TT Claude Memory

project_zeta_q3_eps_iota_S1_2026_05_02

Properties1
descriptionFirst of 10 derisking spikes for ε.ι lock decision (per plan `2026-05-02-zeta-q3-eps-iota-derisking-spikes.md` Task 1). v6.6 confirmed as high-quality 80%-shape SEED — 21 dedicated jurisdiction-or-tradition extensions + 32 jurisdictions enumerated + 11-jurisdiction multi-decade tax-history + 5 faith pillars wired (Christian / Muslim / Jewish / Hindu / customary). Sample mean depth 4.6/5 across 10 sampled extensions. Kill condition NOT-MET on all 3 legs (jurisdictions 7×; per-jurisdiction file density well-above; freshness 18× — oldest content 4 weeks vs 18-month threshold). Cost story revises DOWNWARD (depth-4.6 SEED → 10-15% LLM-REFINE per jurisdiction not 20%). Universal-production-pipeline-sequence reinforced per Rich-directive — UTILISE the SEED material → THEN LLM-research-weaknesses → THEN partner-firm-REVIEW (codified as `feedback_universal_production_pipeline_sequence.md` 2026-05-02). Recommendation: proceed with universal-production-pipeline framing for ε.ι Layer 4. T-file at `~/off-github/library/projects/inherit/T-spike-eps-iota-S1-v66-seed-audit-2026-05-02.md` (23KB; status: complete; maturity: outcome-VALIDATED).

6 min read

S1 spike outcome — VALIDATED 2026-05-02T07:01 BST. T-file written; logging-contract closed 2026-05-02T11:45 BST in cross-link cascade.

Key numerical findings

  • 21 dedicated jurisdiction-or-tradition extensions in v3/extensions/: UK-E&W, Scotland, Ireland, US, Canada, AU/NZ, EU-26, Switzerland, India, Japan, China, Hong Kong, Singapore-Malaysia, UAE, Israel, Brazil, Latin-America, Africa-customary, Islamic-succession, Jewish-succession, Hindu-succession.
  • 32 jurisdictions enumerated in reference-data/jurisdiction-profiles.json (incl. sub-national US states + Indian states + Canadian provinces + Australian states).
  • 11 jurisdictions in reference-data/tax-thresholds.json with multi-decade history (UK NRB/RNRB; US federal estate tax + state; German Erbschaftsteuer; French droits de succession; Indian; Japanese; Australian state; Canadian deemed disposition + provincial probate; Singapore (post-2008 zero); Irish CAT; NZ).
  • 5 faith pillars covered: Christian (implicit baseline), Muslim (Islamic-succession × 11 countries), Jewish (Jewish-succession × 6 countries), Hindu (Hindu-succession × 9 countries), customary (Africa-customary × 6 countries), secular (default).
  • 674 files (yaml/yml/md/linkml/json/ttl, excl. node_modules/.git).
  • ~33,670 lines of structured content amenable to LLM-REFINE: v3 core schema 17,683 lines across 81 schema files + reference-data 15,987 lines across 34 files.
  • Sample mean depth 4.6/5 across 10 sample-read extensions; sample floor depth 4 (Canada, EU, Africa-customary); sample ceiling depth 5 (UK, Scotland, Islamic, Jewish, Hindu, US, India).
  • Repo head: 19ae973a733fb125db206f4c36d02e12a394af2f 2026-04-19; oldest content 2026-04-07; newest 2026-04-14. Active development, not stale archive.

Kill-condition outcome (predefined: <3 jurisdictions OR avg <5 files/jurisdiction OR >18 months stale)

NOT-MET on any leg by margins of 7× / well-above-threshold / 18× respectively.

Strongest evidence (verbatim from T-file §0)

  • UK-E&W extension cites IHTA 1984 / AEA 1925 / IFPA 1975 with monetary thresholds and dates; nilRateBand/RNRB/40%-rate as temporal-rules; IFPA-1975 eligibility 6-class enum; cohabitant-rights / localPropertyTypes / domicile complications.
  • Islamic-succession extension carries Quranic 4:11-12 ground; 22-value heir enum with Arabic terminology; 15-rank asaba hierarchy; awl/radd; wasiyya 1/3 rule; iddah/waqf/mahr; multi-school fiqh (Hanafi/Shafi’i/Maliki/Hanbali/Jafari) with Quranic verse references.
  • Scotland extension distinguishes prior rights (s.8/s.9) from legal rights (jus relictae/legitim/dead’s part); sheriff-court confirmation 3-type taxonomy; Trusts and Succession (Scotland) Act 2024 amendment; cohabitation Family Law (Scotland) Act 2006; crofting; LBTT; whole-blood/half-blood policy note.
  • Switzerland extension cites ZGB Art. 457–640 + 2023 Erbrechtsrevision Pflichtteil + three-parentele structure.
  • US (longest at 629 lines): IRC + TCJA 2017 + SECURE 2.0; community/separate property regimes; elective share; homestead; per-stirpes variants; no-contest; pretermitted-heir; RUFADAA; Louisiana forced-heirship; UPC.

