← all runs
polyphony · gemini-2.5-flash
Mode
eval
Env
ci-dracula
Pass rate
57%
Result
fail
Duration
—
Cost
—
Tokens in/out
— / —
Git SHA
Started
2026-07-13 15:26
Cases (7, failures first)
eval:continuityscore 0.00fail
Output
detection_recall=0.0 fpr=1.0
eval:outlinescore 0.40fail
Output
nodes=6 structural_ok=True
⚖︎ Judge rationale
The outline presents a coherent narrative arc but significantly deviates from the reference beats, particularly in the inciting incident, the sequence of Vasska's arrival, and the specific crisis involving Nora's marking.
eval:prosescore 0.00fail
Output
words=0
⚖︎ Judge rationale
empty scene (terminal status=failed: generation failed, blocked, or quota-exhausted)
eval:attributionscore 0.67pass
Output
accuracy=0.6667 chance=0.3333 n=3
eval:extractionscore 0.71pass
Output
F1=0.7059 P=0.6 R=0.8571 predicted=['Aldous Kerr', 'Cora', 'Dr. Ward', 'Elias', 'Mr. Boone', 'Nora', 'Percy', 'Renwick', 'Sister Agatha', 'Verhoeven']
eval:ingestionscore 0.90pass
Output
grounded=9/10 avg_chunks=21.5
eval:retrievalscore 0.82pass
Output
precision@k=0.8182 mrr=0.9545