ai-memory v0.7.0 — 2026-05-28 Ship Campaign Dossier

What this is

The 2026-05-28 ship campaign dossier for ai-memory v0.7.0. The branch release/v0.7.0 (HEAD be3347d70) was put through the integrated post-#1174 + post-CI-flake-closure-sweep sweep against the dogfooded binary, the rebuilt lan-parity-test Docker stack, and the full postgres+AGE regression — plus a 100% sweep of the docs + GitHub Pages drift surface (closes #1197 + #1198) and a scope-disciplined CoALA prior-art citation (#1380). Six tracks were exercised under one matrix:

• Track A — Build + 1hr dogfood loop. Live MCP daemon PID 10338 stayed alive 1h32m+ continuously against the integrated-HEAD binary (started 2026-05-27 16:55:25Z; dogfood window 2026-05-28T11:44:53Z → 12:44:53Z; RSS ~18 MB throughout, no leak, no anomalies). PASSED.
• Track B — A2A in-host regression via lan-parity stack. alice + bob daemons exercised via the postgres+AGE test suite (federation paths probed via cargo test --features sal,sal-postgres against the lan-parity URLs). Cross-node Track D (192.168.50.100 ↔ 192.168.1.50) remains operator-blocked per #836.
• Track C — Postgres + Apache AGE full regression. cargo test --features sal,sal-postgres --release --no-fail-fast against lan-parity pg-age (127.0.0.1:15432). Result: 8,028 passed / 9 failed / 27 ignored / 312 suites. Triage: 5 of 9 were cargo-target races (cleared on isolated re-run); 4 of 9 were real test-isolation defects (auto_migrate_384, issue_1213 ×2, migrate_links_roundtrip), filed as #1381 and fixed in PR #1382 (per-test PostgresTestEnv schema isolation helper).
• Track D — Docs + Pages drift remediation. Codegraph-driven detection across all v0.7.0 docs + GitHub Pages surfaces. 41 unique drift sites across 23 distinct files (initial pass) → 6 more from Audit A round 1 → 7 more from Audit B + Audit A retest = ~55 total fix sites across 27 distinct files in 3 commits on PR #1379. Closes #1197 + #1198.
• Track E — CoALA prior-art citation. PR #1380. 3 files touched (ROADMAP.md, docs/positioning.md, docs/strategy/coala-mapping.md NEW). Scope-test-disciplined per operator work order: 3 rejected proposals, no substrate code, no capability-block touched. ZERO-DEFECTS-CONFIRMED.
• Track F — Heterogeneous AI-NHI assessment v3. Parallel- dispatched to a fresh Opus 4.7 agent against the post-drift-fix HEAD. Report expected at docs/v0.7.0/heterogeneous-ai-nhi-assessment/report-claude-opus-4-7-v3.md.

5 GitHub issues were touched this campaign: #1197 (docs drift NO FAIL MISSION, closed via #1379), #1198 (Pages drift NO FAIL MISSION, closed via #1379), #1378 (install codex TOML support — filed during install verification, open, not blocking), #1381 (postgres test-isolation defects, closed via #1382), #1382 (the fix PR itself). The 3 audits under pm-v3.2 NO FAIL MISSION discipline (Audit A / B / C) verified the docs+Pages fix batch. Audit C: ZERO-DEFECTS-CONFIRMED on the load-bearing pinning tests, with a latent-drift-risk flagged for the v0.8 doc-drift-detection workflow.

Verdict at a glance

SHIP-CLEARED. The Track A (build + 1hr dogfood), Track B (A2A in-host via lan-parity), Track C (postgres+AGE regression + #1381 fix PR #1382), Track D (docs + Pages drift 100% remediation closing #1197 + #1198), and Track E (CoALA citation #1380) checkboxes for #836 are agent-completable and now GREEN against tip be3347d70 (with PR #1382 + PR #1379 + PR #1380 pending operator merge). The remaining Track D cross-node routing (192.168.50.100 ↔ 192.168.1.50) and the 24h dogfood-rebuild loop are explicitly operator-gated by #836’s own checklist.