Implications for ε.ι Layer 4 cost story (per T-file §5)

  1. SEED is broader than 8+ jurisdictions. Plan assumed 8+; SEED provides 21 dedicated + 32 enumerated. Cost-per-jurisdiction approximately preserved or slightly reduced.
  2. SEED depth higher than 80%. Sample mean 4.6/5 implies LLM-REFINE step is doing 10-15% of work per jurisdiction, not 20%. Cost story revises slightly DOWNWARD from working assumption.
  3. Faith × jurisdiction cross-matrix already wired via applicableJurisdictions + compatibleWith arrays. REFINE step does not need to reconstruct.
  4. Tax-regime scaffolding mature — 11-jurisdiction tax-thresholds with multi-decade history. REFINE extending to e.g. Saudi Arabia is incremental, not greenfield.
  5. Temporal-rule pattern already universal. $ref: ../../common/temporal-rule.json consistently applied to every monetary threshold + rate. Mission-critical structural decision already solved in v6.6.

No upward revision required to cost story. If anything, cost-per-jurisdiction estimate could revise downward by ~25%.

Suggested next-step framings (per T-file §6 — flagged for ζ-Q3 lock context)

  1. Confirm REFINE/REVIEW pipeline cost model is calibrated to depth-4.6 SEED (10-15% LLM work per jurisdiction × per artefact-type), not depth-4.0 (20%).
  2. Tag the 4-5 weakest jurisdiction extensions (Switzerland, Brazil, Latin-America, possibly HK/PRC) as priority for Phase-1 partner-firm REVIEW so they’re brought up to depth-5 alongside LLM-REFINE pipeline output.
  3. Preserve the dataProvenance + legalSystems + compatibleWith shape in any v2 schema migration — these are the LLM-pipeline’s structural handles.
  4. Treat maturity: draft|candidate|stable as a first-class signal flowing into the REVIEW tier.

What S1 does NOT establish (honesty caveat)

S1 validates that the SEED is high-quality. It does NOT validate that the LLM-assisted REFINE/REVIEW step itself is cheap-and-reliable enough to economically populate Catala/Cedar/SKOS at scale. That validation is the job of S2-S10 in the derisking suite. ε.ι Layer 4’s cost story rests on BOTH legs (high-quality SEED + economical LLM-pipeline); S1 only handles the first.

Universal-production-pipeline-sequence directive (from Rich-directive 2026-05-02 BST)

“there is a huge amount of valuable material revealed by S1, and we need to ensure that our plans utilise it, before using LLMs to then research any weaknesses, and finally passing to a jurisdiction partner to review”

Codified as feedback_universal_production_pipeline_sequence.md (2026-05-02). Sequence: STEP 1 (SEED utilise) → STEP 2 (LLM-REFINE) → STEP 3 (partner-REVIEW). Strictly sequenced; do NOT skip or reorder.

  • ✓ T-file at ~/off-github/library/projects/inherit/T-spike-eps-iota-S1-v66-seed-audit-2026-05-02.md (was already present 07:01)
  • ✓ arch-state §11 row added pointing at T-file
  • ✓ Q-003 file (_pre-final/Q-003-zeta-asset-taxonomy-CCO-BFO-rooted-9-classes-locked.md) cross-link section added
  • ✓ MEMORY.md one-liner pointing at this memory file
  • ✓ This memory file written
  • ✓ active-work-log entry updated (was claiming “NEXT: dispatch S1”; now reflects S1 done + logging closed)
  • ✓ plan v1.1 → v1.2 §1.5 added (S1 outcome briefing for downstream spikes)

Implications for downstream spikes (S2-S10)

  • S2 (LinkML structured_imports CCO/IAO/IOF) — independent of S1; testing substrate-import not jurisdiction content. Currently in-flight 2026-05-02 BST; hit schema-automator parse error on CCO; diagnosis in progress.
  • S3 (OntoUML-in-LinkML phase encoding) — independent of S1; testing OntoUML-LinkML annotation transfer.
  • S4 (UK&W NRB pilot end-to-end pipeline) — DIRECTLY consumes S1 output. Should explicitly seed from v3/extensions/uk-england-wales/uk-england-wales.json (depth 5; IHTA 1984 cited; 352 lines) per universal-production-pipeline-sequence STEP 1.
  • S5 (FIBO SSSOM alignment) — semi-dependent; can leverage S1’s reference-data/local-term-mappings.json (4027 lines) as a SEED for jurisdiction-keyed terminology mappings.
  • S6 (AM-CDM precedent inspection) — independent of S1.
  • S7 (Mondo Disease Ontology workflow precedent) — independent.
  • S8 (Catala HMRC golden vectors) — semi-dependent; SEED material is HMRC IHT Manual which is external, but should reference UK-E&W extension’s IHTA 1984 citations as cross-validation.
  • S9 (OntoGPT/OG-RAG F1) — semi-dependent on S1 (OG-RAG grounding corpus could include v6.6 content).
  • S10 (Partner-recruitment MLP conversation) — independent of S1 mechanically but directionally informed: partners are STEP 3 of the universal-production-pipeline-sequence; S1’s depth-4.6 finding strengthens the partnership pitch (“we have rich SEED material; you provide REVIEW depth at the production-grade frontier”).

Lock-decision implications (Phase E Task 13)

ε.ι lock decision should explicitly cite S1 outcome as evidence that:

  • Universal-production-pipeline framing for Layer 4 is grounded
  • Cost story does not require upward revision (and may revise downward by ~25%)
  • 21+ jurisdictions exist as SEED material (broader than the 8+ working assumption)
  • Partner-firm engagement work product is REFINED output, not v6.6 itself

Graph View

Backlinks