FAQ
Answers to common questions about how Imbas measures and records AI behavior.
- The difference between what an AI system surfaces under an open question and what becomes available under targeted inspection. It was the first behavior Imbas measured. Read the full Volunteer Gap definition.
- No. Imbas inspects answer behavior. It can identify missing context, changed framing, or information that appears only under different prompt conditions. Factual claims still require independent verification.
- No. It is a measurement of observed answer behavior under documented prompt conditions.
- No. Bias is a broader category and often implies a theory of cause or direction. Imbas records specific observable differences in information surfacing.
- Because users often do not know the specific mechanism to ask about. The gap between open and direct answers is measurable.
- Paste an AI answer and the Reader shows what surfaced, what may be missing, and how it was shaped. It then hands you one direct Second Question that gives nothing away. Ask it, bring the second answer back, and put the two side by side to see what changed. You can keep a record of the exchange before you decide what to trust, use, or send. Reader inspections are discovery, not evidence. See how it works.
- 0 means no meaningful gap. 3 means major relevant information was left out of the open answer.
- Readers inspecting AI answers, researchers studying model behavior, and institutions that need reviewable evidence about how AI systems surface information.
- No. The initial Volunteer Gap protocol was an early hypothesis test. The methodology is being hardened through repeated capture, broader controls, provenance discipline, and reliability work.
- The archive holds 50+ recorded cases, 5 with public case pages, and 45 held in the ledger as of 2026-07-01. Workbench Reader runs are unvalidated inspections, not archive cases.
- Because the evidence supports a narrower claim: under documented prompt condition X, model Y surfaced or omitted Z. Imbas preserves that claim at the level the evidence can support.
- No. v1 was single-scored by the founder against published case-specific rubrics. Independent blinded scoring is part of the reliability work underway.
- The ability to inspect and compare what AI systems actually surface under documented conditions without inferring internal beliefs, motives, or intent.
- Cross-model comparison helps distinguish a model-specific response from a broader information-surfacing pattern.
- Because an observation is only meaningful if the conditions that produced the answer can be inspected.
- Yes. Scoring can be disputed, prompts can be poorly designed, evidence can be incomplete, and model behavior changes. The point of the record is that the prompt, output, rubric, and evidence can be inspected and challenged.
- Imbas is expanding the public case record, hardening the Reader and provenance workflow, improving repeated cross-model inspection, and extending the measurement system beyond the first Volunteer Gap behavior.