MEMORY

• Active work log SUPERSEDED 2026-05-25 — Frozen historical record; tt-inherit-v2 TaskList is canonical per QW-4

• 📚 Archive — Older memories, superseded decisions, reference material; search here when active-memory misses

• Master plan §17 Model C SOTA fold-in LIVE 2026-05-26 (v1.17; 704b705) — 12-subsection Model C operating model in master plan body (3-axis arch / Linear hierarchy / 3-tier audit / PR conventions / spec-plan-tasks split / native Claude Tasks / multi-repo / Phase α-ε / §17.10 SOTA-8-topic synthesis from background agent / §17.12 4 PRE-WAVE-D-DISPATCH actions). Process correction: §16 forcing-function alone wasn’t enough; need substrate-currency too. Companion to §16.

• Master plan §16 Wave LP authoring contract LOCKED 2026-05-26 (da9755b + 2069810) — forcing-function discipline: master plan FIRST in §0 substrate-read + REQUIRED in §4 close-out + companion_files + §7 refs; §16.3 bash audit caught its own first-day violation (4 stale v1.15 refs in v2.0.2 → v2.0.3 self-fix). Floor-pin semantic-satisfaction ≠ explicit currency.

• Wave D LP v2.0.3 LIVE 2026-05-26 (Model-C-native + §16-compliant; 2069810) — 17 audit-finding edits + §0a pre-flight battery + §3.5 spec/plan/tasks split per Linear Issue + §3.6 native Claude Tasks discipline + §4 linear_project_id REQUIRED + §4 master-plan-update-IN-FLIGHT step per §16 contract + §8 dispatch 3 paths. Cross-refs Linear Project 5627ed2c + TT-13/14/15/16. Dispatch-ready.

• Wave C.1 closed 2026-05-26 (~1h 30min actual) — 5 deliverables LIVE: Del 1 backfill 109 specs + Del 1.5 wire + Del 2 cascade × 6 + Del 3 Linear-aligned audit × 4 + Del 4 prettier × 6 + Del 5 SKILL.md transplant. 8 honesty caveats; 4th velocity datapoint (5-7× under paper); scripted-batch waves compress more than novel-per-step. SCF: conftest verify no-op vs warn-style rego (wired-but-not-validating); YAML escape bug; SKILL.md description >120 chars per Anthropic format vs TT rule.

• TT Linear setup state 2026-05-26 (Phase α Model C migration) — Workspace mediahq2, team TestateTechnologies (TT) ID 94c6bdcd-…, Business plan confirmed, 14 labels created (repo:* × 8 + type:* × 6 with captured IDs), 3 Initiatives queued for Rich Linear-Agent creation, Steven account ID 18faabef-… captured, Code Intelligence + Project Slack channels available

• Linear-aligned 3-tier audit architecture (locked 2026-05-26) — Drops CHANGELOG.md auto-regen; replaces with Linear issues (operational) + wave-execution-records with linear_project_id: (strategic) + GitHub Releases auto-gen at v1.0+ (public). Empirical: Spike A VALIDATED-WITH-CAVEATS (SHA cascade); Spike C v2 EMPIRICAL REJECT (~35-50% audit-trail loss); SOTA = Vercel/Linear/Stripe/Anthropic all dropped CHANGELOG.md years ago; Linear-fit research = Project completion view IS the changelog equivalent. Wave C.1 Del 3 cascade-complete 2026-05-26 (4-for-4 rebase tax as pilot predicted).

• Cloudflare Pages tt-inherit-docs PROVISIONED 2026-05-26 — Wave D deploy target LIVE; project ID b876b622-e401-49aa-be99-06deaad4af80; URL tt-inherit-docs.pages.dev; source testatetech/docs-strategy main; build pip install mkdocs-material==9.7.6 && mkdocs build → site/; PYTHON_VERSION=3.12; first build will fail until Wave D Stage 10 ships mkdocs.yml (expected pre-Wave-D state)

• Wave C closed 2026-05-26 (3494467; ~2-3h actual) — 4 of 5 mandatory LIVE (EARS x7 + memorylint NOT-FIT + frontmatter v1.5 spec + Conftest+Rego); Del 6 SKIPPED; backfill + cascade Wave C.1-deferred; 10 honesty caveats; 3-point velocity convergence (A=10× / B=2-3× / C=2-3×)

• Conftest+Rego dogfood self-validation pattern (locked 2026-05-26) — When authoring schema + validator concurrently, run validator against schema-defining doc FIRST. Caught at Wave C Del 5 when frontmatter-conventions.md itself missed v1.5-REQUIRED description. Schema-validates-its-own-author paradigm = cheapest-empirical-signal pattern.

