TT Claude Memory

feedback_bold_front_loaded_synthesis_preference

Properties1
descriptionLocked 2026-05-02T05:30 BST per Rich directive after fragility-signal at ζ-Q5 surfaced that the original CORE-over-Hybrid rule had been subtly biasing option-generation toward conservatism even when noted as Phase-3-specific. New rule favours BOLD-FRONT-LOADED SYNTHESIS for foundational architectural decisions where cost-of-change-later is high; Phase-3 minimum-change pattern stays valid only for per-module authoring with low cost-of-change. Codified after Rich's analysis that aspirations were being restricted in Phase 2 by κ.θ-LITE weight choices + asymmetric devil's-advocate inclusion + Rich-pattern row in stakeholder sanity check.

4 min read

Rule: Favour bold-front-loaded synthesis options over minimum-change variations when generating + scoring + recommending architectural options. Phase-context modulates the strength:

  • Phase 2 (foundational architecture): STRONG bold-synthesis preference. Cost-of-change-later HIGH (cascades through 27 primitives + 23 SKOS schemes + 19 CI gates). Reversibility EXPENSIVE (Full Reversion 2026-05-01T20:45 cost ~10-11 hours). Acquirer-DD load-bearing. Bold-front-loading payback HUGE.
  • Phase 3 (per-module authoring): CONTEXTUAL preference — minimum-change appropriate when cost-of-change is genuinely low AND the module sits cleanly within locked Phase-2 architecture; bold-synthesis appropriate when the module surfaces gaps in the locked architecture or has high cross-module knock-on.
  • Phase 4-5: bold-synthesis preference returns (closure prep + stress-test execution define the build-start ceiling).
  • Phase 6-9: contextual; per-brand productisation often minimum-change against locked architecture.

Why: Original feedback_phase_3_q_minimum_change_preference (codification candidate; never formally written; informal references in PRE-REFINED-PROMPT-archived ζ-Q11..Q14) was correctly Phase-3-scoped. But its mere presence biased option-generation in Phase 2 via:

  1. κ.θ-LITE weight choices at ζ-Q1: criterion #11 TT commercial velocity weighted 0.15 (highest single weight); criteria #6 Maintainability + #8 Tooling ecosystem add another 0.15. Sum 0.30 weight on conservatism-axes vs 0.18 on aspiration-axes. Subtle structural bias of 0.12 toward conservatism.
  2. Asymmetric devil’s-advocate inclusion: every Phase-2 Q included a “frozen / do-nothing / minimal” option (κ.α / ξ.α / ε.α / ω.ζ / ψ.α) but NEVER the symmetric “go bigger than the obvious next-step” devil’s-advocate.
  3. Stakeholder sanity check Rich-pattern row: noting Phase-3 CORE-over-Hybrid pattern as a stakeholder-row in EVERY Phase-2 Q applied subtle counter-pressure against bold options.
  4. Cost-bias framing: cost-rows skewed toward “cheaper is better” when for a £200K+ acquirer-narrative-defining 2027-2029 DD artefact, +£15-30K cumulative differentials are rounding errors.
  5. Anchoring on PRE-REFINED-PROMPT options: option-generation tended to be SLIGHT extensions from PRE-REFINED-PROMPT-archived picks rather than fresh re-derivation.

How to apply:

