TT Claude Memory

feedback_research_artefact_forward_traceability

Properties1
descriptionEvery TT research artefact (T-files / cumulative-state docs / scorecards / Q-cascade files / spike T-files / amendment registries / arch-state rows / risk register entries / brand-architecture sections / partner-model sections / etc.) MUST be authored with FORWARD readers in mind — not just the immediate-task reader. Add `phase_q_relevance:` frontmatter naming the future Phase Q lock-decisions / sprint commitments / acquirer-DD claims this evidence will inform; add `companion_files:` block with role / relationship / version_pinned / pin_kind for upstream + downstream docs. Without forward-traceability, prior research silently fails to compound — Phase Q-formulation re-derives conclusions already reached, and ~£2-5K of prior research per artefact gets re-spent.

6 min read

Every TT research artefact must be authored as if a future reader (Claude or Rich) will arrive 3-12 months from now with a question we have NOT yet framed. The artefact’s frontmatter must give that future reader enough metadata to:

  1. Decide whether this artefact is relevant without reading the body
  2. Find the upstream evidence it was built on (T-files / spike T-files / books / web sources)
  3. Find the downstream consumers that have already cited it (cascade-Q files / arch-state rows / scorecard rows / amendment rows)
  4. Locate future Phase Q work this evidence is load-bearing for (Phase 2 ζ-Q lock-decisions / Phase-1 Sprint commitments / acquirer-DD claims / standards-engagement narratives)

Why: Reuse of prior research is the highest-leverage discipline in long-running engineering projects. A T-file authored in Round 7 (October 2025) might be load-bearing for a Phase 2 ζ-Q lock decided in Phase E (Q4 2026) — a 12+ month gap. If the T-file’s frontmatter doesn’t say “this evidence is load-bearing for ζ-Q5 alignment-axiom strength + ζ-Q15 Wills follow-through + Phase-1 Sprint S2 SSSOM build”, the future Q-formulator may not find it. The 22-spike Q&A-formulation suite (May 2026) surfaced this concretely: ~5 of 22 spikes had ≥50% prior-work coverage in T-files but the prior research was not signposted forward — the spikes were nearly re-runs.

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

  1. 2026-05-03 Spike 22 prior-work audit — Identified Spike 8 (~80-90% prior coverage in Q-003 + S1 + S2.5 + S2.6), Spike 22 (~70% prior in T59+T60+T52+T58), Spike 20 (~70% prior in T82), Spike 12 (~70-80% prior in arch-state §17 + Q-cascade files), Spike 11 (~50-60% prior in T82 + R-NEW). All 5 had T-files in ~/off-github/library/projects/inherit/ but NONE had phase_q_relevance: pointing to the upcoming spikes — discovered only via post-hoc Tier-2 grep audit.
  2. Round 7 T-files without phase_q_relevance — T1-T82 were authored as research deliverables but lack metadata pointing forward to which Phase 2 / Phase 3 / Phase 1 Sprint decisions they inform. Future Q-formulators have to guess via topic-mapping table at docs-strategy/CLAUDE.md — works for keyword matches; fails for cross-cutting dependencies.
  3. ε.ι spike T-files (12 files; 2026-05-02) — Each captures empirical evidence (S1-S8) but does NOT name the 22-spike-suite spikes that will consume the evidence (e.g., S4 UK&W NRB → consumed by Spike 13 / Spike 16; S2.10 Cedar Analysis → consumed by Spike 11; S5 FIBO SSSOM → consumed by Spike 8 + Spike 14). Forward-pointing absent.
  4. Cumulative-state docs (6 files; 2026-05-02) — Capture state-as-of-date but don’t name which future Phase Q lock-decisions will reference them. Acquirer-DD package cite-back relies on memory rather than frontmatter.

How to apply (5-step authoring discipline for every new research artefact):

  1. Before writing any body content, draft the frontmatter phase_q_relevance: list:

    phase_q_relevance:
  - "Phase 2 ζ-Q5 alignment-axiom strength lock-decision (Q-005 cascade-Q file)"
  - "Phase 3 ω.η SSSOM-canonical alignment build (arch-state §A-22)"
  - "Phase-1 Sprint S2 production SSSOM commit-time work"
  - "Acquirer-DD package §3.4 standards-engagement evidence"

    If you cannot name at least ONE forward consumer, the artefact’s load-bearing-ness is unclear — pause + decide whether it should be authored at all.

