project_phase2_q14_q23_brainstorm_complete_2026_04_27

Date: Sunday 27 April 2026 (one long brainstorm session, ~12 question-locks)

Scope: 10 architecture decisions Q14-Q23 covering substrate + application + 5 GitHub repos + 2 new Apache primitives + 1 SKOS classifier scheme.

Key locks:

• Q14 D: Cloudflare primitives stack (Workflows + DOs + Code Mode + Workers + AI Gateway routing to Anthropic) — explicitly NOT Project Think Agents SDK preview
• Q15 D: Tier-aligned 4-tier OpenAPI 3.1.2 + Arazzo 1.0.1 + Overlay 1.1.0 SDK matching Q5 G commercial tiers
• Q16 H: Bowtie L1 + light-Inferno L2 + minimal-L3 + coverage-floor self-enforcement (responding to Rich’s elevated concern about claude-code building incomplete code with un-built tests)
• Q17 G: catala-explain + Pandoc/WeasyPrint + React + i18next/Crowdin
• Q18 G: AI-augmented document-centric workspace (Tiptap+Y.js + Monaco + AI side-pane + audit timeline + dashboard + Cmd-K)
• Q19 F: per-firm DO-SQLite FTS5 + shared Supabase EU pgvector + voyage-law-2 + Cohere Rerank
• Q20 F: Resend + Twilio + Cloudflare Workers orchestration + per-tenant preferences
• Q21 G: Stripe Identity + Thirdfort two-vendor
• Q22 F1: BoldSign-led 4-layer (BoldSign UK+EU QES via Evrotrust QTSP / Skribble Swiss / Notarize/Proof US RON / Documenso fallback) + WitnessEventProvenance + JurisdictionalSubmission Apache primitives
• Q23 F: TrueLayer + DVLA VES + HMLR Open Data + OPG LPA-Use + HMRC IHT400 PDF + Settld upstream + Companies House + RoS + zefix + AssetDiscoveryProvenance field-group

5 GitHub repos confirmed under testatetech/ org (Rich-locked TT-org-only):

• code-inherit-v2 (Apache 2.0; Tier-3 standard)
• code-inheritkit (AGPL+commercial; Tier-2 SDK + AI agents + signing adapters + asset adapters)
• code-ias (commercial; Tier-1 IAS solicitor app)
• code-inheritv2-www (Apache; INHERIT v2 website www.openinherit.org)
• code-inheritv2-test-suite (Apache; test pyramid)

2 new Apache primitives:

• #36 WitnessEventProvenance (jurisdiction-agnostic PROV-O bundle + N WitnessCredential VCs + JAdES; from T56 finding 5)
• #14 JurisdictionalSubmission (lifted from T52/T60 candidate-status; SKOS classifier with 7 sub-classes)

1 new field-group on A-29: AssetDiscoveryProvenance (4 fields + 1 SKOS classifier)

1 new SKOS classifier scheme #10: AssetDiscoveryChannelScheme (11 enumerated values)

Cumulative Phase-1 costs:

• Build: ~£259,450 (~493 engineer-days; ~2 engineers × 6 months)
• Monthly: ~£2,584-9,039/mo Phase-1 ongoing

Why this matters: Phase-2 brainstorm completes the substrate + application architecture. Phase-1 build can now begin. Outstanding decisions are operational (build cadence + commercial GTM) not architectural-substrate.

Canonical doc: docs/superpowers/specs/2026-04-27-phase-2-q14-q23-brainstorm-locks.md v1.0

arch-state: v2.78 → v2.79 (this session adds the v2.79 changelog row + primitive registry updates).

IAS UX mockup committed: docs/superpowers/specs/2026-04-27-q18-ias-ux-mockups/g-document-centric-workspace.html (~28KB; demonstrates Q18 G + Cloudflare DO + Tiptap+Y.js + Monaco + AI side-pane + audit timeline + Cmd-K palette + per-tier ConformanceCredential status badges).

Recently-applied disciplines (referenced repeatedly across all 10 locks):

• Q16 H implementability lens (c8 = 14 weight on every scorecard from Q16 onwards) — Rich’s elevated concern
• Cost transparency (Build £ + Monthly £ separately from Q19 onwards)
• No phased options (Rich rejected Q14 onwards; all locks must operate Phase-1 Day-1)
• feedback_actively_use_t_files_in_scorecard_authoring — verbatim T-file grounding
• project_law_vs_practice_split — Apache (LAW) primitives; AGPL+commercial (PRACTICE) vendor adapters
• feedback_first_vertical_slice_symmetric — 8 Phase-1 jurisdictions including Texas+Florida RON