• Pilot belongs in spike phase, NOT execution (locked 2026-05-26; VALIDATED mid-Wave-C) — README-only validated tools/extensions MUST NOT land as Wave deliverables; empirical pilot belongs in spike phase BEFORE COMMIT. Caught at Wave C 2026-05-26 when memorylint NOT-FIT discovered DURING Del 4 execution. Companion to feedback-lp-t-file-substrate-audit-pre-dispatch.

• Wave B closed 2026-05-26 (5c5372b; ~3-4h actual) — 3 deliverables LIVE + §1.5 caveat closures; 7 honesty caveats; 3 substrate-correcting findings (host-repo miscitation, additional_dependencies trap, LoC under-target on conservative counting); velocity 2-3× under (convergence from Wave A’s 10×)

• LP T-file substrate audit pre-dispatch (locked 2026-05-26) — Pre-dispatch grep LP for T-file citations + verify against actual T-files; Wave B caught U2 host-repo miscitation that would have produced worse architecture. Apply pre-Wave-C + pre-Wave-D dispatch.

• Concurrent-session sham-fold-in pattern (locked 2026-05-25) — Sibling-session doc-version-bump may be FRONTMATTER + CHANGELOG only; body still reflects prior version. /review-plan before next session folds on top. Caught 2026-05-25 on audit v1.2; v1.2.1 surgical-fix landed 10 items

• Spike-suite-4 CLOSED 2026-05-25 (T1 mkdocs + T2 EARS; 2 COMMIT) — Session B Stage 10 (mkdocs+Cloudflare Pages) + Session D1 Stage 6 (EARS hook v2 TT-aware) scope LOCKED; Path E v1.2.1; 4-suite ~95% decision-locked

• TaskList replaces project_active_work.md (QW-4) — Native Claude Code TaskList canonical for parallel-session coordination 2026-05-25; atomic; cross-device synced; queryable; closure narratives go in per-batch memory files

• Recursive review-and-revise pattern (session-validated 2026-05-25) — Substantive substrate: author → /review-plan → revise → spike-validate → commit. 3-checkpoint discipline; visible §0 body checklist is the forcing function; ~5-10% cost premium prevents lock-then-revise cycles. Empirically validated v1→v6 design iteration + 6-spike suite (3 COMMIT + 5 DEFER + 1 REJECT).

• Grep workspace before authoring substrate — Closed-set lists from memory are anti-pattern; grep -rlE first; empirically validated 3× in 24h (PE-47 v1.13 + v1.15 + dispatch v1.0); generalises sister memories

• Split-personality sweep discipline — enumerate ALL repeated-fact locations BEFORE editing; grep-verify zero stale instances; Phase D” 6-commit chain caught 3 times 2026-05-24

• Stable-ID-first cross-references — 2026 SOTA (OpenSpec Requirement-ID + Spec-Kit T001); default stable-IDs over file-paths in TT spec/plan/tasks; Phase D” D-6 lock 2026-05-24

• Methodology-migration cascade-Q pairing mandatory — arch-state version bumps for methodology / infrastructure / closure-amendment ba…

• Phase D” missed-opportunities sweep (11 items; 2 new stages) — Sweep 2026-05-23 evening surfaced 11 items nearly missed at Phase D” author…

• Frontmatter policy SDD alignment (Phase D” Stage 7) — TT’s v1.3 frontmatter is ~60% SDD-aligned; v1.4 adds 8 fields (description REQUIRED + globs + …

• Adopt GitHub Spec-Kit (Path E; REINSTATED) — CURRENT LOCK 2026-05-24T21:20 BST; Path E = Spec-Kit per-repo + TT-native coordination; restored after Path D substrate-corrections collapsed 96% → ~78-82%

• Adopt Spec-Kit git extensions (standardisation directive) — Rich-directive 2026-05-24T21:50 BST during Path E smoke-test: install Spec-Kit’s git extension (auto-commit + create-new-feature + git-common + initialize-repo + 5 SKILLs) rather than maintain TT-native git automation; PE-2 revised to INCLUDE git extension; Mode A vs Mode B workflow choice deferred to Path E launch-prompt authoring

• Verify framework-extension maintenance before lock — Discipline from 2026-05-24 Path D collapse; 6-check battery (1st-party / recency / peer-dep currency / dep-footprint / README-grep / open-issues) for every REQUIRED extension in scorecards

