polyphony · claude-opus-5-5
✓ passMetrics
| outline.cast_fidelity | 1 |
| outline.score | 0.792 |
| outline.unknown_rate | 0.033 |
Labels
| cadence | on-demand |
| experiment | w15-pre-regrade |
| harness_sha | b46d695794a77670e2a34b6d9a4520e1ccb00850 |
| judge_model | claude-sonnet-5-5 |
| judge_provider | anthropic |
| judge_self_graded | True |
| repeat | 1 |
| trigger | workflow_dispatch |
Compared with the previous run
No earlier complete run in this series (polyphony · eval · cel · ci · claude-opus-5-5), so there is no baseline yet.
Cases (1)
nodes=6 structural_ok=True
R1: All principals (Milo, Zara, Edric/CEL, Prava, Old Fen, Halvard Bell) appear with exact names, the CEL is read correctly, Edric calls Milo 'Mr. Voss' without contractions, and no protagonists are invented; the small deductions are for the Retention team being shown only via Halvard, Milo's mild early | R6: The outline faithfully restates the premise, lands each act turn in its final chapter (the discrepancy in ch.2, the Delete Draft decision in ch.4, Edric's filing and the honest email in ch.6), escalates steadily, and meets the six-chapter request, but a few beats are slightly static (e.g., 'Off Sche | R3: All six chapters realise every planned beat in the planned order, and each beat's change lands (for example the report is drafted in ch.3 and deleted in ch.4, and the ledger is discovered in ch.4 and handed over in ch.5). The only deviations are minor unplanned additions: Edric's offer of the climb,