  2. Always include companion_files: block with structured role / relationship / version_pinned / pin_kind:

    companion_files:
  upstream_t_files:
    role: "Source evidence consulted at authoring time"
    relationship: "Body §3 cites; conclusions derived from"
    paths:
      - "../../../off-github/library/projects/inherit/T54-OAuth-Keycloak-DPoP.md v1.2"
      - "../../../off-github/library/projects/inherit/T58-id-and-v-vendors.md v1.0"
    pin_kind: snapshot
  q_cascade_file:
    role: "Phase 2 cascade-Q file consuming this evidence"
    path: "../../docs/superpowers/specs/2026-04-29-multi-phase-audit/current-questions/Q-014-...md"
    version_pinned: "v1.1 (2026-05-02T11:00)"
    pin_kind: floor
  arch_state_row:
    role: "Tier-1 arch-state §11/§12/§13/§14 row referencing this artefact"
    path: "../../docs/superpowers/specs/inherit-v2-architecture-state.md"
    version_pinned: "v3.34 (2026-05-03T11:50)"
    pin_kind: snapshot

  3. At authoring time, run the bidirectional cite-back: edit the upstream T-files / cumulative-state docs to mention the new artefact in their consumers: block (a structured frontmatter list of downstream consumers); edit downstream cascade-Q files / arch-state rows to cite the new artefact in their evidence list.

  4. Index in Tier 2 (where applicable): the pgvector library index at ~/off-github/library/projects/inherit/tier-2-pgvector-library-index/ should ingest the artefact (run query_library.py ingest cycle). Long-term: add a phase_q_relevance JSONB column to the library_chunks table for filterable retrieval (--phase=2-zeta-q5 flag).

  5. Update topic-mapping quick-reference at ~/testatetech/docs-strategy/CLAUDE.md “T-file topic-mapping quick-reference” if the new artefact introduces a NEW topic-keyword that future readers will search by.

Required frontmatter shape (canonical recommendation):

---
title: "..."
version: "1.0"
status: approved
date: 2026-05-03T12:30
lastmod: 2026-05-03T12:30
author: "..."
source: "..."
phase_q_relevance:                                          # NEW per this rule
  - "<future Phase Q lock-decision / sprint commitment / DD claim>"
  - "<another forward consumer>"
companion_files:                                            # NEW per this rule
  q_cascade_file: { role / relationship / path / version_pinned / pin_kind }
  arch_state_row: { role / path / version_pinned / pin_kind }
  upstream_t_files: { role / relationship / paths / pin_kind: snapshot }
  downstream_specs: { role / relationship / paths / pin_kind: snapshot }
gold_value: high|medium|low
related:
  - "..."
---

Cross-references:

  • Sibling rule (narrower trigger; READ-time discipline): feedback_check_t_files_first_for_any_inherit_v2_work (locked 2026-05-03) — covers READING T-files BEFORE work; this rule covers AUTHORING artefacts FOR future readers.
  • Sibling rule (narrowest trigger): feedback_actively_use_t_files_in_scorecard_authoring (elevated 2026-04-27) — only triggered on scorecards.
  • Substrate (canonical convention): ~/testatetech/docs-strategy/frontmatter-conventions.md v1.3+ §3.6 documents phase_q_relevance: as a recommended field.
  • Substrate (global discipline): ~/.claude/CLAUDE.md §17.7 codifies forward-traceability as cross-session policy.
  • Substrate (retrieval index): Tier 2 pgvector library index at ~/off-github/library/projects/inherit/tier-2-pgvector-library-index/.
  • Plan integration: 22-spike plan v1.15 §1.1 frontmatter contract gains phase_q_relevance: + companion_files: block; §1.6 logging contract 8 → 9 steps with NEW Step 9 forward-traceability cascade.
  • Backfill task (deferred, ~2-3h): backfill phase_q_relevance: into 21 substantive recent artefacts (12 ε.ι spike T-files + 6 cumulative-state docs + 5 closed 22-spike T-files including Tier 2). Track at richard-tasks.md.
  • Backfill task (deferred, ~3-4h, lower priority): backfill into T1-T82 (74 files).
  • Tier 2 column extension (deferred, ~½ day): add phase_q_relevance JSONB to library_chunks table; extend query_library.py with --phase filter.

Trigger context (broadest of the three sibling rules): ANY new research artefact authored within TT scope — T-files / cumulative-state docs / scorecards / Q-cascade files / spike T-files / amendment registry rows / arch-state rows / risk register entries / brand-architecture sections / partner-model sections / acquirer-DD package sections / standards-engagement notes / book-research synthesis files. The discipline applies whenever the artefact carries evidence that could compound for a FUTURE decision not yet framed.

Failure mode if rule isn’t applied: prior research silently fails to compound; Phase Q-formulation re-derives conclusions already reached; ~£2-5K of prior research per artefact gets re-spent; acquirer-DD narrative weakens because “research investment evidence” is hard to surface; the 22-spike-suite-style “did we do this before?” audits become routine rather than rare.

Strategic framing (Rich-directive 2026-05-03): “the top priority is ensuring that all research is utilised when we resume asking questions. i believe that our policies and processes are very important. frontmatter can make a vital difference.” This rule is the policy-and-process implementation of that directive.

Graph View

Backlinks