TT Claude Memory

project_wave_c_closed_2026_05_26

Properties1
descriptionWave C build-process standardisation closed at 3494467 — ~2-3h actual vs ~5-9h paper / ~3-5h projected (extends Wave A=10× / Wave B=2-3× convergence to 3 datapoints); 4 of 5 mandatory LIVE (Del 6 SKIPPED; Del 2 backfill + Del 5 cascade Wave C.1-deferred); 3 NEW empirical lessons load-bearing for Phase-1.5+ planning

5 min read

Wave C build-process standardisation (spec-authoring layer) closed at commit 3494467 on docs-strategy/origin/main (2026-05-26T~07:55 BST). Coordinator session captured execution outcomes + 3 NEW empirical lessons; this memory is the per-batch closure narrative complementing the execution record at docs/superpowers/specs/2026-05-25-wave-c-execution-record.md v1.0.

Deliverables:

  • Del 1 EARS hook v2 — LIVE × 7 in-scope repos. docs-strategy 279e803 substantive (script + tests + 3 fixtures + .pre-commit-config.yaml) + 6 cascade commits (dcf767e/c4d7404/0438660/2f7f856/273dc7d/7a8c31f). T2 baseline reproduces per-file rates exactly on spec.md files; tasks.md FN count delta (5 → 32) explained by file growth between spike and execution (algorithm correct).
  • Del 4 memorylint pilot — NOT-FIT for TT at 32abe08. Pilot findings doc at docs/superpowers/specs/2026-05-26-memorylint-pilot-findings.md v1.0. Reason: memorylint’s single-AGENTS.md + single-constitution.md model conflicts with TT’s distributed-multi-file constitution (global ~/.claude/CLAUDE.md + parent ~/testatetech/CLAUDE.md + per-repo CLAUDE.md + 15+ arch-state.md files) + PE-13 lock on docs-strategy/.specify/memory/constitution.md (pointer-only, INTENTIONALLY EMPTY). Extension uninstalled (S2 rollback validated); community-allowed catalog override retained for future re-evaluation.
  • Del 2 frontmatter v1.4 → v1.5 SPEC bump — LIVE at 8b7e7ff. §1 description promoted to REQUIRED + §3.8 8 SDD-aligned fields + framework_artifact_type mapping table + §3.9 Structured MADR for spike T-files + §3.10 TAP 14 for smoke-test artefacts. Preserves §3.5 pin-kind / §3.6 forward-traceability / §3.7 placeholder cascade-Q convention verbatim. 603-doc backfill DEFERRED to Wave C.1 per LP §5 fork (variable-tail; spec + engine LIVE so no urgency).
  • Del 5 Conftest+Rego validation engine — LIVE in docs-strategy at c930954 (4 Rego policies / 158 LoC + wrapper / 90 LoC = 248 LoC total). Honest empirical count vs T-spike-s3 NARROW 53 LoC for one policy only; reported per Wave B Lesson 3 conservative-counting discipline. Cross-repo cascade to 6 other repos DEFERRED to Wave C.1 (1-week parallel-run vs Python hook + cascade pattern mirrors Wave A).
  • Del 6 MADR-5 — SKIPPED per LP default (Rich-decision gate; no pre-approval at dispatch).
  • Close-out — execution record v1.0 + Path E audit v1.5 → v1.6 at 3494467. Coordinator handoff doc v1.0 at ff75adf for post-context-exhaustion pickup.

Velocity calibration: Wave A=10× under, Wave B=2-3× under, Wave C=2-3× under (matched B’s ratio). 3-point convergence confirms the trend isn’t noise. Phase-1.5+ master plan §8 footnote’s “Wave B convergence ratio as baseline” projection is empirically validated. Wave D projection: ~1-2h paper likely ~1h actual (continues 2-3× under per current trend).

