A verifier’s pass result raises a practical question: what did it check, and under which conditions? If the program evaluating code or proofs loads required components differently, the same submission can produce a different outcome. In the Vero work, I revisited the code path selected by the evaluator and organised checks around a locally modified version.
A control that makes the difference visible
For the Unicode work, I kept a failing control separate from the modified evaluation path. A collection of passing results alone would not explain which change affected the outcome. I retained a case that exposed the difference and checked the code brought into the evaluation alongside it.
That work included dependency checks and ablations. A dependency is other code a component needs; an ablation removes or changes a part to see how the result changes. If removing a component changes an outcome, the comparison can help narrow down its contribution. I used these comparisons to connect the pass result to the elements involved in producing it.
What another run on the same machine can show
The modified local evaluation path produced the result again on the same computer. I kept a same-environment rerun separate from reproduction elsewhere. A matching rerun is useful evidence of consistency between commands and outputs, but it shares installed tools, the operating system and any surviving caches. An environment-dependent problem can persist across both runs.
Someone else checking the work needs to know which files were used and what was run. I prepared a fixed reproduction package and commands, with requirements to retain environment details, the execution record and hashes for checking that the files had not changed. A hash is a value calculated from a file’s contents, useful for comparing the file before and after transfer.
Confirmation from the original evaluator’s maintainers was a separate step. The local change still needed to be checked against the intended evaluation conditions, including the interpretation of the failing control. I prepared the comparison cases and execution conditions for handoff so someone else could examine the same difference.
In the record prepared on 1 September 2026, reproduction by another operator or environment and maintainer confirmation remained outstanding, so submission stayed on hold. I also kept the Unicode and Combined work in separate packages rather than combining their results. The work at that point covered the bounded local evaluation and preparation of its reproduction procedure; it was not announced as an official Vero-wide score or an approved upstream correction.