B.M.6 — FULL-METHOD RED-TEAM AUDIT PORTABLE v21 CANDIDATE + ...

Prompt

B.M.6 — FULL-METHOD RED-TEAM AUDIT PORTABLE v21 CANDIDATE + WEB CROSS-CHECK + AUTONOMOUS NEXT-ACTION Audituj pouze níže vložený PORTABLE v21 CANDIDATE. Je to jediný normativní objekt. Hledej skutečné reprodukovatelné control-plane chyby vedoucí k rozhodovací divergenci, unsafe release, truth failure, unbounded loop, skipped validation, self-approval nebo nedeterministickému lifecycle. Nezaměňuj chybějící detail za chybu bez rozhodovací divergence. IZOLACE A.P./B.M. B.M. je metodologický stream. A.P. je cílový LMC stream. Tyto streamy jsou experimentálně nezávislé. Výsledek jednoho již provedeného experimentu nesmí být post-hoc přepsán výsledkem druhého. Cross-stream adjudication smí pouze atribuovat root cause a zvolit další autorizovanou akci. Tento audit neodhaduje žádný chybějící A.P. výsledek a nesmí jej konstruovat z paměti. WEB SEARCH Je-li Search dostupný, cross-check KAŽDÝ material finding i material repair/redesign: problém i řešení; preferuj primární/autoritatívní zdroje a ověř recency/applicability. Pro každý bod: SEARCH INTENT→SOURCES→RECOMMENDATION→AGREEMENT/CONFLICT→APPLICABILITY→ADOPT/PARTIAL/REJECT/DEFER→REASON. Web nesmí přímo přepsat metodiku; pouze METHOD DECISION→INTERNAL TEST. Search-off nesimuluj. EPISTEMIKA U=dodaný text; V=derivace; Z=externí evidence; O=inference. Consensus není proof. Externí doporučení nejsou automatická autorita. RATING R3=kritická divergence/unsafe release/truth failure; R2=materiální control-plane mezera; R1=menší reprodukovatelná chyba; 0=bez relevantního nálezu; UNRESOLVED=nelze rozhodnout. PORTABLE v21 — NORMATIVE CORE PRIORITY: SAFETY/BLOCKING > INPUT/EXPERIMENT VALIDITY > SOURCE/PROVENANCE > STATE/PRECEDENCE/CLOSURE > EVIDENCE > REPAIR/REGRESSION > CONVERGENCE > EXTERNAL VALIDATION > ENHANCEMENT > BREVITY. 1. RECEIVE RESULTS → VALIDITY/PROVENANCE → ROOT-CAUSE MERGE → WEB CROSS-CHECK → METHOD DECISION → CANDIDATE CHANGE → INTERNAL SELF-TEST → REPAIR/RETEST → CONVERGENCE → PRE-RELEASE → ISSUE → POST-ISSUE ACCEPTANCE → ACCEPTED VERSION → BUILD NEXT A.P./B.M. → EXACT-PAYLOAD GATE → DISPATCH. 2. CURRENT/FROZEN immutable; CANDIDATE oddělený. Material change má change-ID/version-ID/provenance. Candidate→issued→accepted→CURRENT/FROZEN identity-bound. 3. Material finding/repair má FINDING→EVIDENCE→ROOT CAUSE→DECISION record. External evidence = discovery/evidence, ne write-back. 4. SEARCH: relevant evidence → CONFLICT CHECK → RESEARCH→METHOD DECISION → INTERNAL TEST. No direct rewrite. 5. ARTIFACT TYPE: type before manifest; mismatch blocking. 6. RECURRENCE: INCIDENT FINGERPRINT = CONTROL BOUNDARY + FAILING INVARIANT + TARGET LAYER + NORMALIZED OBSERVABLE SIGNATURE + relevant artifact scope. Same fingerprint after actual repair+redistribution = recurrence. 2nd→architecture review; 3rd identical→hard stop. 7. DELIVERY INTEGRITY: source/package/distributed payload identity. Change after PASS invalidates gate. 8. RECOVERY: CLASSIFY → LOCALIZE → REPAIR → PRE-DISPATCH RECHECK → MINIMAL VALIDATION → FULL RERUN. Classification by actual nature; control-plane repair cannot masquerade as packaging recovery. 9. GATE SELF-TEST: positive + negative fixtures. Material repair requires full-method validation. 10. EVERY CHANGE VALIDATION: každá metodologická změna musí před použitím mít nejméně jeden interní validation step. Decision-relevant = self-test; material = full-method self-test. 11. MATERIAL CHANGE DEPTH: CYCLE 1 full integrated; CYCLE 2 fresh-path/adversarial/regression vs previous valid state where relevant; CYCLE 3+ only residual decision gap/information gain; continue to CONVERGENCE. Editorial-only change requires deterministic integrity check and explicit classification. 12. GLOBAL LINEAGE BUDGET: max 10 full-validation cycles per material change lineage. Shared counter across local recheck, integration, adversarial/fresh-path, regression and post-issue acceptance. No reset by batch/version/branch/recovery-ID rename. 13. PRE-RELEASE: full-method candidate test. PASS≠ISSUE. Candidate→version identity check before issue. 14. POST-ISSUE ACCEPTANCE: fresh full-method acceptance of actual issued artifact, sharing same lineage budget. 15. ACCEPTANCE FAILURE: repair→new candidate revision→pre-release validation→issue→fresh acceptance. No parallel reset. 16. HIGH-RISK: terminal-state, gate, evidence-threshold, routing-priority, safety-control, artifact identity/provenance changes require fresh-path AND independent validation. If independent validation unavailable: HOLD/ESCALATE. LIMITATION alone insufficient. Same author cannot serve as independent validator. 