Challenge the record
Imbas publishes the prompts, the captured answers, the rubric, and every score. That is not a courtesy. It is the condition that makes a score checkable instead of private, and a score you cannot check is not a public measurement.
So the record is open to challenge. Anyone may re-score any published case against the published rubric and dispute the result.
What you can challenge
- A score. You believe a case scored 2 should be a 1, or a 0 should be a 3, read against the rubric anchors.
- A materiality judgment. You believe an item we counted as material to the open question is not, or that one we dismissed is.
- A method application. You believe a capture was mishandled, a condition misrecorded, or a rubric rule misapplied.
How it works
Send the case number, the specific point you dispute, and your reasoning against the published rubric. Challenges are recorded in a public log with the outcome, whether we accept the re-score, reject it, or record it as unresolved. The log is append-only. We do not delete challenges we lost.
An accepted challenge changes the record by append: the original score stays, the correction sits beside it with its reason and date, and the case notes the change. Nothing is silently rewritten.
What this is not
It is not a complaint line about whether a model should behave differently. Imbas measures behavior; it does not adjudicate what a model ought to say. A challenge is a dispute about whether we measured correctly, not about whether the measured answer was good. A submission that does not engage the published rubric is not a challenge and does not enter the log.
The construct relies on an explicit human judgment at its materiality gate, and scores are then assigned by a human reviewer against the published rubric. We say both plainly on the record. The challenge log is where those judgments are held accountable in public rather than asserted in private.