feat: add governed graph observations

.planning/GRAPH-OBSERVATIONS-V1.md

@@ -0,0 +1,63 @@
+<!-- doc-head: active PPR-7 graph-observations implementation ledger -->
+# Graph Observations V1
+# Covers: deterministic discovery, governed synthesis, lifecycle, opt-in recall, and rollout evidence.
+# Key terms: exact signatures, union-find, support fingerprint, candidate observation, steward gate.
+# Read when: implementing, reviewing, testing, or rolling back PPR-7.
+# Status: implemented and locally verified; separate PR and CI evidence in progress.
+<!-- /doc-head -->
+
+## Fixed boundary
+
+The feature derives optional observations only from current, confirmed,
+non-sensitive claims with active evidence and graph-edge supports in one exact
+tenant and scope. Components are deterministic; an LLM may summarize an
+eligible component but cannot choose membership, cite outside it, promote its
+own output, or feed observations back into extraction.
+
+## Work ledger
+
+| Area | Required outcome | Status |
+|---|---|---|
+| Migration 0020 | Add SQLite observation, support, and leased-job tables; PostgreSQL fails closed. | Implemented; migration tests pass |
+| Discovery | Canonical signatures, hub suppression, union-find, bounds, deterministic fingerprints. | Implemented; 48-case evaluator is 100%/100% |
+| Synthesis | Strict structured output, three-call cycle cap, provider/global budget, replay-safe `no_signal`. | Implemented; focused failure/cap tests pass |
+| Lifecycle | Candidate-only creation, deterministic steward gate, immediate archive/stale on support change. | Implemented; promotion/retirement tests pass |
+| Recall | Default byte-equivalent output; separately packed opt-in observations bounded to five. | Implemented; equivalence and opt-in tests pass |
+| Surfaces | Python, CLI, MCP, dashboard, and disposable demo parity. | Implemented; focused surface tests pass |
+| Evaluation | Versioned 40+ case corpus plus structural, citation, leakage, replay, and retrieval gates. | Local gates pass; PR CI pending |
+
+## Acceptance evidence
+
+- Structural component precision and recall are each at least 95% on the
+  versioned corpus; observation precision is at least 90% and root-cause
+  precision at least 85%.
+- Citation/support correctness is 100%, with zero cross-tenant, cross-scope,
+  sensitive, retired, stale, candidate, or observation-generated support.
+- Ordinary recall is byte-equivalent when observations are disabled; graph
+  top-five improves without more than 0.01 overall R@5/MRR regression.
+- `improve()` only queues work and remains below 500 ms p95; an hourly scope
+  cycle performs at most three observation synthesis calls.
+- Focused tests, the full non-ML suite, retrieval gates, Ruff, collection,
+  migration/restore, clean-wheel, supply-chain, and GitNexus checks pass before
+a PPR-7 pull request is proposed.
+
+## Local verification evidence
+
+- Offline corpus: 48 cases, structural precision 1.00 and recall 1.00.
+- Full non-ML suite: 4,463 passed, 72 skipped, 97 deselected, one expected
+  failure; Ruff and 4,633-test collection pass.
+- Focused restore/retrieval gates: 17 passed; release/supply-chain contract
+  tests: 86 passed; regenerated release truth verifies.
+- Clean wheel builds, passes Twine, installs into a fresh venv, initializes
+  migration 0020, and completes the observation promotion/recall/staleness demo.
+- Feature generation remains disabled by default; the configured Dreaming
+  extraction/consolidation pair remains Gemini plus GLM.
+
+## Rollout and rollback
+
+Roll out through disposable SQLite and fake/local providers, feature-off wheel
+installation, offline shadow evaluation, verified authoritative backup/restore,
+then opt-in candidate generation. The 24-hour observation is post-implementation
+operational evidence, not a coding prerequisite. Rollback disables generation
+and recall, archives candidates, marks confirmed generated observations stale,
+and preserves the additive tables and audit history.

