ai-memory v0.7.0 — Track A NHI Re-run + In-Campaign Fix Batch (2026-05-18)

What this is

A focused re-run of the NHI test playbook Track A against the post-PR-#820 binary, plus the complete remediation of every defect surfaced during the re-run, all landed in v0.7.0 (no deferral). Executed per the prime directive testing-loop addendum (memory f1dca8fa-6c33-4139-b0b5-389cca45b921, supersedes 5d703efe) which mandates: discover → file → fix-in-current-release → retest → re-check → close until 100% remediated.

This campaign exercises only Track A (the NHI test playbook). Tracks B (IronClaw A2A 4-domain on Docker + Grok 4.3), C (Postgres + Apache AGE on Node B), and D (cross-node integration) remain on the strategic-tracking backlog, gated on operator pre-flight directives.

How this directory is organised

File	Audience	Purpose
`README.md`	All	Campaign index (this file)
`track-a-nhi-results.md`	Engineering	Track A raw results, all 12 phases + verdict, single file

Verdict at a glance

SHIP v0.7.0 — 85 PASS / 0 FAIL across 12 phases. Zero open defects in scope for this campaign.

10 GitHub issues closed in-campaign with retest + re-check evidence:

#826 memory_update MCP coverage (transitively via #859)
#829 verbose token budget 15570 → 9507
#830 TTL “extend” docs drift
#831 memory_promote step-skip docs drift
#837 NHI-P2 forget scratch-DB discipline
#859 MCP tools/list schema-trim restored optional property discovery
#860 memory_get_links exposes valid_from/valid_until/observed_by/attest_level
#861 memory_archive_list preserves metadata.agent_id + emits tags as JSON array
#862 tool count off-by-one help-text clarification (70 callable + 1 always-on = 71)
#865 playbook boot-directive worktree path drift (memory 081791ae refreshed)

In flight at write time: #863 (CLI ai-memory governance check-action subcommand) and #864 (Family naming clarification) — being closed by a background agent.

Code commits this campaign (branch `local/install-815-816`)

SHA	Author	Scope
`091350c`	Claude (Agent B)	`fix(#860, #861)`: get_links surfaces temporal+attest cols; archive_list preserves metadata + emits tags as JSON array
`d41b8cb`	Claude (Agent C)	`perf(#829)`: trim verbose tool docs from 15570 → 9507 cl100k tokens
`5ab3315`	Claude (Agent A)	`fix(#859)`: MCP tools/list exposes optional property schemas for NHI discovery
`e99fb0e`	Claude	`style(tests)`: is_none_or for attest_level assertion in get_links_temporal (post-#860 pedantic cleanup)
(in flight)	Claude (Agent for #863+#864)	CLI subcommand + Family-naming doc clarification

Reproducibility contract

Pinned binary — git SHA e99fb0e on branch local/install-815-816 in worktree /Users/fate/v07/v07-fixes/. The pre-fix-batch baseline was f612675; the fix batch is the 4 commits above.
Binary location — /Users/fate/.local/bin/ai-memory → /Users/fate/v07/v07-fixes/.cargo-shared-target/release/ai-memory (26 MB).
DB — /Users/fate/.claude/ai-memory.db (the operator’s live MCP DB), schema v43.
Models — nomic-ai/nomic-embed-text-v1.5 (768 d), cross-encoder/ms-marco-MiniLM-L-6-v2, gemma4:e4b. Tier autonomous.
Authoring agent — ai:claude-code@FROSTYi.local:pid-1060.
Fix agents — 4 parallel background agents (Agent A schema-trim, Agent B storage serialization, Agent C verbose trim, Agent D yesterday-issue verification) + 1 follow-up agent (#863+#864).

Hard rules during the campaign

Per the prime directive testing-loop addendum (canonical in memory f1dca8fa + CLAUDE.md §”Testing-loop discipline”):

No deferral to v0.7.1. Every issue surfaced during the test campaign was fixed in v0.7.0.
No banned framings. “non-blocking”, “P2/P3 follow-up”, “surface-level”, “vN+1 polish” are all disallowed.
Audit trail mandatory. Every GH issue body links to ai-memory evidence; ai-memory evidence links to GH issue id; commit messages reference both.
Retest discipline. Each fix was verified via the same scenario that surfaced it (retest) PLUS a fresh probe at a different angle (re-check). No “close as fixed” without both green.
Recompile + batch retest. The campaign rebuilt the release binary once after all fixes landed, then ran the batch retest sweep against the new binary via CLI + raw MCP probes.

Memory namespace convention

Item	Namespace	Title pattern
Track A phase results	`ai-memory/v0.7.0-nhi-testing`	`NHI-P{N}-{name} (v0.7.0 re-run @ f612675, 2026-05-18)`
Verdict	`ai-memory/v0.7.0-nhi-testing`	`v0.7.0 — Full-spectrum NHI verdict (ship-readiness) — re-run @ f612675 + fix batch @ e99fb0e, 2026-05-18 pm`
Strategic checkpoint	`_v070_strategic_tracking`	`iter #19` (initial re-run), `iter #20` (post-fix-batch SHIP)
Prime directive testing-loop	`global/policies`	`f1dca8fa-6c33-4139-b0b5-389cca45b921`
Sandbox sub-namespaces (evidence trails)	`ai-memory/v0.7.0-nhi-testing/{sandbox-2026-05-18, p3-kg-2026-05-18, p4-locked-2026-05-18, p11-2026-05-18}`	Per-test memories preserved

Provenance

Item	Value
Campaign date	2026-05-18 (morning re-run + afternoon fix batch)
Operator	justin@alpha-one.mobi
Authoring agent	Claude (Opus 4.7 1M context)
Authority	Autonomous execution authorized by operator 2026-05-18 (connectivity-loss recovery + fix-everything-in-v0.7.0 directive + testing-loop addendum)
Iter checkpoints	`_v070_strategic_tracking/iter #19` (memory `a1697779`), `iter #20` (this campaign verdict)
Prior campaign	`docs/v0.7.0/test-campaign-2026-05-17/`
Binary at retest	git SHA `e99fb0e` on branch `local/install-815-816`

Drafted by Claude (Opus 4.7 1M context).