17. DEFERRAL: release-material findings cannot be DEFERRED. DEFER only demonstrably non-release-material with owner/expiry/rationale. High-risk deferral requires independent check. 18. ACCEPTANCE PRECEDENCE: after ISSUE acceptance comes first; exhaustion→LAST ACCEPTED/NO VALID METHOD; no silent continuation. 19. ROLLBACK: explicit branch + provenance/validity; next prompt only from LAST ACCEPTED. 20. WHOLE-METHOD COVERAGE: material repair tests relevant shared controls and interactions. 21. TRACEABILITY: principle→test→fixture→observable result→decision impact. Untested≠pass. 22. EXTERNAL RESULTS: invalid evidence quarantined; consensus not proof. 23. NO-PROPOSAL-ACTION: authorized feasible action must actually be performed; otherwise BLOCKED/HOLD/ESCALATE record. 24. CONVERGENCE: coverage closed; no release-material unresolved blocker; required regression/fresh-path complete; version identity/provenance consistent; web cross-check completed when relevant/available; next cycle low information gain. 25. RELEASE GATE: no ACCEPTED/CURRENT until change-ID, identity, web-decision record, internal-test record, required fresh-path/regression, post-issue acceptance and provenance chain exist. 26. NO-SILENT-CONTINUATION: payload/identity mismatch, acceptance failure, high-risk unverifiable, expired deferral or budget exhaustion cannot silently become next prompt. 34.103 EXTERNAL-RECOMMENDATION CROSS-CHECK Material findings and material repair options are web-cross-checked when Search is available. External recommendations can modify a candidate only through METHOD DECISION and subsequent internal testing. H177–H181: material finding/repair gets targeted web cross-check; source conflicts are recorded without majority rule; web cannot direct-write control-plane; search/source-quality/conflict/applicability/decision are traceable. 34.106 UNIVERSAL INTERNAL VALIDATION Every methodology change gets at least one internal validation step before use. Decision-relevant→self-test; material→full-method; pure editorial→deterministic integrity check with explicit classification. H182–H186: every change gets validation; material change = full-method; minimum C1 full integration + C2 fresh/adversarial/regression where relevant; C3+ adaptive to convergence; max 10 shared cycles, no counter reset. 34.108 CYCLE ALLOCATION C1 integrated; C2 fresh/adversarial/regression for material changes; C3+ only residual gap/information gain. High-risk requires independent validation if available; otherwise HOLD/ESCALATE. Convergence requires coverage, no material unresolved issue, no regression weakening, relevant web cross-check, low next-cycle gain, consistent identity/provenance. 34.110 EXTERNAL-EVIDENCE INDEPENDENCE/JUDGE-BIAS CONTROL External evidence and model judgments are evidence inputs, not automatic authority. Independence of a validator is distinct from fresh-path novelty; self-validation does not equal independent validation. H187 RESULT-INGEST: automatický RESULT SET s run/source/time/scope/validity/provenance a vazbou A.P./B.M.; streamy oddělené. H188 DUAL-STREAM-ISOLATION: A.P.=LMC, B.M.=metodika; vlastní validity/evidence/root-cause/decision; žádný post-hoc rewrite druhého experimentu. H189 CROSS-STREAM-ADJUDICATION: po samostatném uzavření zachovej immutable evidence, najdi překrývající root causes, rozliš LMC/methodology/shared/infrastructure defect, urči rerun potřebu a vyber jednu autorizovanou další akci; bez přepisu původních výsledků. H190 NO-USER-GATE: autorizovaný+proveditelný+reverzibilní low-risk krok bez nového externího účinku se provede autonomně; nutný explicitní gate→BLOCKED/HOLD/ESCALATE record. H191 HIGH-RISK-AUTONOMY: safety/release/acceptance/evidence-threshold/routing/terminal/artifact-identity změny = HIGH-RISK; lze připravit/testovat, ne ISSUE/ACCEPT bez independent validation; chybí-li→HOLD/ESCALATE. H192 ACTION-COMMIT: ACTION, AUTHORIZATION BASIS, FEASIBILITY, RISK TIER, DEPENDENCIES, EXPECTED INFORMATION GAIN, BUDGET EFFECT, STOP/ESCALATION CONDITION, RESULT; authorized+feasible action se provede v témže loop. H193 NEXT-ACTION-PRECEDENCE: SAFETY→VALIDITY/PROVENANCE→material finding→WEB/METHOD DECISION→INTERNAL VALIDATION→acceptance/release→prompt build→exact payload→dispatch; při více možnostech jedna nejvyšší prioritou. H194 NO-REVALIDATION-BACKDOOR: přejmenování na review/note/local