movies-drama-imbalance-sentmismatch-text

Task gallery / Machine learning / Version 1

Reference solution (oracle): Pass
No-op resistance check: Pass
QA verdict: Not recorded
Aggregate eligibility: Eligible version

The task

Take the review pile in [local evidence] and predict the star rating of each Movies and TV drama review from its text. The reviews are gzipped JSON lines carrying the rating, the body, and the title; the held-out reviews you score arrive without their rating, which is recorded nowhere in the data. Ratings lean hard toward five stars and the one and two star classes are rare, but those low scores are the ones that matter m…

Excerpt only; not a complete executable task specification. Task ID task_25583b5ca1fa4cc3 · version 1. Selection status: not fully recorded.

Results by model and effort

This exact task version only. Every rate is passes / valid runs with a Wilson 95% interval; n≤2 remains counts only. Macro aggregation does not apply within a single task. Baselines are summarized in the quality gates, not ranked. A passed no-op resistance check means that doing nothing did not solve the task.

Current evidence

Model / effort	Verifier result [95% Wilson]	Diagnostic labels / valid	HARNESS_ERROR · excluded	Outcome classes
Claude Opus 5 · default (unverified)	3/5 · 60.0% [23.1%, 88.2%]	5 / 5	0	GOOD_FAILURE: 2 BAD_FAILURE: 0 GOOD_SUCCESS: 3 BAD_SUCCESS: 0

Legacy evidence

Model / effort	Verifier result [95% Wilson]	Diagnostic labels / valid	HARNESS_ERROR · excluded	Outcome classes

Difficulty, with uncertainty

Selected-best difficulty: 40.0% [11.8%, 76.9%]. This is one minus the best measured current-row pass rate among rows with at least five valid runs.

The maximum is selected from the observed rows, so the estimate is optimistic. Its Wilson interval is not adjusted for model selection. “Hard” requires qualifying GOOD_FAILURE evidence; the threshold flag additionally requires the interval and recorded quality gates to meet the documented rule. This is not independent validation. Threshold flag: no.

Every recorded run

7 recorded trials, including controls. Diagnostic labels do not replace verifier reward. Labels and quality checks are source declarations, not independently attested here. Internal evidence remains private; a missing public URL is shown explicitly. Records with invalid or absent rewards are not scored.

Run / model	Date / source	Reward / eligibility	Diagnostic class / source	Resources	Evidence
`trial_85b85287cb284685` nop@@	Oct 11, 2026 console	0 · valid	No diagnostic label derived	Tokens: — Turns: — Cost: —	Not publicly available
`trial_a02a73adadef486a` oracle@@	Oct 11, 2026 console	1 · valid	No diagnostic label derived	Tokens: — Turns: — Cost: —	Not publicly available
`trial_14067e3dd7f541e2` Claude Opus 5 · default (unverified)	Oct 11, 2026 console	0 · valid	GOOD_FAILURE frozen-classifier	Tokens: — Turns: — Cost: —	Not publicly available
`trial_435a8333c8c94575` Claude Opus 5 · default (unverified)	Oct 11, 2026 console	0 · valid	GOOD_FAILURE frozen-classifier	Tokens: — Turns: — Cost: —	Not publicly available
`trial_1fc6b37a4cb44907` Claude Opus 5 · default (unverified)	Oct 11, 2026 console	1 · valid	GOOD_SUCCESS frozen-classifier	Tokens: — Turns: — Cost: —	Not publicly available
`trial_004649ce299d449c` Claude Opus 5 · default (unverified)	Oct 11, 2026 console	1 · valid	GOOD_SUCCESS frozen-classifier	Tokens: 45,356 Turns: 26 Cost: $2.4871	Not publicly available
`trial_3a2dcb51777b43f6` Claude Opus 5 · default (unverified)	Oct 11, 2026 console	1 · valid	GOOD_SUCCESS frozen-classifier	Tokens: 35,357 Turns: 19 Cost: $1.8202	Not publicly available