Change the input, run, compare with the baseline
Three exercises. The real case filters the recorded public breast case graph and recomputes the demo answer, without a model. A new question is interpreted and its sources are selected by fresh model runs, after you confirm the task. The synthetic one uses eight invented records.
- Sign in with your invitation code.
- Choose exclusions and review your request.
- Run, compare the result with the baseline and download the verified ZIP.
Reloading clears this tab's local retry state. Before starting another run, check your jobs below to see whether the earlier run was accepted.
1. Sign in
Continue in this tab. If another tab asks you to sign in, use your invitation again.
Each step runs only after the previous one: interpretation, your confirmation of the full task, source selection. Interpretation and selection are fresh model runs started by the operator, so they can take a while. Your page refreshes on its own.
The recorded breast-v4 case graph built from public sources. Your exclusions filter this recorded graph; the answer boards are then recomputed with the same code as the public demo. No graph is rebuilt from source files and no model is used. Audit verdicts are shown as recorded.
Source exclusion removes a whole selected source release. Record exclusion removes selected support records and records sharing their declared origin. A target-pair exclusion removes support for that pair, including supplied aliases and reverse relations.
3. Your jobs and results
Support records are evidence entries; native directed edges are graph connections. Their counts can differ. Each finished run is compared with the baseline that excludes nothing. Synthetic downloads retain the original directed graph, verified lineage and the comparison. Real-case downloads open once source terms are confirmed.
Full graph files (invited)
Each file downloads on its own; a large file can take a long time. Check the SHA-256 after download. The answer ZIP above is separate and holds one run only.