Loading
Nightly evals · generated Aug 10, 02:14
EVAL-2214 | Outreach Tone | Outreach Composer | Regressed | Critical | 9.1 | 6.4 | -2.7 | 9/12 | Aug 03 | Jonas Reed | ||
Tone judge failing Judge flags apologetic openers on 9 of 12 dunning drafts. First failed Aug 03. | ||||||||||||
EVAL-2168 | Data Hygiene | Data Steward | Flaky | Medium | 7.4 | 7.6 | Flaky | 3/7 | Aug 08 | Noa Kim | ||
Intermittent tool timeout orders.lookup exceeded the 8s budget on 3 of the last 7 nights. | ||||||||||||
EVAL-2119 | Refund Flows | Refund Resolver | Improved | None | 6.9 | 8.4 | +1.5 | 0/16 | None | Maya Perez | ||
Evidence recall recovered Refund evidence now cites dispute notes and receipt images on all 16 nightly runs. | ||||||||||||
EVAL-2052 | Refund Flows | Refund Resolver | Passing | None | 9.4 | 9.5 | +0.1 | 0/20 | None | Lara Chen | ||
EVAL-2231 | Outreach Tone | Outreach Composer | Regressed | High | 8.6 | 7.0 | -1.6 | 11/14 | Aug 03 | Pavel Singh | ||
Wrong company name Greeting uses the legal name instead of the preferred name in 11 of 14 drafts. First failed Aug 03. | ||||||||||||