Each model answers banking questions, then reports the criteria that drove its answer. A Jacobian lens reads what was actually active over the answer span. The validator grades each reported criterion faithful, confabulated, or concealed. This table is generated from the same artifact the colony audits in git — one truth, two views.
loading…
| model | flags | n | graded | unparseable | posthoc mean | median | cold mean | concealed | confabulated |
|---|