DOCS-MAP.md

@@ -1,16 +1,17 @@
-<!-- doc-head: canonical documentation map; P5 verifier passed, PR #189 open -->
+<!-- doc-head: canonical documentation map; P5 integrated, PPR-7 active -->
 # DOCS-MAP - memorymaster
 # Covers: trust verdicts and replacements for every canonical documentation surface.
 # Key terms: CURRENT, SUPERSEDED, ABANDONED, GENERATED, roadmap, paper radar, ADR.
 # Read when: locating authoritative project documentation before reading doc bodies.
-# Updated: 2026-08-12 after repair5 verifier replay and PR #189 creation.
-# Rule: PR #189 is open; merge, release, deploy, and PPR-7 remain blocked pending separate review and integration.
+# Updated: 2026-08-12 after PR #189 merge, governed runtime deploy, and PPR-7 authorization.
+# Rule: PPR-7 is isolated and opt-in; public release and automatic recall remain separately gated.
 <!-- /doc-head -->
 
 | File | Verdict | Last change | Reason |
 |---|---|---|---|
 | CHANGELOG.md | CURRENT | 2026-08-04 | Public release history; v4.6.0 records governed universal capture, measured quality changes, security evidence, and known follow-ups. |
-| ROADMAP.md | CURRENT | 2026-08-12 | Sole authoritative roadmap; elapsed P5 evidence, pinned Gemini+GLM replay, repair5 verifier pass, and PR #189 open. |
+| ROADMAP.md | CURRENT | 2026-08-12 | Sole authoritative roadmap; P5 is merged and locally deployed, while isolated PPR-7 implementation is active. |
+| .planning/GRAPH-OBSERVATIONS-V1.md | CURRENT | 2026-08-12 | Bounded PPR-7 implementation ledger for deterministic supported components, governed synthesis, lifecycle, opt-in recall, and verification. |
 | .planning/PAPER-RADAR-REVIEW-2026-08-08.md | CURRENT | 2026-08-08 | Primary-paper ledger covers 57-paper triage, 18 deep reviews, exact MemoryMaster gaps, and ordered PPR-1 through PPR-6 decisions subordinate to ROADMAP.md. |
 | .planning/PAPER-RESEARCH-IMPLEMENTATION-2026-08-08.md | CURRENT | 2026-08-08 | Executable status ledger for PPR-1 through PPR-6; records acceptance criteria and evidence without competing with ROADMAP.md. |
 | .planning/HERMES-SCOPE-SKILLS-INTEGRATION-2026-08-07.md | CURRENT | 2026-08-12 | Executable Tencent-derived ledger; repair5 passed and PR #189 is open without merge authority. |
@@ -24,7 +25,7 @@
 | .planning/VNEXT-GOVERNED-CAPTURE-SPEC.md | CURRENT | 2026-07-27 | Bounded implementation specification that explicitly implements, and does not compete with, ROADMAP.md. |
 | .planning/VNEXT-BASELINE-2026-07-27.md | CURRENT | 2026-07-27 | Reproducible pre-change retrieval, test, latency, package, capture, graph, and scheduler baseline at d33a268. |
 | docs/adr/0015-governed-universal-capture-lineage.md | CURRENT | 2026-07-27 | Accepted data-flow decision fixing producer-to-source-to-evidence-to-claim-to-supported-graph lineage and retirement semantics. |
-| docs/public-v1.md | CURRENT | 2026-08-08 | Stable facade, capture boundary, additive approved-skill recall, retirement semantics, dashboard inbox, and demo. |
+| docs/public-v1.md | CURRENT | 2026-08-12 | Stable facade, capture boundary, opt-in approved skills and observations, retirement semantics, dashboard panels, and demo. |
 | .planning/audits/2026-07-27-vnext-governed-capture/audit-delta.md | CURRENT | 2026-08-04 | SQLite activation, LifeAgent retirement, OAuth capture quality, public v4.6.0 release, rollback evidence, and the final seven-day observation gate. |
 | docs/archive/IMPROVEMENT_PLAN.md | ABANDONED | 2026-06-20 | The doc serves as a generated audit and roadmap from March 2026 but is not referenced by any current docs; it contains specific version claims and 'P0' bugs that likely represent a historical snapshot rather than a living plan.  |
 | docs/archive/v315-experiments/E02-results.md | ABANDONED | 2026-06-20 | This is a negative result experiment from a past version (v315) where the code was explicitly reverted and not retained.  |

ROADMAP.md

@@ -1,10 +1,10 @@
-<!-- doc-head: sole authoritative roadmap; P5 verifier passed, PR #189 open -->
+<!-- doc-head: sole authoritative roadmap; P5 merged/deployed, PPR-7 active -->
 # MemoryMaster roadmap
 # Covers: post-v4.6 sequence, Tencent-derived work, paper research, and deferrals.
 # Key terms: Hermes, governed skills, paper radar, temporal projection, sustainability.
 # Read when: choosing release scope, accepting a feature, or checking deferrals.
 # Authority: sole roadmap; planning ledgers implement it and never replace it.
-# Safety: SQLite authority and steward promotion remain fixed; PR #189 is open, never authorizing merge, release, deployment, or PPR-7.
+# Safety: SQLite authority and steward promotion remain fixed; PPR-7 stays additive, opt-in, and unreleased pending its own evidence.
 <!-- /doc-head -->
 