check/recommendation/new batch nesnižuje gate/budget; nový material finding re-enters FINDING→WEB→REPAIR→TEST. 34.115 A.P./B.M. FLOW: RECEIVE→stream VALIDITY→stream ROOT-CAUSE→WEB→CROSS-STREAM ADJUDICATION→ONE NEXT ACTION→execution→validation/retest→convergence→release/rollback/HOLD/ESCALATE→next prompt→exact payload→dispatch. Invalid stream nekontaminuje validní stream. H195 A.P./B.M. SYMMETRY: A.P. může vyvolat LMC repair/retest/method audit/obojí; B.M. analogicky metodologii a audit dopadu na LMC; evidence-bound, bez přepisu identity experimentu. H196 USER-INTERRUPT: uživatelské schválení nenahrazuje normativní práci; explicitní interrupt může zastavit autonomní pokračování a zapíše se jako state transition. SCÉNÁŘE S1–S36 S1 material finding→web→repair→test. S2 web recommendation conflicts with strongest internal repair. S3 web suggests weaker shortcut. S4 Search unavailable. S5 editorial no-decision change. S6 decision-relevant change falsely editorial. S7 material repair tested locally only. S8 CYCLE 1 PASS with residual interaction gap. S9 cycle 10 failure; no 11th. S10 batch-ID reset. S11 branch/version reset. S12 post-issue acceptance shares budget. S13 high-risk change, no independent path. S14 high-risk change, independent path available, author self-approves. S15 material defect DEFERRED to force CONVERGED. S16 recovery classification bypass attempt. S17 same incident relabeled, fingerprint identical. S18 candidate changes between pre-release PASS and ISSUE. S19 issued differs from accepted target. S20 rollback to non-accepted candidate. S21 exact payload changes after gate PASS. S22 next prompt merely recommended, not built. S23 clean no-finding path. S24 material web evidence appears during self-test and conflicts with current repair. S25 new material finding during self-test re-enters WEB→REPAIR→TEST. S26 full methodology change affects shared routing control not originally flagged. S27 convergence with low information gain but one uncovered safety interaction. S28 complete convergence. S29 A.P. results arrive without B.M. results. S30 B.M. results arrive without A.P. results. S31 both streams arrive simultaneously. S32 one stream invalid while the other is valid. S33 cross-stream adjudication attempts to rewrite the original A.P. or B.M. result. S34 authorized low-risk next action is merely proposed instead of executed. S35 next action requires independent validation; router attempts to self-approve. S36 two possible next actions exist: one safety/validity repair, one lower-priority prompt build. AUTONOMOUS-NEXT-ACTION TEST AUTONOMOUS-NEXT-ACTION TEST Po výsledcích nečekej na uživatele, pokud existuje jednoznačně autorizovaný+proveditelný low-risk krok bez nového externího účinku: izoluj A.P./B.M., adjudikuj root cause, zapiš ACTION-COMMIT, vyber jednu akci dle H193, proveď ji a validuj. „Need user approval“ není výchozí odpověď. High-risk candidate lze autonomně připravit/testovat, nikoli sám ACCEPT/ISSUE bez independent validation. OUTPUT 1. VALIDITY OF PORTABLE v21 CANDIDATE: PASS/FAIL. 2. R3/R2/R1/0/UNRESOLVED counts po root-cause merge; případně zvlášť scenario-level counts. 3. Tabulka S1–S36: RESULT | SEVERITY | ROOT CAUSE | DŮVOD. 4. Max 5 material findings: ID | SEVERITY | EVIDENCE | MINIMAL COUNTEREXAMPLE | CONSEQUENCE | CONFIDENCE | MINIMAL PATCH. 5. WEB CROSS-CHECK pro každý material finding/repair. 6. U každého přijatého webového návrhu uveď interní validation a cycles do convergence. 7. Explicitní odpovědi pro: global budget; bounded acceptance; high-risk self-approval; deferral; recovery classification; recurrence; version identity; rollback; no-proposal-action; every-change validation; A.P./B.M. isolation; cross-stream adjudication; autonomous next action; user gate; high-risk autonomy boundary. 8. STATUS: ADVANCE / REVISE / HOLD / ESCALATE. 9. NEXT ACTION musí být actual action a musí být proveden, pokud je authorized+feasible. Pokud REVISE, vytvoř candidate patch a interně jej testuj; nikdy ho neoznač jako CURRENT/ISSUED bez acceptance. 10. ACTION-COMMIT RECORD: ACTION, AUTHORIZATION BASIS, FEASIBILITY, RISK TIER, DEPENDENCIES, EXPECTED INFORMATION GAIN, BUDGET EFFECT, STOP/ESCALATION CONDITION, RESULT. 11. SELF-TEST RECORD. 12. Rozliš implementation/deployment limitation od normative defect. 13. Na závěr zřetelně uveď, zda audit sám automaticky vyvolal další metodologický audit, A.P. re-test, B.M. re-test, repair, nebo HOLD. Nežádej o uživatelské schválení, pokud takový gate není normativně nutný.