← polyphony · eval · dracula · ci
polyphony · gemini-2.5-flash
✗ fail↓ regression 83% → 25%Pass rate
25%
Cases
4 (3 failed)
Mode
eval
Suite
dracula
Env
ci
Started
2026-07-27 10:22 UTC
Duration
—
Cost
—
Tokens in/out
— / —
Commit
—
Branch
—
Source
—
Labels
| legacy_env | ci-dracula |
Compared with the previous run
Baseline 2026-07-20 09:40 UTC · 83% → this run 25%
✗ newly failing · 2
- eval:extraction
- eval:ingestion
cases added / removed
- + eval:outline
- − eval:attribution
- − eval:continuity
- − eval:prose
Cases (4, 3 failed, failures first)
eval:extractionscore 0.00✗ fail
Output
F1=0.0 P=0.0 R=0.0 predicted=[]
eval:ingestionscore 0.00✗ fail
Output
grounded=None/0 avg_chunks=None
eval:outlinescore 0.00✗ fail
Output
nodes=0 structural_ok=False
⚖︎ Judge rationale
The response is an empty outline, failing to recover any of the reference beats or provide a coherent narrative structure.
eval:retrievalscore 0.82✓ pass
Output
precision@k=0.8182 mrr=0.9545