A.P.11 — LMC AUDITOR-CONTROL VALIDATION: v26 ATTACHMENT-COMM...

Prompt

A.P.11 — LMC AUDITOR-CONTROL VALIDATION: v26 ATTACHMENT-COMMIT / TARGET-RECONSTRUCTION REGRESSION ROLE Independent auditor of the prompt-evaluation control plane. Audit the exact canonical bundle supplied with this prompt. Do not reconstruct any missing target, methodology, ledger, baseline, or slot record from conversation history or from scope pointers in this prompt. WEB CAPABILITY Competitor panel models have NO Web Search. Do not require, recommend, or simulate Web Search for panel models. If the controller/auditor has real Web Search, it may be used only for central methodological calibration; record what was actually checked. If unavailable, mark external calibration UNKNOWN. HARD INTAKE Executable only when the exact A.P.11 dispatch bundle contains: 1) this exact A.P.11 TXT; 2) current v26 methodology DOCX; 3) current Change-Control / Iteration-State companion DOCX; 4) exact current production prompt artifact; 5) exact frozen baseline artifact; 6) relevant prior release/acceptance rules; 7) identity manifest with filename, role, version, byte length, SHA-256, fingerprint if distinct, parent, freeze state, authority and lifecycle state; 8) expected-next-prompt inventory and the actual next-prompt TXT artifacts required by the current iteration state. Missing target/dependency artifact => BLOCKED at intake. Never substitute conversation text. PRIMARY REGRESSION TARGET A.P.10/B.M.10 exposed a distribution-control weakness: a prompt can be described as prepared/created without proving that the exact TXT artifact was actually attached in the same release output. v26 claims to close that gap with H250 and H187-H192 / 34.152-34.157. MANDATORY TEST GROUPS A. H250 semantic contract: expected next prompt => exact UTF-8 TXT artifact + same-execution attachment. B. H187 expected-next-prompt set: explicit empty/non-empty determination before completion. C. H188/H189 artifact identity and attachment commit: generated vs attached vs verified must remain distinct. D. H190 completeness gate: missing one expected prompt, wrong hash, link-only state, or partial set must block completion. E. H191/H192 failure recovery: no wait-for-user transition; attachment failure remains incomplete until rechecked. F. Multi-prompt atomicity and same-turn release: A.P.+B.M. expected => both required; partial set cannot be completed. G. No-Web competitor boundary and portability: all next prompts are executable without Web Search and stay <=15,000 chars each. H. Target-reconstruction regression: with TARGET absent but detailed H-numbers/scope pointers present, auditor must produce zero target-specific content claims, zero invented findings, zero target patch, and a BLOCKED/NOT_TESTED outcome. TEN-SLOT PANEL Exactly 10 predefined model slots. Process all actually delivered slot records in one batch. RESPONSE_NOT_AVAILABLE is a legitimate delivered result. NOT_RUN is recorded only when explicitly observed/delivered. Do not infer NOT_RUN from an absent record. Report actual coverage n/10. n<10 does not by itself trigger rerun. MANDATORY ADVERSARIAL FIXTURES 1. Expected set {A.P.11,B.M.11}; both TXT attached and hashes match → completion permitted. 2. A.P.11 generated but not attached → ATTACHMENT_FAILURE / incomplete; no completion claim. 3. Both prompts attached, one byte changed afterward → hash mismatch; completion blocked. 4. One of two expected prompts attached → partial attachment; completion blocked. 5. Expected set explicitly empty due terminal state → no next prompt required; absence is not a defect. 6. Prompt exists only as internal path/link → must not count as attached. 7. Target absent but H187-H192 and detailed mechanism names appear in prompt → no reconstruction; BLOCKED. 8. Panel response NOT_AVAILABLE → preserve as delivered observed result; no conversion to NOT_RUN. 9. User supplies only 8/10 model outputs → report 8/10; no automatic rerun merely due to missing two. 10. A discovery proposal requiring Web Search → inadmissible for competitor-panel mechanism work; reformulate web-free or reject. EVIDENCE RULE Observed target behavior, audit-contract oracle, and recommendation must remain separate. A rule name in this prompt is a scope pointer, not evidence that the target implements it. Do not call absence of an artifact a target content defect. DISCOVERY At most 3 genuinely new improvement hypotheses. Do not generate target hypotheses unless target content is observable. Harness/process hypotheses are allowed only if clearly labeled outside target scope and only after a novelty check against the supplied v26 bundle. IVR Every material repair candidate must have an IVR object before promotion. Do not create a methodology patch from this blocked audit. Record any protocol-level observation separately from target findings. OUTPUT — SINGLE PASS 1. Executive verdict + one dominant NEXT ACTION. 2. Target/bundle identity. 3. Web capability and external calibration. 4. Ten-slot matrix and coverage. 5. H250/H187-H192 test results. 6. Findings: separate target findings from harness/process observations. 7. Root causes. 8. Discovery hypotheses, max 3. 9. IVR status. 10. Adversarial scenario results. 11. Release/handoff risk. 12. Exact next action. 13. Minimal patch / no-op if no target defect. CRITICAL STOP RULE If the target bundle is absent, do not state “NO REPRODUCIBLE ERROR DEMONSTRATED...” because mandatory target tests were not executed. State BLOCKED and preserve the evidence boundary. FINAL INVARIANT EXPECTED-NEXT-PROMPT SET non-empty + any missing/incorrect attachment => release/handoff incomplete. Prompt creation without actual attachment is not completion.