project_audit_brainstorm_in_progress

Where we paused

Rich invoked /superpowers:brainstorming with a major audit request on 2026-04-24. Subsequently added: v1→v2 regression-completeness is a HARD constraint. Exploration pass completed; clarifying question #1 just asked; Rich paused to update Claude. This memory captures state for seamless resumption.

Rich’s original brainstorm prompt (verbatim)

“Ok we are now at a very significant moment in the development of inherit v2 and inheritkit. Before we do anything more, we need to look at the current plan and conduct a thorough analysis of the technical specification that we have so far defined. I am far more interested in technical excellence and future proofing, as well as being ready for the succession and traditions you see at https://www.openinherit.org/extensions/ I believe there are going to be more opportunities to improve the plan. Please look at it through the eyes of someone who carefully watches the latest standards and trends in legal software and ontology’s, structured data, semantic markup - someone who knows where things will be in 2029. Ask them for their suggestions on how to help us improve our plan and simultaneously reduce the lines of code inherit 2 and inheritkit repos will contain. Our library is an important starting point but web research can reveal much more current thinking and data. Look at https://github.com/testatetech/code-ll and the plans we have to allow comparison of succession and tax and faith via https://github.com/testatetech/code-mfi/tree/main/docs/superpowers/specs”

Second instruction (from Rich, mid-session): “we also need to look at https://github.com/openinherit/code-standard and ensure that inherit v2 and inheritkit included all the functionality that is in inherit v1”

Git state at pause

• Branch: main, working tree clean
• HEAD = origin/main = af66260 (brief v1.1.3 N1+N2 naming-consistency fix)
• Prior significant commits: 71f1ebb (v1.1.2 property-domain fix), dd2877c (v1.1.1), 6d3b4d3 (v1.1), c5a731c (v1.0)
• Fourth-pass review verdict: CONVERGED — next review pass should be authored-code review against inherit-core.ttl, not further brief iteration

Brainstorm-skill task ladder state

#	Subject	Status
52	Explore project context	✅ completed
53	Ask clarifying questions (scope, authority, output format)	🔄 in_progress (Q1 asked)
54	Propose 2-3 approaches with tradeoffs	pending
55	Present design sections + get approval	pending
56	Write design doc + spec self-review + user review	pending
57	Transition to writing-plans skill	pending
58	v1→v2 regression-completeness audit (INHERIT v6.6 → v2 + InheritKit)	pending (hard constraint)

Exploration findings (from task #52)

v1 (openinherit/code-standard, private)

• Tech: JSON Schema 2020-12 + OpenAPI 3.1 + TypeScript SDK
• 28,207 LOC across 98 schemas — 37 top-level + 15 common + 42 extension-specific
• 21 extensions: 16 geographic + 4 tradition + 1 EU umbrella
  • Extensions by LOC: us-estate (668), scotland (640), islamic-succession (490), latin-america (441), uk-england-wales (419), australia-nz (416), japan (405), prc-china (401), singapore-malaysia (394), africa-customary (388), jewish-succession (380), hong-kong (365), india (360), hindu-succession (345), uae (341), eu-succession (336), ireland (310), brazil (302), canada (298), israel (262), switzerland (224)
• Tradition extensions at “outline” maturity: Africa Customary / Hindu / Islamic / Jewish
• Core top-level schemas (37): acknowledgement, asset-categories, asset-collection, asset-interest, asset, attestation, bequest, catalogue, common, conformance-declaration, context, dealer-interest, dialect, document, estate, event, executor, extensions, guardian, insurance-policy, kinship, liability, lifetime-transfer, nonprobate-transfer, notification, organisation, person, pet, power-of-appointment, property, proxy-authorisation, relationship, schema, space, subscription, trust, valuation, wish
• Common schemas (15): address, ai-provenance, audit-event, completeness, cultural-disposition, error-codes, field-provenance, identifier, jurisdiction, media, money, provenance, tax-position, temporal-rule, visibility
• Language SDKs: csharp, go, java, python + TypeScript

v2 planned (brief v1.1.3 at af66260)

• OWL 2 DL + SROIQ + EL publication variant via ROBOT 1.9.7
• 4-layer standards stack: OMG Commons Classifiers (cmns-cls) + SKOS + XKOS + CIDOC E55 Type bridge (A-23 vi-6)
• 19 Core-hosted primitives (14 canonical + 5 cross-module per feedback_v2_asset_catalogue_scope)
• Seq G1 authoring order: Core → Assets → Wills E&W minimal → Delegation → Catalogue → Trusts → Probate
• Phase-1 LOC estimate: ~32K canonical / ~85K total — currently ROUGHLY EQUAL OR LARGER than v1
• Phase 1.5 hard gate: 21-extension stress-test against v6.6 (A-25)

code-ll (testatetech/code-ll, private)

• Stack: Next.js 15 + Turbopack + TypeScript
• Consumer-facing LegacyLists MLP
• No README at top level; docs present (MLP stage plan, caching strategy, founding user journey, action lists)
• Josh is building; runs on INHERIT v6.6 (not v2)

code-mfi (testatetech/code-mfi, private)

