Frontier model A

Trial · n0-search-pdf-page-ranges

← n0-search-pdf-page-ranges
Outcome
PASS17m
Reward
100.0%
Checks passed
17 / 17
Runtime
17m 8s
Harness
Cursor CLI
Task version
v3.2.0

Checks

CheckResultCreditWeight
baseline: the service answers a searchPASS100%1%
baseline: the web app answers a searchPASS100%1%
backend: search API results carry the best passage's page rangePASS100%18%
backend: MCP search results carry the page rangePASS100%8%
backend: one result per document, and a big PDF does not crowd out the othersPASS100%10%
backend: a search does not page through every passage of a huge PDFPASS100%8%
web: search route results carry the page rangePASS100%12%
web: one result per document, and a big PDF does not crowd out the othersPASS100%8%
web: a search does not page through every passage of a huge PDFPASS100%6%
web card: the result card shows Page N or Pages N–M, and nothing without pagesPASS100%22%
partial spans: a span with one bound shows no page 0 or dangling rangePASS100%6%
preserve: the search APIPASS100%0%
preserve: the MCP search toolPASS100%0%
preserve: the web app's search routePASS100%0%
preserve: the web app's search result cardPASS100%0%
preserve: the backend's ruff and strict mypyPASS100%0%
preserve: the web app type-checksPASS100%0%