ALL MEMOS Download .docx

QA Backlog — 2026-04-20

HERALD · 2026-04-20 20:30 UTC

Honest state: every deliverable shipped this session is single-author. The peer-QA spec exists; enforcement does not. This file is the backlog + plan to fix.

---

Deliverables shipped today — pending peer QA

| # | Deliverable | Author | QA reviewer (per rule) | QA Status |

|---|---|---|---|---|

| 1 | backend/src/conversation_store.py (DynamoDB persistence) | FORGE-equivalent (me) | SCOUT | ❌ pending |

| 2 | handler.py persistence wiring + hydration + landing page + crisis path | FORGE-eq | SCOUT (code) / GUIDE (UX) | ❌ pending |

| 3 | prompts/system_v1.md emotional-participation + memory revision | SCOUT+FORGE-eq | ORACLE (recency) / VAULT (voice-consistency) | ❌ pending |

| 4 | _INNERVERSE_CHAT_HTML (chat UI) | FORGE-eq | GUIDE | ❌ pending |

| 5 | infra/lib/mirror-stack.ts (DynamoDB + marketplace IAM + CloudFront) | FORGE-eq | SCOUT | ❌ pending |

| 6 | agents/herald.md (HERALD spec, Owner role, peer-QA rule) | HERALD | DARWIN | ❌ pending |

| 7 | plans/ROUGH-ASKS-LOG-2026-04-20.md (39 entries captured) | HERALD | DARWIN | ❌ pending |

| 8 | plans/AGENT-AUDIT-2026-04-20.md + AGENT-INCENTIVE-PLAN-2026-04-20.md | HERALD | DARWIN | ❌ pending |

| 9 | plans/WORK-DISTRIBUTION-innerverse-covering-asses-20260420.md | HERALD | DARWIN | ❌ pending |

| 10 | plans/WORK-DISTRIBUTION-identity-and-signup-funnel-20260420.md | HERALD | DARWIN | ❌ pending |

| 11 | knowledge/memory/hot/feedback/feedback_checkpoint_email_20260420.md | VAULT-eq | HERALD | ❌ pending (but HERALD is author; DARWIN instead) |

| 12 | knowledge/memory/hot/feedback/feedback_inbox_read_body_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 13 | knowledge/memory/hot/feedback/feedback_features_first_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 14 | knowledge/memory/hot/feedback/feedback_innerverse_voice_and_disclaimers_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 15 | knowledge/memory/hot/feedback/feedback_gentle_signup_funnel_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 16 | knowledge/memory/hot/feedback/feedback_modes_as_unlockable_skus_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 17 | knowledge/memory/hot/feedback/feedback_honesty_from_day_one_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 18 | knowledge/memory/hot/feedback/feedback_godly_ego_flywheel_20260420.md | VAULT-eq | HERALD | ❌ pending |

| 19 | knowledge/memory/warm/strategic/vision_multiverse_20260420.md | HERALD | DARWIN | ❌ pending |

| 20 | knowledge/memory/warm/strategic/concept_scoring_20260420.md | HERALD | DARWIN | ❌ pending |

| 21 | knowledge/memory/warm/strategic/idea_15_apps_competitive_20260420.md | HERALD | DARWIN | ❌ pending |

| 22 | knowledge/memory/warm/strategic/strategy_godly_flywheel_20260420.md | HERALD | DARWIN | ❌ pending |

| 23 | knowledge/memory/warm/strategic/system_mechanism_explained_20260420.md | HERALD | DARWIN | ❌ pending |

| 24 | knowledge/memory/warm/case-studies/case_crisis_save_20260420_1952utc.md | HERALD | DARWIN | ❌ pending |

| 25 | server/routes/cost.py + expenses.py + tasks.py | FORGE-eq | SCOUT | ❌ pending |

| 26 | clients/web/command.html dashboard | FORGE-eq | GUIDE | ❌ pending |

| 27 | agents/herald.md — the agent spec itself | HERALD | DARWIN | ❌ pending |

| 28 | SCOUT's legal-gating research research_innerverse_legal_gating_20260420.md | SCOUT (actually spawned) | ORACLE | ❌ pending |

Count: 28 deliverables waiting for QA. Plus ~6 more I haven't listed.

