Evaluation Run 20260923T020538

Rubric 2026-09-23a · typesafe/jev-1.13 · 431 of 431 answered · 22.9s · $0.0432

No entry file was written. Results live in this run's ledger until Phase 4 adds the write path, after calibration sets the thresholds.

Per-dimension distribution

DimensionTypeDistributionMean confidenceAgrees
answer_key_correctnoulmean 0.94 · level 1 holds 429/431 FLAT——
connection_to_themescoremean 0.92 · level 1 holds 415/431 FLAT0.02—
difficultyscoremean 1.22 · level 1 holds 247/4310.42308/431
distractor_pullscoremean 1.41 · level 2 holds 214/4310.51—
engagementscoremean 1.36 · level 2 holds 214/4310.64—
one_defensible_answernoulmean 0.84 · level 1 holds 298/302 FLAT——
options_scan_cleanscoremean 1.58 · level 2 holds 303/4310.55—
passage_realismscoremean 1.35 · level 1 holds 177/3020.55—
reads_in_one_passscoremean 1.85 · level 2 holds 420/431 FLAT0.78—
sat_choice_conventionsscoremean 1.83 · level 2 holds 408/431 FLAT0.77—
sat_stem_conventionsscoremean 1.83 · level 2 holds 406/431 FLAT0.85—
sectionchoicereading 178/4310.99431/431
single_movenoulmean 0.74 · level 1 holds 364/431 FLAT——
skill_codechoiceinfo-and-ideas 133/4310.95405/429
stem_self_containednoulmean 0.95 · level 1 holds 431/431 FLAT——
survives_speednoulmean 0.70 · level 1 holds 418/431 FLAT——
tested_construct_is_satnoulmean 0.96 · level 1 holds 431/431 FLAT——

A dimension marked FLAT holds more than 80% of the run on one level. That is either a clean bank or a dimension reporting a constant, and only a negative control tells them apart.

Run

Served by{'typesafe/jev-1.13-20260917 via TypeSafe': 431}
Input tokens1,028,567
Failed0
Preflight notes0
Skipped entries0