• No SDD framework solves multi-repo for TT — Empirical 2026-05-24: BMAD + Spec-Kit + OpenSpec ALL lack native multi-repo for TT’s 7-sibling-repos; TT’s parent-workspace design + arch-state.md IS the load-bearing layer regardless of framework choice

• SUPERSEDED Path D BMAD+OpenSpec hybrid — SUPERSEDED 2026-05-24T21:20 BST by Path E pivot SCF; 96% scorecard collapsed under empirical findings; retained for audit trail only

• git mv with unstaged edits loses modifications — git mv A B while A has unstaged working-tree edits stages the rename of HEAD-content…

• BUILD-PLAN hybrid Shape A: narrative + native Tasks — BUILD-PLAN.md files retained as strategic prose + Phase/Milestone grouping…

• No forward calendar projection in plans or conversation — Calendar slots (“Q4 2026” / “M7-M8”) + forward calendar (“by next mo…

• Surface phase calendar when authoring phase implementation launch-prompts — Phase-N implementation launch-prompts need calendar-check before dispatch; “Sprint N” = Phase-1 Sprint N per master plan NOT current-calendar sprint; precedent: batch-imp-21 Sprint 1 server dispatched 6-9 months early May 2026 + killed by Rich after §0 + dispatch-mode pause

• pnpm drift 2026-05-23 — Per-repo CLAUDE.md files cite Node v22.x + pnpm 10.x

• claude -p dispatch pattern with execution directive — `( cat ; echo; echo ”---”; echo ""

• Prompts comprehensive and fully detailed — **RULE FLIPPED 2026-05-23T~17:30 BST per Rich-directive “i want prompts to be comprehensive and…

• Compound prompts × context degradation — Above 75% context, prefer single-action launch-prompt over compound multi-action

• Strategic position: operator-first 2026-05-22 — Build TT to run + profit; passive acquirer-optionality (not chasing exits)

• Steven Abols (kinetic-integrations.co.uk) — Local engineer at UK nuclear power co + spec-writing day-job expertise + AI interest

• TT core tool stack 2026-05-22 — Google Workspace + Stripe + Cloudflare as core infra (locked); ScoreApp + Facebook ads for quiz-led marketing (intent)

• Free-stack fidelity validation outcome — Spike-X-prime 2026-05-22: 4 PASS + 2 PARTIAL + 1 substrate-deferred + 0 FAIL across 7 criteria

• Stainless acquired by Anthropic 2026-05-18 — Anthropic acquired Stainless 2026-05-18 per anthropic.com/news

• Substrate search locations file is canonical — substrate-search-locations.md (in docs-strategy/…/2026-04-29-multi-phase-audit/) is sing…

• Reflog cherry-pick recovery pattern — git reset --hard drops local-only commits; recover via git reflog HEAD@{N} + git cherry-pick <hash>

• Worktree isolation under concurrent storm — when 3+ concurrent autopilot sessions touch shared arch-state §15 + ledger

• Verify after author via directory ls — §13.6 method-7 candidate; ls -la <dir>/ immediately before downstream consumption catches mid-flight fil…

• Concurrent burst race condition count 24h — project-state observation; 5 race-recoveries in 24h signals 4-5+ concurrent autopilot sessions =…

• GSD installer preserves user baseline since 1.41 — /gsd-update no longer wipes custom skills/hooks/agents

• InheritKit Anthropic Skills have real consumers — SKILL.md files at code-inheritkit/skills/{faraid,hsa,halakhic,…} ship to partner-firm Cl…

• Session-start misses MCP version skew — installed_plugins.json grep + ctx-doctor both miss the case where MCP server is running from an older pl…

• BATCH acquirer-dd-package §3.x synthesis CLOSED 20 May — A-231 arch-state v4.23→v4.24 + ledger v1.65→v1.66 task-118 DONE

• Backlog-saturation declaration 2026-05-19→20 burst — Claude-EXECUTABLE saturated; ~26h / 15 batches + 4 Q-locks + 7 meta-Qs + 112 commits + 4…

• Defer-cost arithmetic in recommendations — locked 2026-05-20 Q-044 trigger; now-cost + refactor-cost × probability = expected total

• Substrate-quality is axis-velocity conditioning — locked 2026-05-20 Rich-trigger; (a)/(b) decomposition — axis count vs axis velocity

• Claude does research inline at Q-lock time — locked 2026-05-20 Rich-directive; do research/spike work INLINE during Q-lock authoring