10 substrate-correcting findings (per execution record §3) — load-bearing for Wave D + Phase-1.5+ Wave E planning:

  1. §3.1 EARS hook tasks.md FN count discrepancy — T2 spike: 5 FN; Wave C: 32 FN. File grew between spike + execution; per-file 100% FN rate matches T2 (algorithm correct; absolute count grew with content).
  2. §3.2 memorylint NOT-FIT for TT — empirically refines S2 spike’s PARTIAL classification to FAIL for TT-fit purposes. Both findings correct in their scope.
  3. §3.3 NEW workspace-locked memory [[feedback-pilot-belongs-in-spike-phase-not-execution]] — sibling session locked discipline mid-flight; Wave C empirically validated it. README-only-validated tools/extensions MUST NOT land as wave deliverables.
  4. §3.4 Rego LoC honest reporting — 158 LoC across 4 policies (vs T-spike-s3 NARROW 53 LoC for one policy only); apples-to-apples Python comparison needs split first. Honest disclosure per Wave B Lesson 3 pattern.
  5. §3.5 CI changelog regen rebase tax — Wave A=5, Wave B=6-8, Wave C=5+ rebase cycles confirmed across 3 waves. Wave A.1 catch-up (changelog.yml github-actions[bot] guard) trigger met (≥10 per-repo commits in <2h). Phase-1.5+ Wave E item 1.
  6. §3.6 Wave A caveats #1+#2 → option (c) deferral to Wave A.1 catch-up. Low-severity; both orthogonal to Wave C scope.
  7. §3.7 Wave C velocity convergence — 2-3× under projection. Under-runs because Del 4 NOT-FIT collapsed scope + Del 2 backfill + Del 5 cascade shifted to Wave C.1.
  8. §3.8 Spec inventory delta — 603 specs at Wave C dispatch (LP expected 601; +2 since LP authoring). Phase-1.5+ Wave E backfill scope is 603-doc baseline.
  9. §3.9 Dogfood finding — frontmatter-conventions.md itself missing v1.5-REQUIRED description. Conftest+Rego caught it on first smoke-test; fixed in-commit at c930954. NEW load-bearing pattern captured at feedback-conftest-rego-dogfood-self-validation-pattern.
  10. §3.10 Two Wave C.1 deferrals explicit — backfill (~3-5h Claude) + cross-repo cascade (~30-60 min) per LP §5 fork; both LP-sanctioned.

Wave C.1 scope (now Phase-1.5+ Wave E fold-in per coordinator handoff §2):

  • Original Wave A.1 items (changelog regen guard + prettier patterns + S9 SKILL.md transplant) — ~3-5h paper
  • NEW Del 2 frontmatter v1.5 backfill — 603 specs; phased option (c) ~136 depth=1 first — ~2-3h paper
  • NEW Del 5 Conftest+Rego cascade × 6 repos + 1-week parallel-run vs Python — ~3-4h paper
  • Re-framed Wave E total: ~8-12h paper / ~3-5h projected (was ~3-5h / ~1-2h)

3 NEW empirical lessons captured as durable memories (this coordinator-session pass):

Wave D readiness (per execution record §4):

  • frontmatter v1.5 spec LIVE → mkdocs nav can leverage description field (once backfilled in Wave C.1 / Wave E)
  • 601→603-doc inventory updated → T1 spike’s 230-doc extrapolation needs 603-doc actual at Wave D close
  • Conftest+Rego engine LIVE in docs-strategy → Wave H OPA pin-drift migration (Phase-1.5+) can leverage same patterns + cascade is one of the Wave E items
  • Velocity projection: ~1-2h paper / ~1h actual per convergence trend

Wave C LP frozen at v1.9 + execution record frozen at v1.0 per Wave A F-15 precedent. Substrate-correcting findings live in record §3 (forward-traceable per frontmatter-conventions.md v1.5 §3.6). See project-wave-b-closed-2026-05-26 for Wave B precedent + feedback-lp-t-file-substrate-audit-pre-dispatch for the audit-before-dispatch lesson + feedback-pilot-belongs-in-spike-phase-not-execution for the empirical-pilot-pre-COMMIT discipline.

Graph View

Backlinks