 ## Shipped in v4.6.0
@@ -59,8 +59,15 @@ its governed-claims authority:
 - The pinned Gemini Flash Lite plus GLM 5.2 scheduled replay passes: exit 0,
   eight extraction/consolidation/application decisions, zero errors, and one
   stale crash run recovered. Repair5 independently replays the bounded gate
-  against the elapsed baseline and passes; [PR #189](https://github.com/wolverin0/memorymaster/pull/189)
-  is open, while merge, release, deployment, and PPR-7 remain prohibited.
+  against the elapsed baseline and passes. [PR #189](https://github.com/wolverin0/memorymaster/pull/189)
+  merged as `d34b80c` and its clean wheel is active in the governed local
+  runtime with healthy MCP HTTP probes; the scheduled Dreaming pair remains
+  Gemini Flash Lite plus GLM 5.2.
+- PPR-7 governed graph observations are implemented and locally verified on an
+  isolated feature branch. Implementation is additive, candidate-first, and
+  opt-in; its separate PR/CI, feature-off deployment, and rollout evidence are
+  still required before activation. No public release or automatic
+  ordinary-recall inclusion is implied.
 - The invalid earlier window remains incident evidence only because it included
   a VM OOM/gateway interruption and did not contain P5.
 - Keep v4.6.0 operational while the post-release Obsidian opt-in and OpenCode
@@ -78,7 +85,12 @@ its governed-claims authority:
   progressive approved-skill reuse are implemented, verified, and active.
   Windows snapshot/readiness gates, consoleless P5 runtime replacement, VM
   package rollback preparation, and live functional probes passed. The repair5
-  bounded verifier now passes; create a PR only, with no merge or follow-on PPR-7 work.
+  bounded verifier passed, PR #189 merged, and the governed local runtime was
+  upgraded with a clean wheel and healthy post-restart probe.
+- Review and integrate PPR-7 according to `.planning/GRAPH-OBSERVATIONS-V1.md`;
+  exact support signatures, deterministic discovery, candidate-only synthesis,
+  observation-specific steward validation, and explicit recall inclusion are
+  implemented with local evidence, while PR/CI and feature-off rollout remain.
 - Improve personal/local backup guidance beyond the already verified disposable
   backup/restore and migration procedure.
 - Keep semantic recall optional and disabled unless a local user deliberately