---

What WAS QA'd (defense-in-depth layers that did fire)

1. Code via pytest — 582 drift-lock tests run on every deploy. Caught the OPTIONS-CORS bug, the __init__.py issues, the env-var audit. That's automated QA and it worked.

2. CDK synth — generates the CloudFormation template; caught missing fields + syntax errors.

3. Smoke tests post-deploy — GET /health / POST /invoke / DynamoDB query. Caught the pydantic wheel issue + SK collision + marketplace IAM gap.

4. Real user test (Harnoor, as himself) — the godly-flywheel conversation. Caught the cid-persistence bug + the overpromising memory prompt.

5. Real user test (Harnoor-as-adversary) — crisis simulation. Caught the regex guardrail gap.

So code-level QA is strong. Documentation + strategic docs + memory files are what's un-QA'd.

---

The fix — how to enforce going forward

Option A — Synchronous QA on every ship (rigorous)

Every commit waits until a peer-agent reviewer spawns and returns PASS. Adds ~5-20 min per deliverable. Slows the session dramatically.

Cost: +300-500% cycle time.

Benefit: absolute confidence in every shipped artifact.

Verdict: too slow for "features > security / iterate fast" directive (R0002).

Option B — Async QA with rolling backlog (recommended)

1. Every deliverable ships with a QA ticket in F:/TITAN/qa-queue/<deliverable-slug>-<date>.md — stating what needs review, who the reviewer is, deadline.

2. HERALD runs a daily QA sweep — spawns each agent in turn to clear its QA queue. If Friday's sweep finds >10 items overdue, HERALD escalates.

3. If QA returns FAIL or CONCERNS, HERALD creates a follow-up commit with the fix and re-routes for re-QA.

4. Scorecard penalty: author loses 3 points per QA-FAIL; reviewer loses 5 points per weak-QA (letting a failing artifact pass).

Cost: +30-60 min/day HERALD + ~10 min/agent/day.

Benefit: reasonable rigor without blocking the ship path.

Verdict: correct for our cadence.

Option C — Tiered QA (practical hybrid) — PROPOSED

Cost: moderate. P0 slowdown is small (code reviewers are fast). P1/P2 amortized across week.

Benefit: high rigor where it matters (safety + correctness), reasonable everywhere else.

Verdict: shipping this option as today's policy.

---

HERALD's first QA sweep (starting now, pre-committed-against-option-C)

Immediate actions HERALD takes TONIGHT:

1. Spawn DARWIN with the 15 strategic-doc deliverables (numbered above) — one batch prompt. DARWIN reads all, writes ONE consolidated QA report at agent-memory/herald/qa-sweep-1-2026-04-20.md with PASS/CONCERNS/FAIL per item.

2. Spawn SCOUT (second instance) with the 5 code deliverables. Writes agent-memory/forge/qa-sweep-1-2026-04-20.md.

3. Spawn ORACLE with the SCOUT legal-research deliverable. Writes agent-memory/scout/qa-20260420-legal-gating.md.

4. HERALD consolidates findings into logs/herald-qa-report-2026-04-20.md + Harnoor checkpoint email.

Timing: Can start tonight, complete by tomorrow morning. None of this blocks the current chat path — it's retrospective peer review of what already shipped.

---

Proposal for Harnoor (decision ask)

1. Approve Option C (tiered QA: sync for P0 code/safety, async for P1 docs, weekly batched for P2 memory)?

2. Approve spending ~4-6h of agent runtime (SCOUT + DARWIN + ORACLE × tonight's sweep)? Budget impact: ~$0.50-1.50 (inside $100/mo cap).

3. Approve HERALD auto-spawning peer-QA on future commits without asking per-commit (enforcement via per-commit hook + orchestrator trigger)?

If silent → HERALD executes Option C tonight + codifies the hook tomorrow.