  1. In option-generation: include AT LEAST ONE option that is “go BIGGER than the obvious next-step” — symmetric to the “frozen / do-nothing / minimal” devil’s-advocate. Phase-2 Qs especially should include an aspirational option like ξ.+ (G+ + Phase-1-graphRAG + Phase-1-Cedar-Analysis + Phase-1-formal-verification) that goes beyond the obvious next-step.
  2. In scorecard weighting: AHP pairwise-comparison session (richard-task #205) is the corrective; pending; consider acquirer-narrative (#12) weighted equal-or-higher than commercial-velocity (#11) for foundational decisions.
  3. In stakeholder sanity check: Replace “Rich-pattern (Phase-3 CORE-over-Hybrid preference per feedback_phase_3_q_minimum_change_preference)” with “Rich-pattern (bold-front-loaded synthesis per feedback_bold_front_loaded_synthesis_preference; 4/4 refined-prompt Phase-2 decisions chose synthesis: κ.θ utility-tree + ξ.η G+ + Phase-1-graphRAG + ε.ε CCO/BFO + ω.η SSSOM-mediated)”.
  4. In cost-of-ownership framing: when cost differential ≤£30K cumulative 5y for a foundational decision, treat as “rounding error vs acquirer-narrative-value”; do NOT let it dominate scoring.
  5. In confidence assessment: if option-set feels narrow/clustered/fragile, EXPLICITLY add the aspirational devil’s-advocate AND broader research/spike work BEFORE locking. Per Rich’s ζ-Q5 fragility-signal pattern.

Boundary tests (when bold-synthesis preference applies STRONGLY):

  • ✓ Phase 2 foundational architectural decisions
  • ✓ Phase 4-5 closure-prep / stress-test execution
  • ✓ Cross-module primitive promotion decisions
  • ✓ Substrate-architectural decisions (LinkML / Cedar / Catala / SSSOM substrate choices)
  • ✓ Standards-track positioning decisions
  • ✓ Acquirer-DD load-bearing decisions

Boundary tests (when minimum-change preference applies — narrow Phase-3-specific):

  • ✓ Per-module schema authoring within locked Phase-2 architecture
  • ✓ Per-jurisdiction parameter-binding within locked rule structure
  • ✓ Per-row SSSOM TSV catalog additions within locked alignment-axiom-strength regime
  • ✓ Operational tweaks that don’t touch architectural commitments

Codification trigger: Rich directive 2026-05-02T05:30 BST: “first we need to modify the likes of feedback_phase_3_q_minimum_change_preference-style to favour bold-front-loaded synthesis.” Validates fragility-signal at ζ-Q5 ψ.α-ψ.η option-set + Claude self-audit identifying 4 mechanisms by which the older rule biased Phase-2 option-generation.

4/4 refined-prompt Phase-2 decisions retroactively validate this rule: ζ-Q1 κ.θ utility-tree synthesis + ζ-Q2 ξ.η G+ + Phase-1-graphRAG synthesis + ζ-Q3 ε.ε CCO/BFO + 9 i-ζ classes (convergent with PRE-REFINED-PROMPT but rigour-front-loaded) + ζ-Q4 ω.η BFO-only with SSSOM-mediated synthesis (NEW Greek-prefix synthesis NOT in PRE-REFINED-PROMPT option-set). Pattern fully validated; codified now.

Effect on retroactively reviewing ζ-Q1..ζ-Q4: locks may stand OR may surface bolder alternatives if explicit aspirational-devil’s-advocate had been included. Rich-judgement-call whether to run Option D re-validation. Default: apply prospectively from ζ-Q5 onward.

Related memories (forward references; this rule downgrades these):

  • project_zeta_q11_locked_pi_beta_2026_05_01.md (PRE-REFINED-PROMPT-archived)
  • project_zeta_q12_locked_theta_beta_2026_05_01.md (PRE-REFINED-PROMPT-archived)
  • project_zeta_q13_registrations_phase_1_scope_2026_05_01.md (PRE-REFINED-PROMPT-archived)
  • project_zeta_q14_alignment_axiom_strength_fresh_under_r2_2026_05_01.md (PRE-REFINED-PROMPT-archived)

These all referenced the codification candidate feedback_phase_3_q_minimum_change_preference informally. Under bold-front-loaded synthesis preference, those Phase-3 picks may stand (genuinely Phase-3 contextual) but the underlying rule is downgraded from “codification candidate after 4 consecutive picks” to “narrow Phase-3-conditional pattern”.

Graph View

Backlinks