project_session_2026_04_17_v4_architecture

Summary

Long research session executing the v3.5 foundation-architecture handoff prompt. Thesis (JSON Schema + TemporalRule + policy-as-code option B + selective Layer 2/3/4/5 additions + agent-readiness layer) ACCEPTED — no falsification criterion met; 8 Phase 2 composite renderings all held together; internationalisation stress-test passed under Islamic faraid synthetic SCR; multi-entity DADE integration clean with 3 new DADE-SCRs surfaced.

35 parallel subagents for Phase 1 (one per research subject + 1 DADE PDF reader), 8 parallel subagents for Phase 2 renderings, 4 parallel subagents for Phase 3 positioning + Phase 4 v4.0.0 architecture spec + Deliverable 7 licence memo + Deliverable 8 AI-vendor memo.

Outputs

Artefact	Version	PDF pages	Location
v4.0.0 Architecture + Integration Roadmap	v1.0	50	docs/superpowers/specs/2026-04-17-inherit-v4-architecture.md
Institutional Positioning memo	v1.0	27	docs/superpowers/specs/foundation-arch-research/institutional-positioning.md
Licence Strategy memo (Deliverable 7)	v1.0	25	docs/superpowers/specs/foundation-arch-research/licence-strategy-memo.md
AI-Vendor Commercial Strategy memo (Deliverable 8)	v1.0	36	docs/superpowers/specs/foundation-arch-research/ai-vendor-strategy-memo.md
DADE Report Ground-Truth Extract	v1.0	15	docs/superpowers/specs/foundation-arch-research/dade-report-ground-truth.md
34 Phase 1 research files	v1.0	—	docs/superpowers/specs/foundation-arch-research/01-...36-*.md
8 Phase 2 SCR renderings	v1.0	—	docs/superpowers/specs/foundation-arch-research/renderings/*.md

PDFs in Downloads: inherit-v4-architecture.pdf, inherit-v4-institutional-positioning.pdf, inherit-v4-licence-strategy.pdf, inherit-v4-ai-vendor-strategy.pdf, dade-report-ground-truth.pdf.

Headline decisions (Rich sign-off)

1. Foundation: JSON Schema + TemporalRule + Rego policies + JSON-LD 1.1 (option 12b, sole wire serialisation). Policy-as-code Option B reconfirmed. Paradigm not changed; this is selective enhancement to the endorsed foundation.

2. Foundation licence: RETAIN Apache 2.0. Decisive matrix score (34 Apache vs 25 MIT vs 14 AGPL). Bruno-Lowagie-synthesised iText playbook applied unchanged at Tier 2 (InheritKit AGPL+dual commercial). CLA + trademark registration + open-core-boundary commitment before first external commit.

3. DADE Joint Implementation Guide (JIG) v0.1 co-authored with OIDF DADE CG — pin stable anchors (Legacy Manager entity, State Change Event, on-behalf-of RFC 8693, UMA 2.0, VC 2.0 envelope, civil-registration VC reference pattern); forward-stub deferred items (DASDT Final, Sovrin, full posthumous-AI field packs).

4. AI-vendor commercial model: Route 1 (embedded licence £750k-£1.5M/yr) + Route 2 (hosted metered £0.02→£0.008/call) primary. Route 3 (training data) DEFERRED — threat-model review flagged Route 3 as materially weakening T2 (domestic abuse) + T4 (inheritance without intent) mitigations; rework to public-conformance-content-only + synthetic-corpus approach preserves £15-30M breakout upside in v4.1+.

5. Phase 1 verdicts summary:

   • ADOPT v4.0.0: 8/8 Lau Pa Sat Recs (1,2,3 language, 4 partial, 5 Py+Go+Java, 6 spec, 8, 10); 3/4 Layer-2 (Akoma Ntoso, JSON-LD 12b, FHIR); 2/3 Layer-3 (OpenAPI depth, MCP); most Layer-5 as interop seams (RFC 8693, UMA, Legacy Manager, posthumous-AI entity shape)
   • DEFER v4.1+: LegalRuleML (tooling thin), Arazzo (young), DASDT (Implementer’s Draft), Sovrin (SSI wrong shape), C# SDK, full OASIS/ISO engagement (2027-2030), Rec 6 implementation
   • 0 REJECTIONS.