benchmarks/fixtures/graph_observations_v1.json

@@ -0,0 +1,55 @@
+{
+  "corpus_version": "graph-observations-v1",
+  "algorithm_version": "graph-observations-union-find-v1",
+  "ontology_version": "personal-v1",
+  "cases": [
+    {"id":"dep-01","category":"dependency_chain","template":"eligible","expected":[[1,2,3]]},
+    {"id":"dep-02","category":"dependency_chain","template":"eligible","expected":[[1,2,3]]},
+    {"id":"dep-03","category":"dependency_chain","template":"merge","expected":[[1,2,3,4]]},
+    {"id":"dep-04","category":"dependency_chain","template":"split","expected":[[1,2,3],[4,5,6]]},
+    {"id":"root-01","category":"root_cause","template":"eligible","expected":[[1,2,3]]},
+    {"id":"root-02","category":"root_cause","template":"merge","expected":[[1,2,3,4]]},
+    {"id":"root-03","category":"root_cause","template":"eligible","expected":[[1,2,3]]},
+    {"id":"root-04","category":"root_cause","template":"split","expected":[[1,2,3],[4,5,6]]},
+    {"id":"pattern-01","category":"recurring_pattern","template":"eligible","expected":[[1,2,3]]},
+    {"id":"pattern-02","category":"recurring_pattern","template":"merge","expected":[[1,2,3,4]]},
+    {"id":"pattern-03","category":"recurring_pattern","template":"eligible","expected":[[1,2,3]]},
+    {"id":"pattern-04","category":"recurring_pattern","template":"split","expected":[[1,2,3],[4,5,6]]},
+    {"id":"unrelated-01","category":"unrelated_similarity","template":"unrelated","expected":[]},
+    {"id":"unrelated-02","category":"unrelated_similarity","template":"unrelated","expected":[]},
+    {"id":"unrelated-03","category":"unrelated_similarity","template":"one_signature","expected":[]},
+    {"id":"unrelated-04","category":"unrelated_similarity","template":"insufficient_claims","expected":[]},
+    {"id":"hub-01","category":"hub","template":"hub","expected":[]},
+    {"id":"hub-02","category":"hub","template":"hub","expected":[]},
+    {"id":"hub-03","category":"hub","template":"hub","expected":[]},
+    {"id":"bounds-01","category":"bounds","template":"oversized","expected":[]},
+    {"id":"bounds-02","category":"bounds","template":"oversized","expected":[]},
+    {"id":"conflict-01","category":"conflict","template":"excluded","state":"conflicted","expected":[]},
+    {"id":"conflict-02","category":"conflict","template":"excluded","state":"candidate","expected":[]},
+    {"id":"conflict-03","category":"conflict","template":"excluded","state":"stale","expected":[]},
+    {"id":"retired-01","category":"retired_evidence","template":"excluded","state":"retired","expected":[]},
+    {"id":"retired-02","category":"retired_evidence","template":"excluded","state":"retired","expected":[]},
+    {"id":"retired-03","category":"retired_evidence","template":"excluded","state":"retired","expected":[]},
+    {"id":"scope-01","category":"scope_boundary","template":"cross_scope","expected":[[1,2,3]]},
+    {"id":"scope-02","category":"scope_boundary","template":"cross_scope","expected":[[1,2,3]]},
+    {"id":"scope-03","category":"scope_boundary","template":"excluded","state":"wrong_scope","expected":[]},
+    {"id":"tenant-01","category":"tenant_boundary","template":"cross_tenant","expected":[[1,2,3]]},
+    {"id":"tenant-02","category":"tenant_boundary","template":"cross_tenant","expected":[[1,2,3]]},
+    {"id":"tenant-03","category":"tenant_boundary","template":"excluded","state":"wrong_tenant","expected":[]},
+    {"id":"secret-01","category":"sensitive_data","template":"excluded","state":"sensitive_claim","expected":[]},
+    {"id":"secret-02","category":"sensitive_data","template":"excluded","state":"sensitive_evidence","expected":[]},
+    {"id":"secret-03","category":"sensitive_data","template":"excluded","state":"sensitive_source","expected":[]},
+    {"id":"candidate-01","category":"lifecycle","template":"excluded","state":"candidate","expected":[]},
+    {"id":"stale-01","category":"lifecycle","template":"excluded","state":"stale","expected":[]},
+    {"id":"observation-01","category":"feedback_loop","template":"excluded","state":"observation","expected":[]},
+    {"id":"skill-01","category":"feedback_loop","template":"excluded","state":"skill","expected":[]},
+    {"id":"summary-01","category":"feedback_loop","template":"excluded","state":"summary","expected":[]},
+    {"id":"observer-01","category":"feedback_loop","template":"excluded","state":"observer_agent","expected":[]},
+    {"id":"minimum-01","category":"eligibility","template":"insufficient_claims","expected":[]},
+    {"id":"minimum-02","category":"eligibility","template":"insufficient_evidence","expected":[]},
+    {"id":"minimum-03","category":"eligibility","template":"one_signature","expected":[]},
+    {"id":"symmetric-01","category":"canonicalization","template":"symmetric","expected":[[1,2,3]]},
+    {"id":"replay-01","category":"determinism","template":"eligible","expected":[[1,2,3]]},
+    {"id":"replay-02","category":"determinism","template":"merge","expected":[[1,2,3,4]]}
+  ]
+}

docs/generated/release-truth.json

@@ -134,7 +134,7 @@
     "console_entrypoints": 8,
     "mcp_tools": 50,
     "ops_cli_commands": 5,
-    "pytest_test_functions": 3767
+    "pytest_test_functions": 3781
   },
   "feature_profile_matrix": {
     "capture_hook": [

docs/generated/release-truth.md

@@ -7,7 +7,7 @@ Do not edit this file by hand. Run `python scripts/generate_release_truth.py`.
 - Main CLI commands: **118**
 - Operations CLI commands: **5**
 - Console entrypoints: **8**
-- Pytest source test functions: **3767**
+- Pytest source test functions: **3781**
 
 ## MCP tools