Update — the entry is unchanged; this is a place to check.
Our submission remains 02969cb46a86c44a7b411ff98d9e05c4f6fd3c93. main has not moved, and nothing has been pushed since the deadline.
We kept building, and one result is worth a minute if you have one. Sidq's verdict was always meant to be reproducible; it is now re-derivable without Sidq. A second implementation of the decision — 99 lines of jq, no Python, no DataHub, no network, sharing no code with the engine — recomputes the published BLOCK from the verdict's own evidence, and a build failure is what disagreement produces. It runs in your browser at sidq2.mlki.app, alongside three other read-only demonstrations, and the whole repository is downloadable there with every branch.
That page also carries what we could not do: 99% precision is reachable and costs 20 points of recall, and at our corpus size it is not statistically distinguishable from 97%. We publish the confidence intervals rather than the headline.
Nothing there is part of the entry, and none of it is asked to be.
Log in or sign up for Devpost to join the conversation.