Build the comprehensive V7 memory platform

.gitattributes

@@ -0,0 +1,7 @@
+* text=auto
+*.json text eol=lf
+*.jsonl text eol=lf
+*.md text eol=lf
+*.py text eol=lf
+*.yaml text eol=lf
+*.yml text eol=lf

.github/CODEOWNERS

@@ -0,0 +1,13 @@
+* @Da7-Tech
+
+/src/mind/00_prelude.py @Da7-Tech
+/src/mind/20_embedding.py @Da7-Tech
+/src/mind/30_graph.py @Da7-Tech
+/src/mind/70_policy.py @Da7-Tech
+/src/mind/80_lifecycle.py @Da7-Tech
+/src/mind/90_cli.py @Da7-Tech
+/SECURITY.md @Da7-Tech
+/.github/workflows/ @Da7-Tech
+/tools/build_single.py @Da7-Tech
+/tools/claims.py @Da7-Tech
+/tools/privacy_scan.py @Da7-Tech

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,7 +1,28 @@
 ## What & why
 
+## Risk and compatibility
+
+- Storage or format change:
+- Security or process-boundary change:
+- Backward-compatibility impact:
+- Rollback or recovery path:
+
+## Evidence
+
+- Red reproduction or absence detector:
+- Dynamic/adversarial result:
+- Differential or unchanged-path result:
+- Raw benchmark artifact, if applicable:
+
 ## Checklist
-- [ ] `python3 -m unittest discover -s tests` passes (all of it)
+- [ ] `python3 tools/build_single.py --check` passes
+- [ ] `python3 tools/claims.py check` passes
+- [ ] `python3 tools/privacy_scan.py --tracked` passes
+- [ ] `python3 -m unittest discover -s tests` passes in full
 - [ ] behavior changes come with a regression test that fails before the fix
-- [ ] measured claims in README/SKILL still true (rerun the relevant bench)
+- [ ] storage changes cover crash, concurrency, old format, oversize, and privacy
+- [ ] retrieval changes cover fallback, noisy, multilingual, and benchmark paths
+- [ ] measured claims point to immutable committed JSON
+- [ ] English and Arabic sections remain substantively equivalent
 - [ ] zero new dependencies (the whole point of the tool)
+- [ ] all review conversations are resolved

.github/dependabot.yml

@@ -0,0 +1,10 @@
+version: 2
+updates:
+  - package-ecosystem: github-actions
+    directory: /
+    schedule:
+      interval: weekly
+    open-pull-requests-limit: 5
+    labels:
+      - dependencies
+      - github-actions

.github/workflows/ci.yml

@@ -3,10 +3,14 @@ on:
   push:
     branches: [main]
   pull_request:
+concurrency:
+  group: tests-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
 permissions:
   contents: read
 jobs:
   test:
+    timeout-minutes: 35
     runs-on: ${{ matrix.os }}
     strategy:
       fail-fast: false
@@ -20,10 +24,26 @@ jobs:
       - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
         with:
           python-version: ${{ matrix.python }}
-      - run: python3 -m unittest discover -s tests -v
-      - run: python3 bench/bench.py
-      - run: python3 bench/multilang.py
-      - run: python3 bench/discrim.py
-      - run: python3 bench/longmemeval.py --data tests/fixtures/longmemeval_tiny.json --limit 2
-      - run: python3 bench/soak.py
-      - run: python3 bench/fuzz.py --quick
+      - name: Structural and claims gates
+        run: |
+          python tools/build_single.py --check
+          python tools/claims.py check
+          python tools/privacy_scan.py --tracked
+          python -m compileall -q mind.py src/mind bench contrib tools
+      - name: Complete test suite
+        run: python -m unittest discover -s tests -v
+      - name: Retrieval and lifecycle benchmarks
+        run: |
+          python bench/bench.py
+          python bench/multilang.py
+          python bench/discrim.py
+          python bench/slots.py
+          python bench/soak.py
+          python bench/fuzz.py --quick
+      - name: Immutable-harness smoke tests
+        run: |
+          python bench/longmemeval.py --data tests/fixtures/longmemeval_tiny.json --limit 2
+          python bench/longmemeval.py --data tests/fixtures/longmemeval_tiny.json --limit 2 --engine bm25
+          python bench/paraphrase.py --server-cmd "python contrib/concept_embed_server.py"
+          python bench/bulk.py --records 1000 --serial-sample 50 --serial-repetitions 1
+          python bench/autonomy.py --quick

.github/workflows/release-dry-run.yml

@@ -0,0 +1,34 @@
+name: release-dry-run
+on:
+  workflow_dispatch:
+  push:
+    tags:
+      - "v*"
+permissions:
+  contents: read
+jobs:
+  verify:
+    timeout-minutes: 45
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          fetch-depth: 0
+      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
+        with:
+          python-version: "3.14"
+      - name: Rebuild and verify public claims
+        run: |
+          python tools/build_single.py --check
+          python tools/claims.py check
+          python tools/privacy_scan.py --tracked
+      - name: Full local release gates
+        run: |
+          python -m unittest discover -s tests -v
+          python bench/bench.py
+          python bench/multilang.py
+          python bench/discrim.py
+          python bench/slots.py
+          python bench/soak.py
+          python bench/fuzz.py --quick
+          python bench/autonomy.py --quick

