TT Claude Memory

feedback_claude_does_research_inline_at_q_lock_time

Properties1
descriptionWhen Q-lock authoring surfaces a research/spike need, do it INLINE; do NOT spawn a Paul-bound Phase-1.5+ ledger task to defer it. Paul becomes validator, not researcher.

4 min read

Rule: When Q-lock authoring (or any substrate-producing batch) surfaces a research/spike/verification need, do the work inline AT-LOCK rather than spawning a Phase-1.5+ Paul-bound ledger task to defer it. Promote findings directly into the Q-lock substrate (cascade-Q amendment / SKOS scheme / Catala body draft / SSSOM rows / arch-state prose). Paul becomes validator, not researcher.

Why: Locked 2026-05-20T19:45 BST. Rich-directive in response to my framing that “each Q-lock will spawn its own batch of Paul-bound substrate work” — Rich said: “I am keen to allow Claude to do the research or spike work, as soon as it has identified it is required.” The implication: deferring research-class work into Paul-bound ledger tasks fragments the cascade + leaves substrate gaps that compound. Substrate-quality compounds at Q-lock-time (per [[substrate-quality-is-axis-velocity-conditioning]]); the right time to fill a substrate gap is when it’s identified, not later.

Extends [[spikes-inline-not-tasked]] (≤30 min inline) and [[do-now-over-task-list-addition]] (parent discipline). Threshold lifts from “≤30 min spikes” to “research/spike work identified during Q-lock authoring, regardless of duration within reason”. The implicit time budget is “as long as it takes to fill the gap, time-boxed at ~2h before surfacing to Rich”.

How to apply:

  1. At Q-lock authoring time: when “we need to verify X doctrine / check Y statute section / find Z case-law authority / bridge to AA AAOIFI standard / etc.” surfaces — DO IT INLINE. Use authoritative-source fetch:

    • UK statute → legislation.gov.uk direct fetch
    • UK case-law → BAILII direct fetch (with conjoined-appeal alias note per BATCH χ Pitt v Holt / Futter precedent)
    • Swiss law → fedlex.admin.ch via Playwright MCP (BATCH ψ pilot 3 precedent)
    • EU law → EUR-Lex (UNVERIFIABLE-ONLINE flag if returns empty; document as expert-review-pending)
    • HMRC manuals → gov.uk hmrc-internal-manuals direct fetch
    • Faith-tradition → AAOIFI (paid; UNVERIFIABLE-ONLINE + rapporteur queue) / IndiaCode (best-effort) / Singapore SSO (often 403; flag + rapporteur) / Israeli (Hebrew non-standard; flag + rapporteur)

  2. Promote findings DIRECTLY into Q-lock substrate: cascade-Q amendment row / SKOS scheme delta / Catala body sketch (or PROVEN-VIA-BRIDGE if rule-shape permits per MQ-014) / SSSOM mapping_set row / arch-state §15 amendment prose / fixture additions per MQ-011 Axis-2 γ.

  3. DO NOT spawn a Paul-bound Phase-1.5+ task UNLESS:

    • The work is genuinely partner-firm-DD-class — requires £-coded Paul-time + SRA-regulated opinion + contextual case-law judgement Claude cannot defensibly produce alone
    • The work hits a substrate genuinely Claude can’t access (paid databases like Westlaw / LexisNexis / AAOIFI Sharia Standards behind paywall)
    • The work requires rapporteur engagement (faith-tradition cross-school doctrinal judgement; route via task-91 framework per BATCH φ task-87 substrate)

  4. Time-box at ~2h inline research before surfacing to Rich: if the gap is so large that filling it inline would take >2h, ntfy DECISION-NEEDED: research gap at Q-lock substrate, est. <hours>h to fill inline OR spawn task + let Rich pick. Default behaviour is fill-inline; ask-Rich only at >2h threshold.

  5. Documentation discipline: when filling a research gap inline at Q-lock time, capture the verification chain in the cascade-Q file’s §B.0 verification-evidence (per MQ-015) — every fetch URL + every authoritative source. The substrate is then defensible without Paul-side validation as the primary trust signal; Paul-validates the Claude-research findings rather than originating them.

  6. Anti-pattern: do NOT write phrases like “Phase-1.5+ task spawned: verify Re Goodman 2013 removal grounds doctrine + structure taxonomy” during Q-lock authoring. Instead: open BAILII, fetch the case, extract the doctrine, structure the taxonomy, promote into Q-lock substrate, lock with the work done. The task spawn is a DEFERRAL signal; Rich’s directive eliminates that deferral whenever feasible.

  7. Connection to acquirer-DD §3.x: each inline-research lock produces a defensibly-thick Q-lock; an acquirer auditor sees research-in-the-lock not research-deferred-in-a-ledger-task. Substrate density per Q-lock is a load-bearing acquirer-DD signal.

  8. Connection to verify-before-author 22+/22+: every inline-research finding is itself subject to MQ-015 INSTALL-VERIFY-BEFORE-CLAIM + §13.5 recursive author-side mirror. The pattern-stability test extends with each Q-lock-inline-research cycle.

Triggering pattern: if I’m about to write a Paul-bound task spawn that contains the phrase “verify X” / “research Y” / “establish Z taxonomy” / “check authoritative source for AA” — STOP and re-read this memory. Default to inline; only spawn Paul-bound task if criteria in (3) genuinely apply.

The 4 substrate-pre-staging batches in flight 2026-05-20 (ο + ε + ι + λ) already align with this directive — they’re DOING research inline that would otherwise have sat as Paul-bound tasks. This memory codifies the pattern so future Q-lock sessions apply it as default discipline.

Graph View

Backlinks