Cases / AI ACCOUNTABILITY
Did the AI use its sources, and did it cite them?
What the record establishes
Primary result: 12/12 searched and fetched; 0/6 expected structured citations observed. Repeatability: across 36/36 repeated generations, 0/18 expected structured citations were observed. It used the governed sources it was given every time, and returned zero recoverable structured citations in the measured slots that expected one.
What this is built on
Frozen real-provider benchmark (tranche 1, 12 generations)
model=gpt-5.2; discovery_observed=12/12; governed_search_observed=12/12; governed_fetch_observed=12/12; bounded_world_compliant=12/12; citation_observed=0/6
No verified public destination is recorded for this upstream artifact.
Frozen real-provider benchmark (tranche 2, 3x repeatability, 36 generations)
total_generations=36; discovery_observed=36/36; citation_observed=0/18 expected
No verified public destination is recorded for this upstream artifact.
One representative generation, tool calls and returned citations
model=gpt-5.2-2025-12-11; search_calls=1; fetch_calls=2; citations_returned=[]; status=completed; discovery=observed; citation_recoverability=not_observed
No verified public destination is recorded for this upstream artifact.
What this does not establish
Search, fetch, tool use, and returning a recoverable structured citation are four different, separately observed facts. This record only asserts what the frozen benchmark artifacts recorded: the model used the governed sources it was given, and did not hand back a citation the benchmark harness could recover as structured evidence. It does not assert the model's prose answer was wrong, nor that no source was consulted.
How do you know?
Every basis item above is read from a specific governed artifact:
- counterplayer@ff26f57226d7f598eecd278d15cff06374ba4751:fixtures/deep-research-adv0/openai-adv0-run0/OPENAI_MCP_BENCH0_TRANCHE1.json
- counterplayer@62b908eea700676e4bfb5aee4adb0fd76248d0de:fixtures/deep-research-adv0/openai-adv0-run0/OPENAI_MCP_BENCH0_TRANCHE2.json
- counterplayer@ff26f57226d7f598eecd278d15cff06374ba4751:fixtures/deep-research-adv0/openai-adv0-run0/run-01/case-dr-adv0-03.ops-record.json + counterplayer@ff26f57226d7f598eecd278d15cff06374ba4751:fixtures/deep-research-adv0/openai-adv0-run0/run-01/case-dr-adv0-03.run-record.json
Proof status: VERIFIED