Locked by Rich, Wednesday 22 April 2026 after reviewing Q2 + Q5 scorecards + jurisdictional-scope scorecard + INHERIT-Core scorecard.

The lock-ins

AreaDecision
Architecture paradigmⅢ (Full Option C+) with v6.6 inheritance — JSON Schema + JSON-LD + SPARQL 1.1 (Oxigraph) + SHACL + OWL 2 DL + Catala + InheritKit
Build pathD-ii sequenced single release — layer-by-layer build, single v2.0 release at end
Phase-1 Catala jurisdictionsOption D — E&W + Scotland + Switzerland (common + mixed + civil)
Phase-1 faith-tradition Catalaα_max — all three (Islamic + Jewish + Hindu) Phase-1 + ε plug-in marketplace committed Phase-3+
Q5 admin-surfaceε Day-1 + δ Phase-2+ — single codebase + tenant-scoped navigation + content-type-per-row licensing + consortium handover committed
Phase-1 timeline30–48 months (realistic under α_max + D; accepted by Rich)
Module decomposition7 published modules: @inherit/core + Wills + Trusts + Probate + Assets + Catalogue + Delegation
Catalogue position6th top-level module (Rich’s rationale: contains wishlist items testator does NOT yet own)
INHERIT-CoreOption A — explicit 7th published module (@inherit/core). Scorecard: 53/65 = 81.5%; B distributed = 51.5%; C hybrid = 73.5%
LLM postureScenario 2 (vendor API via Cloudflare Gateway, Haiku default, £1,500/mo ceiling) — see feedback_phase1_llm_scenario2_lock.md
v6.6 relationshipClean-break at runtime; inheritance at data-model layer — see feedback_v2_clean_break_from_v6_6.md

Reification-risk mitigations (α_max)

MANDATORY and ENFORCEABLE (not decorative). Per Q2 memo v1.1 + Rich’s 2026-04-22 acceptance:

  1. Per-rule provenance metadata {school, generation, endorsing-body, endorsement-date, tier, geographic-scope, version} — modules without complete provenance metadata do not compile
  2. Non-authoritative-compilation clause on every faith-tradition module (“This Catala represents {tradition} as understood by {named panel} on {date}; not the tradition simpliciter”)
  3. Bilingual domain-expert translation-audit leg (separate from endorsement)
  4. Women’s-rights + non-dominant-school counter-expert review (addresses Bhattacharyya-Panda women’s-rights documented casualty)
  5. Runtime-feedback-to-panel loop (real-case outcomes re-ratified periodically)
  6. Deprecation affordance — endorsing body retains unilateral version-bump + withdrawal rights
  7. Candidate-status publication (inherited from v6.6 posture) — initial maturity: candidate / legalReviewStatus: self_verified

What runs in parallel under F7.8 = 3 months per tradition

Under α_max + D, six Phase-1 Catala modules plus InheritKit plus 4 other D-ii layers (schema port, SPARQL, SHACL, OWL 2 DL). F7.8’s 3-month-per-tradition lock only works with concurrent scholar-panel engagements (Slater Heelis Islamic + Manchester Beth Din Jewish + HCUK-equivalent Hindu all running simultaneously; E&W succession + Scottish SLC + Swiss cantonal partners all running concurrently). This is ~6 live partner relationships at once for Rich personally. feedback_ai_raises_delegation_bar.md is load-bearing — AI-assisted drafting is the only route to tractability at this bandwidth.

Companion memories

  • project_partner_model_three_shapes.md — Shape A (legal-advisory triad: succession + tax + faith per jurisdiction) / Shape B (Catalogue competitive-marketplace) / Shape C (Assets platform-integration, TBD)
  • project_z_research_programme_24_files.md — 24-Z research dispatch plan (6 waves)
  • project_scotland_as_second_jurisdiction.mdnow superseded by this memory (Scotland is Phase-1, not Phase-2)

All decisions + preparatory commits landed (2026-04-22T21:15)

  • Repo-naming convention — locked Option (ii) consistent code-{brand}
  • A4.40 three-shape partner-model claim wording — locked v1.6
  • Z1-Z5 disposition — hybrid (c) 19 new + 1 staleness-review
  • GitHub repos CREATED 2026-04-22T07:20 UTCtestatetech/code-inherit-v2 (PRIVATE) + testatetech/code-inheritkit (PRIVATE). Both match stealth discipline per project_openinherit_release_discipline. Not yet cloned locally; cloning is a forward-looking task (Year-1 build start).
  • architecture-proposal v1.6 committed + pushed (post-independent-review fixes)
  • handoff v1.2 committed + pushed (autonomous-mode operating rules + Sourcemeta-not-a-concern note)
  • 24 Z-notes stubs exist at ~/off-github/library/projects/inherit/
  • 24 research-request YAML blocks in research-requests.md v1.1
  • Memories saved + MEMORY.md index updated

Pending the fresh session dispatching Wave 1

Rich will spawn a fresh Claude Code session with the autonomous-mode kickoff prompt (see §“Kickoff prompt” in project_session_2026_04_22_architecture_lock.md). That session executes Waves 1-6 autonomously per handoff v1.2 §AUTONOMOUS MODE operating rules.

What this supersedes

  • project_scotland_as_second_jurisdiction.md — Scotland was Phase-2 there; now Phase-1
  • docs/superpowers/specs/2026-04-18-inherit-v2-architecture-options-scorecard.md v1.9 status best-current-proposal-pending-product-requirements — now decided
  • Option G+ recommendation superseded by Ⅲ per scorecard re-read under v6.6 inheritance (Ⅲ scored 57/62.5 = 91% vs Ⅱ at 49/62.5 = 79% post-inheritance)

Cascades to commit when v1.5 writes

  • architecture-proposal v1.4 → v1.5 with Module-decomposition section + Option-D jurisdictional scope + α_max mitigation enforcement + 30-48 month timeline
  • Q2 memo → v1.2 (recommendation overridden from β→ε to α_max + ε-committed)
  • Q5 memo → unchanged (v1.2 already recommended ε+δ, now locked)
  • Roadmap spec → v1.6 with MFI launch M24-30 (shifted from M11)
  • Mega-spec v1.0 → v2.0 — full rewrite absorbing all cascades