Tracer
← polyphony · eval · frankenstein · ci

polyphony · claude-opus-5-5

✓ pass≠ not comparable
Pass rate
100%
Cases
1
Mode
eval
Suite
frankenstein
Env
ci
Started
2026-10-24 09:49 UTC
Duration
—
Cost
—
Tokens in/out
— / —
Commit
Branch
main
Source

Metrics

prose.score0.691
prose.words3,565

Labels

cadenceon-demand
experimentw16-r2-1
harness_sha47632720c43762568ae9087c9bc6abedfe3aee3f
judge_providergroq+anthropic
judge_self_gradedTrue
repeat1
triggerworkflow_dispatch

Compared with the previous run

Baseline 2026-10-24 06:21 UTC · 100% → this run 100% · code diff ↗

cases added / removed
  • + eval:prose
  • − eval:attribution
  • − eval:continuity
  • − eval:extraction
  • − eval:ingestion
  • − eval:retrieval

Cases (1)

eval:prosescore 0.69✓ pass
score=0.691words=3,565queued=4deferred=0n_scenes=4r2_lines=25cast_named=1prose_chapter=1r2_applicable=0.750cast_from_beat=1r1_capped_scenes=0estimate_over_actual=0.772beat_seam_max_overlap=0scene_seam_max_overlap=0
Output
words=3565 path=planned rubrics={'R1': 0.9275, 'R2': 0.31, 'R3': 0.89, 'R4': 0.81, 'R5': 0.825} seams=0/0
⚖︎ Judge rationale

scene 1 R1: Both principals, Rowan Merrow and Hollis Crane, appear exactly as planned and behave in line with the brief (Crane as a plain-spoken, worried reporter, Rowan as the ambitious captain), and the only other figures are unnamed crewmen, though the canon gives Crane only a bare description so there is li | scene 1 R3: All five beats are realized in order with their changes landing (the journal and groaning hull, the dream of an open sea with the isolation note, Crane's report and Rowan's hollow reassurance, the deck scene with the realization, and the return to the desk with the wager and the unwritten word), wit | scene 1 R4: The scene keeps close third-person past tense, with journal entries in italics in first person and in-voice, and an exalted-then-uneasy tone; the sentences are concrete and varied, there are no labels, repetitions, or summary-in-place-of-scene, though there are a few minor stumbles (e.g., 'Every one | scene 1 R5: The scene follows the brief, beats, cast, setting and POV with no contradictions in names, objects, timeline or who knows what; the only wrinkle is the unestablished detail that the men signed for 'a share of the oil', which hints at whaling and is not in canon. | scene 1 R2: not applicable: under two cast speakers with voice samples have attributed lines | scene 2 R1: All three beat-cast principals (Rowan, Crane, Emeric) appear with exact names and behave per canon and plan, with only unnamed functional crew and a mentioned surgeon invented. | scene 2 R2: The five lines are terse, modern and generic ("We're bound north", "Rope, And the ladder"), with no ornate Georgian cadence from Rowan, no feverish anguish from Vandel and no chatty warmth from Crane, so attribution would rest on context rather than voice, and an unfilled "[name]" placeholder and st | scene 2 R3: All four beats appear in order with their changes landing (the glimpse, the dawn drift, the rescue and question, the refusal until told 'north' and the realization of pursuit), but small deviations lower it: Rowan wakes and is roused by Crane (fine), yet the narrator states the pursuit conclusion ex | scene 2 R4: The prose holds close third-person past tense from Rowan's POV throughout, with an eerie tone, concrete imagery and varied rhythm, and it shows no beat labels or repetition; minor slips include the 'Crane' vs. 'Hollis' naming, a few generic phrases ('grey on grey', 'wall of fog'), and a slightly abr | scene 2 R5: The scene follows the canon names, the planned beats and the prior text (cabin, journal, the yesterday/today timeline, Crane as Hollis Crane, Rowan as captain) with no major contradictions; only minor wobbles remain, such as 'most' dogs dead against a single 'surviving dog', 'tide' on a polar sea, a | scene 3 R1: Both beat-cast principals, Rowan and Emeric, appear and act in line with canon and the plan, with exact names, no invented characters beyond a passing, unnamed surgeon, and only minor quibbles such as Crane being absent (not required here). | scene 3 R2: The lines can be attributed mostly through context (who addresses whom, who is rescued), not through diction, and both characters speak in a clipped, aphoristic register that contradicts their samples (Rowan's ornate, effusive epistolary voice and Vandel's anguished, gothic cadence), so they differ | scene 3 R3: All five beats appear in the planned order with their changes landing and no major unplanned events; minor weaknesses are the thin 'quoting poets' in beat 2 and beat 5's recurring pattern being shown through a single exchange. | scene 3 R4: The scene keeps close third-person limited POV from Rowan in consistent past tense, with a tender, ominous tone, concrete imagery, and varied sentences, and it follows all five beats without labels or prefixes; minor deductions for a few slightly generic phrases ('set hard as the pack'), some brisk | scene 3 R5: The scene respects the names, the sledge and dogs, the surgeon, the ruined hands, the northwar