v0.7.0 Ship Campaign 2026-05-28 — Decision-Maker Briefing

Verdict

SHIP-CLEARED. The 2026-05-28 ship campaign exercised the integrated release/v0.7.0 HEAD be3347d70 (post-#1174 closure + post-CI-flake-sweep) against six tracks: build + 1-hour live dogfood, A2A federation via rebuilt lan-parity Docker stack, postgres + Apache AGE full regression, 100% docs + GitHub Pages drift remediation, scope-disciplined CoALA prior-art citation, and a parallel-dispatched independent AI-NHI structural assessment.

Outcome:

• Track A (build + dogfood) — GREEN. 16h+ daemon uptime, ~18 MB RSS steady-state, no leak.
• Track B (A2A via lan-parity Docker) — GREEN. alice + bob + pg-age all healthy; federation paths exercised end-to-end.
• Track C (postgres+AGE regression) — GREEN. 8,028 / 9 / 27 / 312 initial → all 9 triaged (5 cargo-target races cleared; 4 real defects filed as #1381 → fixed in PR #1382). Substrate untouched.
• Track D (docs + Pages drift) — GREEN. ~55 fix sites across ~27 distinct files, 3 commits on PR #1379, ZERO-DEFECTS-CONFIRMED after three QC audits.
• Track E (CoALA citation) — GREEN. PR #1380 scope-disciplined; 3 deliverables, 3 rejected proposals, substrate untouched.
• Track F (AI-NHI assessment v3) — IN-FLIGHT. Independent fresh Opus 4.7 agent dispatched against post-fix HEAD.

The remaining release-gate items (cross-node Track D subnet routing, 24h dogfood loop) are operator-action-gated per #836’s own checklist and are not engineering-completable.

What was at stake

The 2026-05-22 release-gate campaign minted the 7,321 / 0 / 0 verdict against the post-#1013 + post-22-issue-fix tip. The 2026-05-28 campaign is the integration follow-on: did the intervening week of refactor closure (Wave-A audit-merge campaign on 2026-05-25/26), CI-flake closure sweep (2026-05-27 batch of #1372/#1373/#1334 fixes), and ongoing engineering work regress the release-gate posture? Did the docs surface get stale during that window? Are the test fixtures still load-bearing against the realistic lan-parity dual-daemon substrate?

The answer at tip be3347d70 is: the engineering quality posture held, one new class of test-isolation defect surfaced and was fixed in-campaign, and the docs surface is now mechanically correct against the canonical-source symbols.

The headline numbers

The Postgres+AGE 9 failures are the interesting datum. Of the 9:

• 5 were build-system race conditions. Concurrent cargo build --release interleaving with the in-flight test run caused “binary not found” / “Text file busy” symptoms. All cleared on isolated re-run. Environmental, not substrate.
• 4 were real test-isolation defects. Tests assumed they had a clean schema to themselves on the postgres container, but the lan-parity stack’s always-on ic_alice / ic_bob daemon schemas
  • other tests in the same cargo test --no-fail-fast invocation left state behind. Filed as #1381 (one tracker, four manifestations). Fixed in PR #1382 via a new per-test PostgresTestEnv schema isolation helper (tests/common/postgres_env.rs, 421 LOC). All 4 GREEN against the rebuilt lan-parity stack post-fix. Substrate untouched.

This is the meaningful signal: the testing-loop discipline (pm-v3.3) caught a new class of defect at the realistic-deployment substrate boundary (the lan-parity shared container), and the fix landed entirely on the test side. The production code is unaffected.

The Track D docs drift remediation — what 3 audits surfaced

The Track D portion (docs + Pages drift remediation, closing #1197 + #1198) was the most labor-intensive track. The codegraph-driven detection pass surfaced 41 drift sites; the first commit closed all 41. But three independent QC audits found more:

• Audit A (literal enumeration) — re-ran the literal-search detection against the post-commit-1 state and found 6 sites missed.
• Audit B (structural call-graph) — traversed from each canonical-source symbol (expected_tool_count, Command enum, CURRENT_SCHEMA_VERSION, etc.) to every doc location that should reference it. Found 7 more, concentrated in the compliance / inventory bundle.
• Audit A retest — confirmed 1 site remained after Audit A round 1; folded into commit 3.
• Audit C (regression-invariance fault-injection) — injected deliberate count drifts into the source-side pin tests and confirmed each one fails loudly. ZERO-DEFECTS-CONFIRMED on the load-bearing pin tests. Flagged latent-drift-risk for v0.8: the doc sites themselves have no enforcing CI gate, so a future source-side bump + correct pin-test update + forgotten doc update would slip. A doc-drift-detection workflow is scoped for v0.8.

This is the testing-loop discipline working as designed: detection → fix → audit → re-fix → audit → confirm zero defects.

Risk profile

Cost

Comparison vs. 2026-05-22 release-gate campaign

v0.7.0 is now mechanically correct against its own canonical source symbols across every doc page. That is the engineering hardening that this campaign minted.

Recommendation

Merge the 3 open PRs (#1379 docs/Pages drift, #1380 CoALA citation, #1382 PostgresTestEnv fix), run the 24-hour dogfood loop, and cut the v0.7.0 tag. The engineering quality gates are met. The 3-audit ZERO-DEFECTS-CONFIRMED on Track D and the ZERO-DEFECTS-CONFIRMED on Track E demonstrate the testing-loop discipline (pm-v3.3) is operating as designed: defects surface at the realistic-deployment substrate boundary, get fixed in-campaign, and the audit trail closes loop-to-loop.

The one open issue from this campaign (#1378 codex install TOML) is non-blocking: the substrate’s manual TOML config works correctly; only the optional installer is affected. The fix is small and lands as a v0.7.x follow-up.

Provenance + audit

Apache-2.0, © 2026 AlphaOne LLC.