New SCRs surfaced (for Framework v1.4 absorption)

• SUCC-SCR-09, SUCC-SCR-10 (Legacy Manager + Digital Estate Service) — HIGH
• POSTHUMOUS-AI-SCR-01..05 (from subagent 26)
• DT-SCR-01..06 (threat-model structural gaps: cooling-off, capacity-challenge, shielded-address, institution-taxonomy, default-override, agent-provenance) — HIGH
• DADE-SCR-01..03 (from Phase 2 DADE rendering: eContestEvent, contentHash on evidence, eAccessPingChannel)
• FW-FAITH-01 HijriDate type — MEDIUM v4.1

Critical cross-cutting findings

• DADE PDF ground-truth extract flagged 4 divergences from memory summary — most important: impersonation NOT categorically deprecated (§5.4 transitional concession); must support both delegation_mode values in v4.0.0.
• @sourcemeta/jsonschema diff doesn’t exist yet — GSoC 2026 project chartered; co-fund opportunity with Sourcemeta as Q3 2026 partnership deliverable.
• LegalRuleML home is Bologna/CIRSFID-AlmaAI not Turin — institutional-positioning memo corrects prompt’s error.
• JSON-LD 12b option (sole) recommended over 12a (concentric) — Grieve’s “JSON only” warning was about FHIR JSON+XML dual-serialisation; JSON-LD 12b is single-serialisation with semantic affordances, not a second format.
• Current JAIRF score ~58 (Level 1 Foundational); target Level 3 (75-90) for v4.0.0. Effort 5-7 dev-days. Erik Wilde engagement opportunity.
• Route 3 AI-vendor push-back rewritten to preserve upside — public conformance content + synthetic corpus, not partner-validated real records.

Standards-body pathway sequenced

• 2026 H2: OASIS Associate (free) + LegalDocumentML TC observer
• 2027 H1-H2: OASIS Contributor ($5.4k)\to Sponsor($25.6k) + TC charter filed
• 2028 Q1: OASIS TC formation; v4.x contributed
• 2030 Q4: Target Approved OASIS Standard
• 2031+: Optional ISO/IEC JTC 1 SC 32 PAS submission; BSI as national body

Total governance cost through 2030: ~$100k fees + Rich 25-80 hrs/quarter.

Top 5 priority institutional engagements for 2026

1. STEP UK & Ireland Technology Committee + Digital Assets SIG — paper target 2026 Q4 with member co-author
2. Uniform Law Commission Study Committee — machine-readable fiduciary directives extending RUFADAA (48 US states)
3. Bologna CIRSFID-AI (Monica Palmirani) — Akoma Ntoso + LegalRuleML alignment preprint; JURIX 2027 paper
4. OpenAPI Initiative / Jentic (Erik Wilde) — JAIRF scorecard publication; reference-example proposal
5. W3C Community Group — zero-cost 2026 H2 launch for JSON-LD + VC adjacency legitimisation

Top 5 actions for Rich in next 90 days (AI-vendor thread)

1. Approve £50-200k infrastructure investment pre-v4.0.0 for hosted MCP/API stack
2. Make Anthropic first-contact personally Q2 2026 — MCP-native pitch + JAIRF Level 3 scorecard
3. Confirm Route 1 £600k floor + Route 2 £0.02→£0.008/call pricing
4. Sign off Route 3 defer — no training-corpus tier in v4.0.0
5. Open Sourcemeta JV Arrangement A conversation with Juan within 30 days — 50/50 validator split + auto-MCP hosting

Interaction pattern worth preserving

Handoff prompt explicitly said: no “shall I proceed” questions; one sanity-check only if 3+ additions killed; dispatch Phase 1 subagents aggressively in parallel. 35 parallel subagents dispatched in one message; all returned clean. 8 Phase 2 renderings dispatched in one message; all returned clean. 4 parallel deliverable subagents (v4 spec, positioning, licence, AI-vendor) dispatched in one message; all returned clean. Main-session context stayed manageable because subagents wrote to files and returned short summaries. Total session wall time ~2 hours. Pattern: aggressive parallelism + file-based persistence + short summary returns works at this scope.

What’s next

1. Rich reviews the 5 primary PDFs away from screen
2. Decide on specific sign-offs flagged in v4.0.0 spec §17 + licence memo §7 + AI-vendor memo §15
3. Queue: dispatch OPA/Cedar spike per PaC spec §11; start JSON profile draft; Juan conversation on doc-generator + auto-MCP + JV Arrangement A; Erik Wilde JAIRF scorecard engagement
4. Framework v1.4 absorb: SUCC-SCR-09/10, DT-SCR-01..06, DADE-SCR-01..03, POSTHUMOUS-AI-SCR-01..05, FW-FAITH-01, FW-V4-NN open questions