• Methodology hot-reload at decision points — locked 2026-05-20 Q-044 trigger; re-grep MEMORY.md at [E]/[F]/[G] checkpoints

• Launch-prompt scope-framing at rule-statement granularity — locked 2026-05-20 BATCH ρ A-226→A-228 partial-reversion trigger

• Planning sequence + dependencies, not hours; capacity flexible — Rich-directive 2026-05-20; Claude over-estimates effort + over-defers s…

• Anthropic Claude for Legal — 3 entry paths into wills — May 2026 launch excludes wills (12 plugins)

• Pre-stage substrate for other in-flight session — sibling session captures spike yield in parallel-safe location

• Spikes inline not tasked — refines feedback_do_now_over_task_list_addition for spikes specifically; ≤30 min spike → do inline

• Batch compression 5-10× lowers defer-vs-do-now threshold — empirical π/σ/χ/ρ pattern refines MQ-017 DO-NOW threshold

• ntfy curl-block — use python3 urllib fallback — context-mode blocks curl/wget by name; urllib POSTs work; silence reads as SIGKILL

• MQ-011 ★ cascade-Q file compression + test-fixture-at-lock L6 discipline LOCKED — A-193

• MQ-010 ★ EW-counter-reset-at-module-boundary + persona-rotation-discipline LOCKED — A-192

• Q-040 (Q-T4) ν.η★ Trusts.TrustDistribution MVP LOCKED 18 May — A-201

• Q-039 (Q-T3) μ.η★ Trusts.BeneficialInterest class shape MVP LOCKED — A-190

• Q-038 (Q-T2) λ.η★ Trusts.Trustee role-bearer expansion MVP LOCKED — A-189

• MQ-009 ★ Module-prefix Q-numbering + BFO-class-alignment LOCKED 18 May — A-188

• Q-037 (Q-T1) κ.η★ Trusts.Trust container class shape MVP LOCKED — A-187

• Q-036 ι.η★ Probate.AdministrationAccount class shape MVP LOCKED — A-186

• Q-035 θ.η★ Probate.AssentToBeneficiary class shape MVP LOCKED — A-185

• MQ-008 ★ Phase-5 closure scope + Catala rule body Q-tracking LOCKED 18 May — A-130

• Q-CM-1 η.η★ Transfer.AssetTransferEvent canonical primitive MVP LOCKED — A-124

• MQ-007 ★ AssetTransferEvent placement Option D LOCKED 18 May — A-124

• Q-033 ζ.η★ Probate.ProbateAdministration state-machine class MVP LOCKED — A-181

• Q-032 ε.η★ Probate.PersonalRepresentative role-bearer expansion MVP LOCKED — A-180

• Q-031 δ.η★ Probate.GrantOfRepresentation MVP LOCKED — A-179

• Q-030 γ.η★ Probate.WillRecognition MVP LOCKED — A-178

• Q-029 β.η★ Probate.Estate canonical container MVP LOCKED — A-177

• Q-009.1 ι.ε.β.α★ IRI BASE MIGRATION 17 May — A-153 → A-176

• Q-028 ν.η★ Wills.SubstitutionClause MVP LOCKED 17 May — A-175

• Q-027.2 ξ.η’.β★ Wills.Bequest PARTIAL AMEND LOCKED 17 May — A-168

• Q-026.2 ψ.η.β★ Wills.Will PARTIAL AMEND LOCKED 17 May — A-166

• MLP call Mon 11 May 2026 — Jane + Stephen enthusiastic on MFI; next-phase

• Q-026 ψ.η ★ Phase-3 ζ-Q1 Wills.Will MVP LOCKED 5 May — A-166

• Q-025 ω̄.γ ★ Phase 2 closure-check META-Q 5 May — A-165

• MLP pilot v1.7 doc-pack delivered Tuesday 5 May 2026 — combined v1.7 + Jane

• Q-024 Φ.η ★ Faith-tradition Phase-1.5+ wave-1 activation 5 May — A-164

• Q-023 ω̃.ι ★ Catala formal-verification activation 5 May — A-163

• refined-prompt v3.12 → v3.13 RELEASED — Named-persona review pass + per-Q

• Q-024.1 + Q-025.1 faith-tradition cluster re-answer under v3.15 — A-171

• Q-020 LOCKED ω.ζ★ REDUX — A-170

• refined-prompt v3.14 → v3.15 RELEASED — V6.6 substrate consultation + — A-141

• refined-prompt v3.13 → v3.14 RELEASED — Wide bundle 5 improvements

• Q-027 ξ.η’★ Phase-3 ζ-Q2 Wills.Bequest MVP LOCKED 17 May — A-168

• Q-026.1 jurisdiction-uplift amendment 2026-05-17 — A-167 + arch-state v3.61 →

• feedback E&W-creep recurrence on Q-027 first v3.13 production use 2026-05-17 — A-124

