker coverage for
-  completed sessions without requiring live diarization.
-- [x] **Phase 21: Approved Draft Playback** - Play only explicitly approved
-  generated drafts through existing safe TTS/playback paths.
-- [x] **Phase 22: Regression Replay** - Re-run saved sessions through current
-  logic and report reliability metrics.
+## Next Step
 
-## Phase Details
-
-### Phase 18: Post-call Review Queue
-**Goal**: Let the user review the highest-value samples from one completed
-session first and apply labels needed for safer live behavior.
-**Depends on**: v1.2 review commands, v1.4 meeting memory
-**Requirements**: REV-05, REV-06, REV-07
-**Success Criteria** (what must be TRUE):
-  1. A user can open a session-scoped review queue that ranks likely handoffs,
-     questions, name mentions, unknown speakers, and classifier-critical
-     samples ahead of ordinary speech.
-  2. A user can apply category, speaker, and answer-decision labels from the
-     queue without editing JSON directly.
-  3. The queue reports remaining unknown speakers, unlabeled decisions,
-     accepted/rejected counts, and classifier-readiness blockers.
-  4. Existing `trigger-samples review/list/category/decision/label` flows still
-     work for manual review.
-**Plans**: 4 plans
-
-Plans:
-- [x] 18-01: Add post-call review priority scoring and queue model
-- [x] 18-02: Add session-scoped review command and batch-safe label actions
-- [x] 18-03: Add review progress and classifier-readiness diagnostics
-- [x] 18-04: Add docs, edge-case tests, and no-raw-audio report checks
-
-### Phase 19: Addressing v2
-**Goal**: Stop treating every recognized name as an addressed-to-me handoff
-while keeping true handoff detection explainable and tunable.
-**Depends on**: Phase 18
-**Requirements**: INT-02, INT-03, INT-04, INT-05
-**Success Criteria** (what must be TRUE):
-  1. Incidental first-name and full-name/surname mentions can be classified as
-     mention-only instead of `addressed_to_me`.
-  2. Direct handoffs such as "John, what do you think?" still produce an
-     answer candidate or confirmation wait.
-  3. `trigger-check` explains matched phrase, local context, decision path,
-     final action, and skip/answer reason.
-  4. Per-profile config can tune handoff and mention-only patterns without
-     hardcoded private names in source.
-  5. Saved false-positive and true-handoff samples can be replayed against
-     Addressing v2.
-**Plans**: 4 plans
-
-Plans:
-- [x] 19-01: Add addressing evidence model and explainable decision output
-- [x] 19-02: Add direct-handoff vs mention-only heuristics
-- [x] 19-03: Add profile-configured handoff and mention-pattern tuning
-- [x] 19-04: Add saved-sample replay tests for false positives and true handoffs
-
-### Phase 20: Speaker Bootstrap
-**Goal**: Turn completed sessions with `speaker=unknown` into usable
-speaker-labeled training evidence through reviewed local workflows.
-**Depends on**: Phase 18, v1.3 diarization sidecars
-**Requirements**: SPKR-04, SPKR-05, QUAL-04
-**Success Criteria** (what must be TRUE):
-  1. A user can bootstrap speaker labels for one completed session from
-     reviewed diarization suggestions or a lightweight manual grouping flow.
-  2. Bulk label application preserves manual labels as authoritative and keeps
-     suggestion/audit metadata inspectable.
-  3. Session output reports `me`, `other`, and `unknown` counts by category and
-     shows classifier-readiness impact.
-  4. Saymo remains usable when diarization is disabled or unavailable.
-**Plans**: 3 plans
-
-Plans:
-- [x] 20-01: Add speaker bootstrap candidates from sessions and suggestions
-- [x] 20-02: Add bulk speaker label apply flow with audit preservation
-- [x] 20-03: Add speaker coverage/readiness reports and disabled-backend tests
-
-### Phase 21: Approved Draft Playback
-**Goal**: Let the user send only explicitly approved generated draft text into
-the live call through existing local TTS and playback safety paths.
-**Depends on**: v1.4 answer cockpit, existing TTS/playback routing
-**Requirements**: COCK-03, COCK-04, AUD-03, SAFE-01
-**Success Criteria** (what must be TRUE):
-  1. `answer-cockpit action --action speak` can route an approved draft through
-     configured local TTS and playback devices.
-  2. Unapproved generated draft text is refused before synthesis or playback.
-  3. A user can preview/monitor, cancel, or timeout approved draft playback.
-  4. Audit events record draft id, action id, TTS engine, output route,
-     playback start/end, cancellation, and blocked reasons.
-  5. Manual takeover, confirmation, and max-duration safety behavior remains
-     intact.
-**Plans**: 4 plans
-
-Plans:
-- [x] 21-01: Add approved playback request model and blocked-reason gates
-- [x] 21-02: Wire approved drafts into existing TTS and playback routing
-- [x] 21-03: Add preview, monitor, cancel, timeout, and max-duration controls
-- [x] 21-04: Extend audit/reporting and safety regression tests
-
-### Phase 22: Regression Replay
-**Goal**: Make saved sessions reusable as a local reliability test for
-addressing, question detection, speaker coverage, and live-assist readiness.
-**Depends on**: Phases 18-21
-**Requirements**: REG-01, REG-02, REG-03
-**Success Criteria** (what must be TRUE):
-  1. A user can replay one saved capture session through current addressing,
-     question, speaker, and classifier logic without re-recording audio.
-  2. Regression output reports false handoffs, missed handoffs, question noise,
-     speaker coverage, review coverage, and live-assist readiness.
-  3. A user can compare before/after results and export a sanitized report.
-  4. Regression reports omit raw audio and secrets and avoid committing local
-     transcripts or recordings.
-**Plans**: 3 plans
-
-Plans:
-- [x] 22-01: Add saved-session replay engine over local sample metadata
-- [x] 22-02: Add reliability metrics and before/after comparison
-- [x] 22-03: Add sanitized regression reports, docs, and fixture tests
-
-## Progress
-
-**Execution Order:**
-Phases execute in numeric order: 18 -> 19 -> 20 -> 21 -> 22
-
-| Phase | Milestone | Plans Complete | Status | Completed |
-|-------|-----------|----------------|--------|-----------|
-| 18. Post-call Review Queue | v1.5 | 4/4 | Complete | 2026-05-21 |
-| 19. Addressing v2 | v1.5 | 4/4 | Complete | 2026-05-21 |
-| 20. Speaker Bootstrap | v1.5 | 3/3 | Complete | 2026-05-21 |
-| 21. Approved Draft Playback | v1.5 | 4/4 | Complete | 2026-05-21 |
-| 22. Regression Replay | v1.5 | 3/3 | Complete | 2026-05-21 |
-
-## Requirement Coverage
-
-| Requirement | Phase |
-|-------------|-------|
-| REV-05 | Phase 18 |
-| REV-06 | Phase 18 |
-| REV-07 | Phase 18 |
-| INT-02 | Phase 19 |
-| INT-03 | Phase 19 |
-| INT-04 | Phase 19 |
-| INT-05 | Phase 19 |
-| SPKR-04 | Phase 20 |
-| SPKR-05 | Phase 20 |
-| QUAL-04 | Phase 20 |
-| COCK-03 | Phase 21 |
-| COCK-04 | Phase 21 |
-| AUD-03 | Phase 21 |
-| SAFE-01 | Phase 21 |
-| REG-01 | Phase 22 |
-| REG-02 | Phase 22 |
-| REG-03 | Phase 22 |
-
-**Coverage:** 17/17 requirements mapped.
+Run `$gsd-new-milestone` when the next milestone goal is ready.