Every row below is verified against the source PDF, page and table cited. Nobody staged these: they're what's actually sitting inside the papers this agent reads.
Sanity Context's own conflict detector raised 14 issues while building this Knowledge Base, across several rebuilds. 6 of them pointed to real errors, covering 3 of the 10 inconsistencies below: the Bori swap, the Choba curve label, and, first caught on a later rebuild, the Etche Odufor curve label. Same-fact detection is exactly right for a case where two documents state different numbers for the same station.
The other 7 are never stated as a contradiction in words: a label that doesn't match its own data, or a depth that doesn't equal the layer above plus its own thickness. Those need a computed check, not a text comparison, so this agent's code catches them at query time instead.
Found by: Context = Sanity Context's build-time conflict detector · Code = this agent's own checks, computed live from the raw layer numbers
| # | Station(s) | What's in question | Verdict |
|---|