CHANGELOG.md

@@ -2,23 +2,36 @@
 
 ## Unreleased
 
-- Added `bench/longmemeval.py`, an independent LongMemEval retrieval harness
-  that downloads the official cleaned oracle dataset by default, maps history
-  sessions to `remember()` and questions to `recall()`, and reports exact
-  evidence-turn retrieval (with answer-session fallback only for unannotated
-  sessions), answer-string retrieval, latency, dataset SHA-256, git commit,
-  and the exact command.
-- CI now smoke-tests the LongMemEval harness against a tiny local fixture,
-  keeping the public dataset download out of the test matrix.
-- Added an optional `MIND_EMBED_CMD` recall re-ranker. The command reads text
-  from stdin and returns a numeric vector as JSON or whitespace-separated
-  floats; failures, timeouts, invalid vectors, and unset commands fall back to
-  the deterministic hash embedder without mixing vector spaces. Transient
-  failures are retried after a short cache window, and oversized output stays
-  off the Python heap.
-- Benchmarks now report the default offline recall column and, when
-  `MIND_EMBED_CMD` is set, a second embedded re-rank column so semantic
-  backend gains can be measured without changing the offline baseline.
+- Development identity is now `7.0.0.dev0`, distinct from the pinned stable
+  `6.2.10` artifact. Verbose version output includes commit and artifact hash.
+- Split development source into ten domain fragments under `src/mind/` while
+  preserving one deterministic, standard-library-only `mind.py` artifact.
+- Closed archive, signals, CRLF/BOM export, cortex ownership, failed
+  transaction, directional relation, journal accounting, FIPS, Windows,
+  command resolution, cache-bound, dash-query, conflict-age, temporary-file,
+  and redundant-write defects from the consolidated audit.
+- Replaced per-candidate semantic processes with one versioned batch per
+  ranking, a total deadline, process-group termination, whole-ranking
+  fallback, bounded caches, explain receipts, and an optional persistent
+  framed sidecar.
+- Added policy-gated automatic capture, quarantine approval, typed/scoped
+  memory, explicit user-global memory, slot conflicts, context recipes, and a
+  same-file stdio protocol server with seventeen tools.
+- Added journal format 2 with stable event IDs and UTC nanoseconds,
+  deterministic three-way journal replay merge, backup/checkpoint/restore,
+  segmentation/compaction, doctor, growth, and crash-resumable
+  forget/unlink/redact/purge.
+- Rebuilt mutation analysis with a green baseline preflight, complete staging,
+  structured classifications, bounded diagnostics, source hashes, and a
+  second LongMemEval-harness target.
+- Added immutable LongMemEval manifests, an identical-mapping BM25 baseline,
+  paraphrase, bulk-ingest, thirty-session, five-year, and public raw-result
+  gates with privacy-safe clean-commit provenance.
+- Replaced hand-maintained volatile claims with generated facts and benchmark
+  tables, restored English/Arabic structural parity, and documented the
+  default/optional security boundary and three host quickstarts.
+- Added code ownership, pinned-action update automation, bounded CI jobs,
+  release and privacy gates, and an explicit solo-maintainer review policy.
 
 ## 6.2.10 — 2026-07-12
 

CONTRIBUTING.md

@@ -1,16 +1,67 @@
 # Contributing
 
-Ground rules (they keep the tool what it is):
-
-1. **`mind.py` stays a single stdlib-only file.** No dependencies, no
-   network calls in the default path. Helper tooling may live in
-   `bench/`/`tests/` and use whatever it needs.
-2. **Every behavior change needs a test** (`tests/`, stdlib `unittest`):
-   `python3 -m unittest discover -s tests`
-3. **Every performance or lifecycle claim needs a measurement**:
-   `python3 bench/bench.py` and `python3 bench/soak.py` must stay green —
-   both run in CI on Linux/macOS/Windows × 3 Python versions.
-4. Scoped, ready-to-pick-up work lives in the
-   [issues](https://github.com/Da7-Tech/mind/issues).
-
-If you're unsure, open a Discussion first — happy to scope things together.
+## Invariants
+
+1. The default kernel is deterministic, offline, and standard-library only.
+2. `src/mind/` is the source of truth. Never hand-edit generated `mind.py`.
+3. `python3 tools/build_single.py` must reconstruct `mind.py` byte-for-byte.
+4. Plain files remain the default storage, and current plus previous formats
+   remain readable.
+5. Recall stays a pure read. Durable writes are atomic, serialized,
+   recoverable, and journaled.
+
+## Change Protocols
+
+Every behavior change needs an absence detector or red regression test.
+
+Storage changes must cover:
+
+- every durable crash point;
+- concurrent writers;
+- the previous format;
+- below, at, and beyond each size boundary;
+- backup, restore, compaction, redaction, and purge behavior.
+
+Retrieval changes must cover:
+
+- clean and noisy corpora;
+- competing vocabulary and near duplicates;
+- multilingual and temporal cases;
+- empty/no-result precision;
+- semantic failure and whole-ranking fallback;
+- immutable LongMemEval evidence.
+
+Every bounded constant needs a crossing test and a lifecycle policy. Every
+quality harness needs a green baseline preflight and a planted-defect
+self-test.
+
+## Required Local Gates
+
+```bash
+python3 tools/build_single.py --check
+python3 tools/claims.py check
+python3 tools/privacy_scan.py --tracked
+python3 -m unittest discover -s tests -v
+python3 bench/bench.py
+python3 bench/multilang.py
+python3 bench/discrim.py
+python3 bench/slots.py
+python3 bench/soak.py
+python3 bench/fuzz.py --quick
+python3 bench/autonomy.py --quick
+```
+
+Performance claims require immutable input, source commit, raw JSON, and a
+README summary generated from that JSON. Do not edit generated facts or
+benchmark tables manually.
+
+## Review
+
+Storage, privacy, semantic-process, protocol, workflow, and release changes
+request code-owner review. Resolve every review conversation. In a
+solo-maintainer repository, an administrator bypass is explicit governance
+debt and is not independent review.
+
+Scoped work lives in the
+[issues](https://github.com/Da7-Tech/mind/issues). Use a Discussion when the
+format, compatibility, or product boundary is still uncertain.