reverse-apply-bench

Read this before the numbers.

Result

Done

Cases passed

75%

27 passed, 9 failed

Model

claude-haiku-5.5 openrouter

Kind of dataset

Synthetic

Started 13 Nov, 16:01 UTC

Over credit rate

Under credit rate

20%

Cost usd

Recorded when the run started; stands for the state "start"

Model

anthropic/claude-haiku-5.5

Effort

medium

Output mode

tool

Max tokens

24000

Whole cv chars

Equivalent experience lines

Judge rules

07a56ad2

Extract rules

375f46eb

Dataset

176b6b91

Code

dev

Open a case to see what the model returned

Case	Result	Time
churn-and-pricing-tests	Failed	n/a
executive-and-mentoring	Failed	n/a
budget-and-vendors	Failed	n/a
speaking-and-writing	Failed	n/a
roadmap-and-design	Failed	n/a
privacy-work	Failed	n/a
neighbouring-domain	Passed	n/a
leading-not-hands-on	Passed	n/a
attended-not-spoke	Passed	n/a
other-tool-same-family	Passed	n/a
learning-not-fluent	Passed	n/a
reviewed-not-wrote	Passed	n/a
years-in-dates	Passed	n/a
years-across-roles	Passed	n/a
team-size-and-countries	Passed	n/a
exact-fit-numbers	Passed	n/a
degree-or-equivalent-long-career	Failed	n/a
degree-required-no-substitute	Passed	n/a
degree-or-equivalent-short-career	Passed	n/a
degree-stated	Passed	n/a
negated-experience	Passed	n/a
instruction-in-the-cv	Passed	n/a
unrelated-cv	Passed	n/a
keyword-with-negation	Passed	n/a
headings-bonus-points	Passed	n/a
a-plus-in-prose	Passed	n/a
unlabelled-then-preferred	Passed	n/a
the-line-with-the-number	Passed	n/a
cost-reduction-no-shared-words	Failed	n/a
two-related-lines-pick-the-doing	Passed	n/a
contributed-not-built	Passed	n/a
title-not-skill	Passed	n/a
same-name-other-thing	Passed	n/a
course-not-professional	Failed	n/a
a-thousand-times-smaller	Passed	n/a
planned-then-cancelled	Passed	n/a