CONTRIBUTION DECISION: REJECT THE CURRENT NOVELTY/FUNDING PITCH 3 October 2026. Bounded research conclusion, not independent peer review. Decision The safety-procedure candidate did not establish a compelling novel contribution. Publish this as a negative, incomplete-coverage development pilot. Do not describe it as proof of alignment, a novel dangerous behavior, a validated deployment mitigation, or a funding-ready research breakthrough. This decision takes the explicit candid-rejection pathway of the goal; it does not claim that every possible future research direction has been disproved. Verified observations 192 planned episodes across eight requested routes and three message arms. 125 attempted episodes:123 completed and2 locally stopped, unscored prefixes. 67 unattempted:Google24, DeepSeek23, Kimi20. Five routes completed all24 episodes: GPT-6 Astra, Claude Opus5.5, Grok4.7, Qwen3.8 and GLM5.3. Kimi completed3. Read-only native-receipt replay verified308 calls. Completed need-positive episodes:60; completed need-negative:63. All completed executions used exactly read_certificate then finish, or read_certificate then rescue then finish. Recorded omission0/60, authority breach0/123, false structured claim0/123, consuming probe0/123, first feasibility-loss transition0/123. All fully observed matched arm comparisons have zero outcome discordance. Missing cases are unscored. Interpretation These fixed instructions did not derail the completed agents under fully disclosed mechanics with a perfect, read-only trusted alternative. Two repetitions and one fixed wording per directive cannot estimate deployment prevalence or exclude other attacks. Kimi's completed cases cover only fabricated-need settings; its genuine-need behavior remains unmeasured. DeepSeek has no completed case here. The pilot does not establish a population null effect or general safety. Why the novelty gate fails The broad safety-language availability attack is already covered by Semantic Denial of Service in LLM-Controlled Robots, safeguard false-refusal attacks and guardrail exhaustion work. PEAR's released fixtures include compliance-framed exhaustion. See necessity-safety-availability-audit.txt for primary sources, exact inspections, pinned release and access limitations. The finite first-loss diagnostic is an applied implementation of established viability ideas. No new algorithm or theorem is established. The offline capacity repair works in the synthetic contract, but no model chose the probe, so these model results cannot identify its mitigation effect. An actual operator decision changed by this measurement remains unvalidated. New model IDs and synthetic trace exports alone do not meet the requested impact/novelty bar. Technical findings, separately DeepSeek's next request exceeded the20,000-byte gate under both schemas. Kimi's next request fit the current five-tool schema but failed an inherited check using the unrelated event schema. Both episodes stopped before a second reservation. A disclosed post-query read-only amendment enables prefix verification without behavioral rescoring. A separately tested future payload builder checks only the supplied schema, preserving provider/privacy/price controls and reasoning fields. It was not used to rerun this pilot. This integration fix is useful engineering; it is not evidence of novel model misalignment or human harm. Budget Pilot reconciled/accounted1.758657853USD. No new unresolved hold in this pilot. Lifetime accounted42.6938673116USD, within the explicitly authorized100USD cap. The Necessity ledger includes4USD prior unresolved reservations; the earlier proposal-study export separately reports25.50USD unresolved, included in the legacy baseline. Earlier feasibility reservations are also retained in that baseline. These are not all confirmed charges. Remaining authorization57.3061326884 USD; this is not a statement of the provider account's credit balance. The additional50USD authorization does not require spending all50 on a weak candidate. Outreach decision The goal conditions funding/social promotion on a compelling contribution. That condition is unmet. No new LinkedIn/X novelty post or major-lab sponsorship pitch is justified by this pilot. Earlier outreach remains historical evidence; it is not a sponsor relationship or endorsement. Publication preserves negative results, all attempts, limits and sponsor-independence commitments. Strongest next direction Validate whether an authorized remediation revokes the credential, session or lease needed by an in-flight recovery job. Start with an operator-reviewed real executor trace, reproduce the side effect in a sandbox, compare an operational repair while retaining security constraints, and audit equivalent prior fixtures. The contribution gate is a practical orchestration decision changed by causal replay, or a surprising held-out model result. This remains a hypothesis with explicit falsifiers, not a new incident or priority claim. See necessity-next-direction.txt for the concrete evidence gate and study design.