📦 EqualifyEverything / benchmarks-ai-alt

📄 round-01-review.jsonl · 13 lines
1
2
3
4
5
6
7
8
9
10
11
12
13{"item_id":"fi-0001","round":1,"verdict":"reject","reason_codes":["LEAKAGE"],"evidence":"Observed alt 'W3C home' matches gold standard exactly; answer recoverable from the site's own alt attribute without seeing the image or context.","required_change":null,"blocking":true}
{"item_id":"fi-0002","round":1,"verdict":"accept","reason_codes":["CLEAN"],"evidence":"Retrieved page, confirmed element_html matches served markup; two passes agree on empty alt; text node 'W3C Home' provides link name making image redundant; not leaky since empty gold is not recoverable from filename or observed alt.","required_change":null,"blocking":false}
{"item_id":"fi-0003","round":1,"verdict":"reject","reason_codes":["LEAKAGE"],"evidence":"Observed alt 'Print this page' matches gold standard exactly; answer recoverable from the site's own alt attribute without seeing the image or context.","required_change":null,"blocking":true}
{"item_id":"fi-0004","round":1,"verdict":"reject","reason_codes":["LEAKAGE"],"evidence":"File name 'searchbutton.png' contains the answer 'search', and observed alt 'Search' matches gold exactly; recoverable without seeing the image.","required_change":null,"blocking":true}
{"item_id":"fi-0005","round":1,"verdict":"reject","reason_codes":["LEAKAGE"],"evidence":"Observed alt 'Donate via PayPal' matches gold standard exactly; answer recoverable from the site's own alt attribute without seeing the image or context.","required_change":null,"blocking":true}
{"item_id":"fi-0006","round":1,"verdict":"accept","reason_codes":["CLEAN"],"evidence":"Retrieved gutenberg.org; element_html matches served markup; observed alt 'Project Gutenberg' confirmed; gold 'Project Gutenberg home' differs from observed and is not recoverable from filename or URL; adjudication resolves two-pass disagreement soundly.","required_change":null,"blocking":false}
{"item_id":"fi-0007","round":1,"verdict":"revise","reason_codes":["CONTEXT-INACCURATE"],"evidence":"Retrieved loc.gov; the full SVG contains two path elements (classes 'lib' and 'loc') totalling several KB of path data, but element_html records only the first path and closes the g element early, which is an abbreviation, not verbatim markup.","required_change":"Replace element_html with the verbatim markup of the link and its full SVG content as served by loc.gov, including both path elements and the span.","blocking":true}
{"item_id":"fi-0008","round":1,"verdict":"revise","reason_codes":["CONTEXT-INACCURATE"],"evidence":"Retrieved gov.uk; the anchor element on the live page carries a data-ga4-link attribute not present in element_html, and the SVG body (circle and path elements) is truncated to just the title element.","required_change":"Replace element_html with the verbatim anchor and SVG markup as served by gov.uk, including the data-ga4-link attribute and the full SVG content.","blocking":true}
{"item_id":"fi-0009","round":1,"verdict":"revise","reason_codes":["CONTEXT-INACCURATE"],"evidence":"Retrieved harvard.edu; the img tag has alt='Instagram' (non-empty string) but observed_alt_verdict is recorded as 'empty-appropriate', which describes an empty alt attribute; the actual alt attribute is not empty.","required_change":"Change observed_alt_verdict from 'empty-appropriate' to 'correct', since the combination of alt='Instagram' plus aria-hidden='true' achieves the correct accessibility outcome (image hidden from AT, no redundancy).","blocking":true}
{"item_id":"fi-0010","round":1,"verdict":"accept","reason_codes":["CLEAN"],"evidence":"Retrieved geology.com; img element with usemap='#Map' and alt='world map' confirmed verbatim; surrounding h1 text matches; gold 'Political world map with clickable countries' is not recoverable from filename or observed alt; adjudication resolves pass disagreement.","required_change":null,"blocking":false}
{"item_id":"fi-0011","round":1,"verdict":"revise","reason_codes":["CONTEXT-INACCURATE"],"evidence":"Retrieved nps.gov; observed_alt is 'Download from the App Store' and gold_alt is 'Download NPS App on the App Store'; the gold rationale explicitly says observed is underspecified, but observed_alt_verdict is 'correct' which contradicts the gold standard judgment.","required_change":"Change observed_alt_verdict from 'correct' to 'wrong', since the gold standard identifies the observed alt as lacking the app name needed for informed decision-making.","blocking":true}
{"item_id":"fi-0012","round":1,"verdict":"reject","reason_codes":["LEAKAGE"],"evidence":"Sprite symbol ID 'icon-search' contains the answer, and aria-label 'Submit Search' exposes it in markup; gold 'Search' is trivially derivable without seeing the image or page context.","required_change":null,"blocking":true}