• feedback v3.12 RELEASED — 9 improvements bundled per Q-022 review

• Q-022 σ̃.η ★ Phase-1 gate sequence dependency 5 May — A-162

• Q-021 χ.η ★ cross-tier serialisation contract 5 May — A-161

• Q-019 ε.ζ ★ cross-tier execution substrate timing 5 May — A-160

• refined-prompt v3.11 RELEASED — Harvard-depth uniform default for ALL Qs

• feedback Harvard-depth uniform default for ALL Qs (locked 2026-05-05)

• ζ-Q15 λ.η ★ Wills follow-through 5 May — A-159

• refined-prompt v3.10 RELEASED — F-spike-loop step 7 SPIKE-CANDIDATE SCAN

• ζ-Q14 V5+ ★ alignment-axiom verify-tier 5 May — A-158

• ζ-Q13 Δ.η ★ registrations Phase-1 scope 4 May — A-157

• ζ-Q12 π.ζ cross-module reference 4 May — A-156

• ζ-Q11 τ.ζ SHACL invariants 4 May — A-155

• ζ-Q10 σ.ζ LinkML schema-evolution 4 May — A-154

• IRI-registry 4 May — A-153

• ζ-Q8 R.δ ODP role-modelling 4 May — A-152

• ζ-Q7 ρ.ζ vocabulary management 4 May — A-151

• ζ-Q6 LOCKED σ.η-on-demand 2026-05-04 — A-150 + arch-state v3.45 + spike T-file — A-150

• ε.κ.0 δ2 CLOSED 2026-05-04 — outcome-VALIDATED — A-162

• ε.κ.0 δ1 CLOSED 2026-05-04 — outcome-VALIDATED-WITH-NOTE

• ε.κ.0 α2 CLOSED 2026-05-04 — outcome-VALIDATED

• ε.κ.0 α1 CLOSED 2026-05-04 — outcome-VALIDATED

• ν.β suite — V1.2 wave-mode + V1.1 harvesting-spike-evidence first

• Hallucination-detection via cross-spike convergence (locked 2026-05-04)

• ν.β E3 CLOSED 2026-05-04 — outcome-VALIDATED

• ν.β E2 CLOSED 2026-05-04 — outcome-VALIDATED-WITH-NOTE

• ν.β E1 CLOSED 2026-05-04 — outcome-VALIDATED-WITH-NOTE

• ν.β D5 CLOSED 2026-05-04 — outcome-VALIDATED-WITH-NOTE

From code-inheritv2-test-suite

• Specific git add when concurrent sessions — git add -A silently scoops other sessions’ work; use git add <files>
• review-plan two-pass pattern — pair every reconciliation with verification; second pass ~5x cheaper, catches fix-introduced drift
• Commit-message must match frontmatter — claiming v1.1 in a message but leaving file at v1.0 = silent 12h drift
• Banner-sweep grep-all-occurrences — when sweeping recurring banner phrases, grep all hits; same framing recurs in subsections
• schedule has narrow legitimate use — only for genuine time-dependent triggers; “same thing but later” is a category error
• Frontmatter modernisation pattern — v3.5+/v1.6+ uses doc_type + companion_files{role/relationship/version_pinned}; legacy related: deprecated
• 5-step Phase-0 hardening cycle — Session 1 → Session 2 → CLAUDE.md → /review-plan + reconciliation → second /review-plan verification
• V2 build repos Phase-0 hardening state — 2026-04-28: 2 at 5-step (www + test-suite), 3 at 3-step (code-inherit-v2 + code-inheritkit + code-ias)

From code-ias

• code-ias Phase-0 hardening complete — Session 1 + Session 2 (2026-04-28); commercial-proprietary single-licence locked; Paul Day-1 ready
• code-ias active work log — parallel-sessions tracker per global CLAUDE.md §7

From code-inheritkit

• Active work log — current sessions + recent completions + Sprint-1 deferred questions
• Two-session hardening pattern — split legal+tooling from docs+scaffolding+BUILD-PLAN; reusable for remaining 3 v2 repos