feedback_check_t_files_first_for_any_inherit_v2_work

At the START of ANY substantive INHERIT v2 / TT-brand work, BEFORE any other research, identify the T-files relevant to the topic via the docs-strategy CLAUDE.md “T-file topic-mapping quick-reference” table + READ them. T-files are TT-specific pre-existing research that compounds: they cite specific tools, vendors, version-pinned conclusions, production-precedent links, and gold-value markers that no general-purpose research substitute can replicate.

Why: Reuse of prior research is the highest-leverage discipline in long-running engineering projects. The TT research library has ~70 T-files at ~/off-github/library/projects/inherit/T1-T82+ covering ~85% of the topics any TT-related work touches. NOT consulting them first means:

• Re-deriving conclusions already arrived at (waste of time)
• Missing version-pinned vendor specifics (T54 names cloud-iam.com + loginfactor.com; not “Hetzner” or “OVH”)
• Producing recommendations that conflict with previously-locked TT positions
• Weakening the acquirer-DD narrative by NOT showing the prior research investment

Anti-patterns this rule prevents (concrete failure modes observed):

1. Spike 22 (2026-05-03) — Rich’s observation: spike override prompt I wrote referenced HMRC/UK statutes pre-fetch + Tier 2 queries, but did NOT call out T59 (Open Banking + HMRC IHT) / T60 (Tier-2 UK gov APIs) / T52 (UK gov agency tech) / T58 (ID&V vendors) which are DIRECTLY load-bearing for trigger-event detection (life-events: marriage / divorce / asset acquisition via Open Banking + Land Registry + Companies House).
2. 2026-04-27 §4 closing-question — named “Hetzner / OVH” as Keycloak managed-provider alternatives when T54 actually names cloud-iam.com + loginfactor.com. Body content was T-file-cited; closing question slipped (codified as feedback_actively_use_t_files_in_scorecard_authoring).
3. Spike 14 (L4M) — risk — without checking T17 (OntoGPT) / T27 (Claude API) / T28 (OpenAI+Gemini) / T82 (architecture-evaluation methodology) first, the spike-runner could re-derive the L4M F1 measurement methodology from scratch instead of building on T-file precedents.

How to apply (5-step pre-flight before any substantive work):

1. Identify topic-keywords from the work-at-hand (e.g., “trigger event”, “Open Banking”, “Will update”, “OWL DL reasoner”, “Cedar policy”).

2. Grep the docs-strategy CLAUDE.md “T-file topic-mapping quick-reference” for each keyword:

   grep -iE "<keyword1>|<keyword2>" ~/testatetech/docs-strategy/CLAUDE.md | head -10

3. Read the matching T-files (typical 2-5 T-files per work-task):

   head -50 ~/off-github/library/projects/inherit/T<N>-*.md
   grep -iE "<keyword>" ~/off-github/library/projects/inherit/T<N>-*.md | head -20

4. Cite the T-files in the work-at-hand:

   • Spike T-file §cross-links section MUST cite each relevant T-file with specific quote + URL/path
   • Phase Q-formulation cascade-Q file §10 MUST cite T-files informing the option-set
   • Scorecard rows MUST cite T-files in the rationale column
   • Acquirer-DD package MUST cite T-files as “research investment evidence”

5. If a T-file is missing on the topic (genuine gap), flag it explicitly:

   • “No T-file exists yet for {topic}; recommend Round-{N+1} spike-avoidance research”
   • Don’t invent vendor names / version numbers / production-precedent claims

Quick-reference (lifted from docs-strategy CLAUDE.md so it’s auto-loaded with this memory):

• Auth / IDP / OAuth / Token Exchange / DPoP → T54
• KMS / pgsodium / Vault / encryption substrate → T55
• Triple-store / graph DB / vector DB → T67 (+T50, T51 for triple-store-specific)
• UK gov agency integration (HMCTS / HMRC / Companies House / RoS) → T52, T59, T60
• UK gov APIs Tier-2 (Pensions Dashboard / Tell-Us-Once / OPG / HMLR / DVLA) → T60 (TWO files; cross-linked v1.1)
• Open Banking + HMRC IHT → T59
• ID and V vendors → T58
• Catala ecosystem + tooling → T36, T43, T57, T81
• Cedar policy → T35, T42, T80
• LinkML schema / pipelines → T13, T37, T44, T81
• Standards bodies (OASIS / W3C / OIDF / AAIF) → T41, T48
• AI-assisted rule authoring → T39
• E-signing / eIDAS / electronic wills → T56
• Verifiable Credentials / Bowtie / Conformance → T63
• Faith-tradition succession → T49
• Architecture-evaluation methodology (ATAM / utility tree / fitness functions / sensitivity-analysis / AI evaluation) → T82 (companion: indexed/bass-software-architecture-4e/)
• PROV-O / temporal model / event-stream → T6
• OntoGPT / AI extraction → T17
• MCP / agent integration → T1
• Jentic JAIRF / agent-readable APIs → T16
• OWL reasoners (HermiT / Pellet) → T50
• Claude API / Anthropic SDK → T27
• OpenAI / Gemini SDK → T28

Cross-references:

• Sibling rule (narrower trigger, same intent): feedback_actively_use_t_files_in_scorecard_authoring (elevated 2026-04-27)
• Substrate enabler: Tier 2 pgvector library index (NOW including T-files per 2026-05-03 update)
• Quick-reference home: ~/testatetech/docs-strategy/CLAUDE.md “T-file topic-mapping quick-reference”
• T-file directory: ~/off-github/library/projects/inherit/T*-*.md (~70 T-files)

Trigger context (broader than the parent rule): ANY spike-runner / phase Q-formulation / architectural-decision / risk-rescore / brainstorming / scorecard-authoring / acquirer-DD / partner-pitch / 22-spike-suite / TT-brand-work that touches an INHERIT v2 / IK / IAS / IW / MFI / LL / OpenInherit topic. NOT just scorecards (which the parent rule covered) — the discipline applies whenever prior TT research can be cited.

Failure mode if rule isn’t applied: re-deriving prior research from scratch + producing inferior recommendations + weakening acquirer-DD narrative + missing TT-specific empirical evidence. Spike 22 was the canary case (2026-05-03).