Add signalbench (In-Band Signal Compliance) community eval
#42
by thamilvendhan - opened
No description provided.
Withdrawing this community eval. On re-audit, 28 of the 75 items for this model were HTTP 404 provider-routing errors on my evaluation side (OpenRouter: "no endpoints support tool use"), not model behaviour, so the tool-graded families (access-deny, bot-policy) did not actually run. The posted score therefore understates the model. Closing rather than leaving an inaccurate number on the model page; I may resubmit a corrected text-only-subset score (clearly marked n=47, not comparable to full-suite results). Apologies for the noise.
thamilvendhan changed pull request status to closed