How this directory is organised

Issues touched in this campaign

Categorical breakdown

The 4 #1381 sub-defects: auto_migrate_converts_384_schema_to_768_on_daemon_bootstrap, issue_1213_unscoped_probe_demonstrates_root_cause, issue_1213_atttypmod_probe_scopes_to_public_schema, migrate_links_sqlite_to_postgres_to_sqlite_roundtrip — all share one root cause (tests assume an empty/known schema state on the postgres container, but other tests + the long-lived ic_alice / ic_bob daemon schemas from the lan-parity stack accumulate state). Substrate itself is GREEN; the fix lives entirely on the test side.

Code commits + PRs this campaign

Canonical counts at HEAD be3347d70

Every authoritative number cited in this dossier resolves through the codebase via the path noted; the docs-drift sweep (Track D) is specifically the closure that brought every doc page in line with this table.

Reproducibility contract

1. Branch + tip. release/v0.7.0 (HEAD be3347d704dad03bcc210c9eb0a517946dbe555f).
2. Binary. Built from release/v0.7.0 tip at target/release/ai-memory; install symlinked to /opt/homebrew/bin/ai-memory per scripts/dogfood-rebuild.sh.
3. Lan-parity stack. infra/lan-parity-test/docker-compose.yml spun up via docker compose up -d --build. Three containers: ai-memory-lan-parity-alice (HTTP 127.0.0.1:19180), ai-memory-lan-parity-bob (HTTP 127.0.0.1:19181), ai-memory-lan-parity-pg-age (postgres 127.0.0.1:15432, PG16 + AGE 1.6.0 + pgvector 0.8.2).
4. Test invocation (Track C). cargo test --features sal,sal-postgres --release --no-fail-fast with AI_MEMORY_TEST_POSTGRES_URL=postgresql://aimemory:****@127.0.0.1:15432/aimemory and AI_MEMORY_TEST_AGE_URL bound to the same URL.
5. Schema version. v51 (current at release/v0.7.0 HEAD; postgres ladder ends at migrate_v51(), sqlite at the if version < 51 arm; per-namespace quota at v50 #1156; federation nonces table at v51 #1255 / PR #1296).
6. Authoring agent. Claude (Opus 4.7, 1M context). Three pm-v3.2 codegraph-driven QC audits dispatched during Track D (Audit A literal enumeration; Audit B structural call-graph; Audit C regression-invariance fault-injection).

Hard rules during the campaign

Per the prime directive pm-v3.3 (memory pm-v3.3-supersedes namespace global/policies) and the testing-loop addendum:

• Testing-loop discipline. Every defect surfaced during the campaign was filed as a GH issue at the moment of discovery, not after the campaign closed. The 4 #1381 sub-defects became one issue + one PR (#1382). The codex-install TOML defect became #1378. The docs+Pages drift became fixes on existing #1197 + #1198.
• Verify-before-claiming. Every “tests pass” claim cites the exact cargo test invocation + result line. Every “I committed X” cites a SHA verifiable via git show <SHA>. No banned phrases (“non-blocking”, “DEFER-TO-V080”, “operator should…”) were used.
• Recompile-retest discipline (pm-v3.3 step 7). The Track C postgres+AGE regression run was against the rebuilt binary at release/v0.7.0 HEAD, not against the long-running MCP PID 10338’s older snapshot. The 4 #1381 failures reproduced deterministically against the freshly-built test binaries.
• NO FAIL MISSION on Track D drift remediation. 3 audit passes (A literal, B structural, A retest + B remaining) until ZERO-DEFECTS-CONFIRMED. The Audit C regression-invariance check flagged latent-drift-risk for v0.8 doc-drift-detection workflow but ZERO-DEFECTS-CONFIRMED on the load-bearing pinning tests.
• Scope discipline on the CoALA citation PR. Per operator work order: 3 deliverables (ROADMAP citation, positioning section, reference doc), 3 rejected proposals (no §2.8 subsection, no §11.4.D/§22 reframing, no coala capabilities-v3 block), no substrate code, no schema change.
• Audit trail mandatory. Every GH issue body links to the commit/PR; every commit references the issue; this dossier links both.

Memory namespace convention

Provenance

Apache-2.0, © 2026 AlphaOne LLC. Drafted by Claude (Opus 4.7, 1M context).