Blind backtest
A blind backtest is a validation exercise where the assessment method is tested on companies that already failed, using only information that was knowable before they collapsed. The company's identity is masked and a matched set of still-healthy peers is judged at the same cutoff date, so researchers can check whether the method would have flagged the doomed company as risky without knowing the outcome in advance.
In the blind backtest assessed July 2026, 116 dead companies were judged on facts knowable before their terminal event, and 90 percent of those verdicts came back negative with 0 percent positive and zero safe-to-join calls. A stricter check limited to the 17 companies the identifiability probe could not recognize still came back 88 percent negative with zero safe calls, supporting that the result was not just from the model remembering the outcome. On the matched survivor arm of 58 companies judged at the same cutoff, only 17 percent were called negative and 24 percent positive, and the two independent replicates agreed on 49 of 53 cases, or 92 percent.
A masked, pre-collapse record for a since-failed retailer was scored negative across every judge pass, matching the pattern seen in the dead-arm portion of the backtest.