ai-memory v0.9.0

ai-memory v0.7.0 — Full-spectrum AI NHI Test Campaign (2026-05-17)

What this is

A reproducible, peer-reviewable test campaign for ai-memory v0.7.0 running against:

Four substantive test tracks are exercised end-to-end:

Track Scope Where it runs
A NHI test playbook (12 phases, P0 → P11 + verdict) Node A, local sqlite
B A2A 4-domain campaign (alice / bob / charlie / dave) Node A, IronClaw + grok-4.3
C Postgres + Apache AGE backend tests Node A binary → Node B postgres
D Cross-node integration (.100 ↔ .1) Both nodes

How this directory is organised

File Audience Purpose
README.md All Campaign index (this file)
setup-reproducible-env.md Engineering Step-by-step environment reproduction recipe
track-a-nhi-results.md Engineering Track A raw results (per-phase)
track-b-a2a-results.md Engineering Track B raw results (per-domain)
track-c-postgres-age-results.md Engineering Track C raw results (per-test)
track-d-cross-node-results.md Engineering Track D raw results (per-scenario)
audience-non-technical.md End users Plain-English summary of what works + what doesn’t
audience-c-level.md Executives Business framing: risk, cost, time-to-ship, ROI
audience-engineering.md Engineers Detailed findings + reproduction + recommendations
final-verdict.md All Ship / fix-first / hold verdict with evidence pointers

Reproducibility contract

Every test result in this campaign meets these criteria:

  1. Pinned binary — exact git SHA + build profile recorded per phase.
  2. Pinned dependencies — Cargo.lock + Cargo.toml SHAs committed with the test artifacts.
  3. Pinned external services — PostgreSQL version + AGE extension version + ollama / xAI model strings recorded.
  4. Pinned data — fixture SHAs or generation seeds recorded; nothing depends on random untracked state.
  5. Pinned environment — env vars, hostnames, IPs, OS versions captured at test time.
  6. Re-runnable shell recipes — every test produces a repro.sh that re-runs the same scenario from a fresh shell.
  7. Per-result commits — each finalized phase/track lands as its own git commit so the GitHub log shows the campaign progression.

How to peer-review

A reviewer should be able to:

  1. Clone alphaonedev/ai-memory-mcp at the recorded SHA.
  2. Read setup-reproducible-env.md end-to-end and have a working Node A + Node B.
  3. Run any repro.sh and observe matching results.
  4. Read the audience-appropriate writeup for their role.
  5. Disagree with any finding and follow the evidence pointer to the raw artifact.

Hard rules during the campaign

Inherited from ai-memory/v0.7.0-nhi-testing playbook hard rules and CLAUDE.md:

Memory namespace convention

Track Namespace Title pattern
A ai-memory/v0.7.0-nhi-testing NHI-P{N}-{name}-2026-05-17
B _v070_grand_slam/a2a_campaign/wave5-2026-05-17 A2A-W5-{domain}-{scenario}-2026-05-17
C _v070_grand_slam/postgres_age_campaign/2026-05-17 PGAGE-{phase}-2026-05-17
D _v070_grand_slam/cross_node_campaign/2026-05-17 XNODE-{scenario}-2026-05-17
Verdict All four above + ai-memory/v0.7.0-nhi-testing v0.7.0 — Full-spectrum verdict (2026-05-17)

Provenance

Item Value
Campaign start 2026-05-17
Operator justin@alpha-one.mobi
Authoring agent Claude (Opus 4.7 1M context)
Authority Autonomous execution authorized by operator while traveling
Source handoff .local-runs/handoff-prompt-next-session-2026-05-17.md (in this repo)
Global track-and-fix rule memory 71ecce23-611b-4984-962d-d37c4309f261 (namespace global/policies)
PR under test #820 (local/install-815-816release/v0.7.0) with 4 fix commits landed 2026-05-17 (#822, #823, #824, #825)

🤖 Drafted by Claude (Opus 4.7 1M context).