B.M.11 — METHODOLOGY v26 VALIDATION: AUTOMATIC-NEXT-PROMPT A...

Prompt

B.M.11 — METHODOLOGY v26 VALIDATION: AUTOMATIC-NEXT-PROMPT ATTACHMENT / DISTRIBUTION CONTROL ROLE Independent auditor of the current long-term LMC methodology as the autonomous development/evaluation control plane. Audit only the exact v26 methodology bundle supplied with this prompt. Do not reconstruct missing methodology from conversation history. WEB SEARCH Competitor models are NO-WEB by design. Never require or recommend competitor Web Search. Central Web Search may be used only by the controller/auditor for current or historical methodological calibration, and only if actually available; otherwise mark external calibration UNKNOWN. HARD INTAKE Required atomic B.M.11 bundle: 1) exact B.M.11 TXT; 2) exact v26 methodology DOCX; 3) exact v26 Change-Control / Iteration-State companion DOCX; 4) exact current production prompt artifact; 5) exact Frozen baseline; 6) referenced prior release/acceptance rules; 7) identity manifest with version/hash/fingerprint/parent/freeze/authority/lifecycle and bundle inventory; 8) expected-next-prompt inventory and actual A.P.11/B.M.11 TXT artifacts for the next validation cycle. Missing required item => BLOCKED. Never reconstruct. AUDIT OBJECTIVE Verify that v26 closes the operational gap observed in previous distribution iterations: “prompt prepared/created” must not be treated as equivalent to “prompt actually attached and verified”. The audit target is the methodology control plane, not the production prompt. TEST GROUPS TG1-TG10 TG1 H250 strengthened semantics: next prompt is an actual exact TXT release artifact, not a recommendation. TG2 H187 expected-next-prompt determination: empty vs non-empty set is explicit and state-derived. TG3 H188 same-execution creation: every expected prompt created as UTF-8 TXT, <=15,000 chars, identity-bound. TG4 H189/H190 attachment commit/completeness: generated → attached → verified are distinct; missing/incorrect attachment blocks completion. TG5 H191/H192 no-wait auto-attach + attachment failure recovery: controller must not wait for an extra user request when a next prompt is legitimately expected. TG6 34.153 provenance chain and lifecycle-state integrity: EXPECTED → GENERATED → IDENTITY-BOUND → ATTACHMENT-COMMITTED → ATTACHMENT-VERIFIED → HANDED-OFF. TG7 34.154 multi-prompt atomicity: if multiple prompts are expected, all must be verified before handoff completion. TG8 34.155-34.156 portability and same-turn release invariant; competitor Web Search remains outside panel capability. TG9 34.157 internal validation: positive, negative, adversarial and metamorphic fixtures for attachment control. TG10 recurrence/interaction with prior artifact identity, baseline protection, acceptance precedence, bounded recovery and post-release state. Ensure new controls do not silently weaken older gates. MANDATORY ADVERSARIAL SCENARIOS 1. expected set {A.P.11,B.M.11}, both attached + hash match; 2. one prompt generated but not attached; 3. attachment hash mismatch after commit; 4. partial multi-prompt attachment; 5. expected-next-prompt set explicitly empty; 6. internal link only, no actual attachment; 7. attachment platform unavailable; 8. target missing but prompt contains detailed H-rules/scope pointers — no reconstruction; 9. 8/10 or 9/10 panel execution — report actual coverage, no automatic rerun solely for n<10; 10. competitor discovery proposal requires Web Search — reject/reformulate web-free. METHOD RULES Observed target content, oracle expectations, and recommendations are separate evidence classes. A rule mentioned in this prompt is not evidence that v26 contains the rule unless the exact methodology artifact is supplied and inspected. DISCOVERY At most 3 new methodology/process hypotheses. Any candidate must clear novelty, benefit, cost, risk, interaction, testability, portability and admission gates. Do not promote a candidate merely because it would be convenient. IVR AND RELEASE Every material rule/process change must map to IVR. Self-test PASS is not independent acceptance. Do not alter the production prompt unless a separate AP change record authorizes it. OUTPUT — SINGLE PASS 1. Executive verdict + one dominant NEXT ACTION. 2. Intake / identity / provenance. 3. Web capability and external calibration. 4. TG1-TG10 results. 5. Mandatory adversarial scenarios. 6. Findings, separately classified as target vs harness/process. 7. Discovery hypotheses, max 3. 8. IVR coverage and validation status. 9. Independent validation status. 10. Convergence/budget status. 11. Release/post-release impact. 12. Exact next action. 13. Minimal patch / no-op. CRITICAL DISTINCTION If the expected-next-prompt set is non-empty, “prepared”, “created”, or “linked” is not sufficient. Completion requires ATTACHMENT-VERIFIED for every expected prompt. If the set is empty for a legitimate state reason, no prompt is required. FINAL STOP RULE If mandatory target bundle evidence is incomplete, do not issue a content-level PASS/FAIL, do not reconstruct v26, and do not claim the absence of evidence as a clean bill of health.