• 7,837 LOC TypeScript
• Spec corpus: mfi-design-spec.md, mfi-spec-part-a-revision.md, mfi-homepage-design-system-alignment.md, 2026-04-02-homepage-hero-redesign.md
• Also: inherit-deep-features.md + brainstorm-progress.md + build-sequence.md + tech-stack-and-patterns.md
• Key audit-relevant concern: cross-jurisdictional/tradition comparison flows (succession / tax / faith) — exact use case for SKOS mappingRelation + XKOS correspondenceTable + CIDOC alignment
• MFI spec v2.3 is known 20-days-stale vs 15-point decision doc; realignment is a deferred task (do NOT touch without Rich’s go-ahead)

openinherit.org/extensions/

• Hybrid three-dimensional: geographic (21 jurisdictions) × tradition (4 groups) × coverage (succession/tax/cultural)
• Groupings: “All / Geographic / Tradition / Hybrid” tabs
• Per-extension fields: legalTradition, extensionType, successionCoverage, taxCoverage, case-study count (0-17)
• Rich concepts modelled: Islamic fara’id / 15 asaba ranks / 22 heir classes; Jewish Numbers 27 order / 12 halachic classes / bechor primogeniture; Hindu Mitakshara coparcenary / Dayabhaga / HUF / 2005 equal daughter rights; African customary / communal land / widow protection

First-pass gap-scan observations

1. v1 has primitives v2 doesn’t yet explicitly host. Not-yet-mapped in v2 Core: pet, subscription, wish, acknowledgement, lifetime-transfer, nonprobate-transfer, power-of-appointment, proxy-authorisation, kinship, cultural-disposition, ai-provenance, completeness, field-provenance, tax-position, temporal-rule. Some fold into Catalogue / Delegation / Capacity / BiTemporal — but mapping is IMPLICIT. Regression-completeness gap.

2. Tradition extensions at outline maturity. Islamic/Jewish/Hindu/African Customary are richest to model but least-covered in v1. Phase 1.5 gate demands stress-test against all 21; vi-6 4-layer facet stack is designed for this but never validated against the richest traditions.

3. LOC arithmetic tension. v2 plan at ~32K canonical / ~85K total is roughly equal-or-larger than v1’s 28,207 despite being semantically richer. Audit’s reduce-LOC goal lives in: ontology pattern reuse (ODPs), schema.org + W3C VC re-export, deferring computation to Catala/SHACL/SPARQL rather than per-primitive machinery.

4. Cross-jurisdictional comparison not first-class in v2 spec. MFI wants succession/tax/faith comparison; v2’s facet stack enables it via SKOS+XKOS+CIDOC but not called out as a first-class concern — implicit in the facet stack.

5. Language SDK coverage. v1 ships SDKs in csharp/go/java/python + TS. v2 InheritKit assumed TS-first per earlier session memory — may need multi-SDK roadmap for enterprise/AI-vendor audiences.

6. ai-provenance schema in v1. v2 has PROV-O peer alignment but no explicit AI-provenance primitive. Gap likely intentional (AI-provenance in InheritKit, not Apache Core) — but needs explicit documentation.

Clarifying question #1 (ASKED, awaiting answer)

Scope of revisability. Multiple-choice:

• (a) Unconstrained — any locked decision reopenable; reframe-beats-re-weight applies
• (b) Structural-locks preserved, details open — 8 Shape X angles + 4-layer stack + alignment-axiom-only modularity + Seq G1 stay locked; amendments A-1..A-25 revisable with strong evidence; primitive counts / facet set / InheritKit split / LOC-reduction levers all open — RECOMMENDED
• (c) Additive-only — audit may propose NEW amendments A-26+ and NEW primitives; cannot reopen locked decisions; LOC-reduction only via consolidation within locked architecture

Rich needs to answer (a) / (b) / (c) when session resumes.

Questions queued after Q1

• Q2 — output format:

  • Single expert-audit document (one big critique)
  • Parallel-dispatch research programme (4-6 expert personas in parallel; aggregate findings)
  • Interactive per-angle brainstorm (Rich weighs in section-by-section; co-produce revised plan)

• Q3 — primary-axis weighting:

  • LOC reduction primary; future-proofing secondary
  • Future-proofing / 2029-readiness primary; LOC reduction secondary
  • Extension-coverage primary (21 extensions including outline-status traditions)
  • All three equal

• Q4 — 2029-watcher persona profile: which standards bodies / trends / critiques the expert follows. Candidate axes: W3C (SHACL/DID/VC) + OGC/ISO/JTC 1 + OASIS LegalRuleML + Akoma Ntoso + Catala + emerging AI-ontology fusion (KB-LLM) + schema.org legal + open-standards ontology-design-pattern (ODP) movements.

Resume-plan on restart

After Rich selects Q1 option:

1. Ask Q2 (output format)
2. Ask Q3 (primary-axis weighting)
3. Ask Q4 (persona profile)
4. Propose 2-3 audit-structure approaches (task #54)
5. Present design sections (task #55)
6. Write design doc to docs/superpowers/specs/2026-04-24-inherit-v2-technical-audit-design.md (task #56)
7. Rich reviews
8. Transition to writing-plans (task #57)

Restart command for Rich

After Claude update completes, Rich can resume this brainstorm by either:

Option 1 (simplest): run claude fresh in ~/testatetech/docs-strategy/ and paste:

Resume the technical audit brainstorm. See memory `project_audit_brainstorm_in_progress.md` for full state. I still need to answer Q1 (scope of revisability — a/b/c). Read the memory, remind me what Q1 said, and I'll answer.

Option 2 (most faithful): claude --resume if the CLI version supports session-resume after an update; then Rich simply answers Q1 directly — Claude re-reads memory automatically.

Related memories (for orientation)

• feedback_always_save_scorecards.md (just-created; scorecards discipline)
• feedback_introduction_module_rule.md (R-1’/AC-1/W-1’ 100% placement rule)
• feedback_reframe_beats_reweight.md (when stalled at 85-92%, reframe)
• feedback_inherit_v2_international_from_day_one.md (no E&W pigeon-holing)
• feedback_v2_asset_catalogue_scope.md (19 Core-hosted primitives)
• project_shape_x_full_lock_2026_04_23.md (8 angles locked)
• feedback_always_check_library_indexed_first.md (library before web-research)
• feedback_international_english_for_inherit_v2.md (US-style for new v2 writing)