rewirebio.iobenchmarks

APIRelease 2026-11-21-adc15377fab2

JSON API

Read the release this site serves as JSON. The API is read-only and needs no key. Every answer names the release it was read from.

Try it

curl https://benchmarks.rewirebio.io/api/v1/use-cases/

Addresses end in a slash; an address without one is redirected. The machine-readable description is at /api/v1/openapi.json (OpenAPI 3.1).

Endpoints

  • GET /api/v1/: Lists every endpoint with its address, and links the OpenAPI description and the docs page.
  • GET /api/v1/models/: The models the release holds, in the order of the site's model index (by name). Query parameters: limit, cursor.
  • GET /api/v1/models/{id}/: A model record with the sources it cites, its recorded hardware and access facts, its rank on each use-case leaderboard it appears on, and the unsettled findings about it. An earlier id of a model also answers here and names the model in alias_of.
  • GET /api/v1/models/{id}/results/: The published results the release links to a model, as the model's page lists them. Each result carries its origin, its review status and a link to every source it cites. Query parameters: limit, cursor, metric, origin.
  • GET /api/v1/benchmarks/: The benchmark suites and challenges the release holds, in the order of the site's benchmark index (by name). Query parameters: limit, cursor.
  • GET /api/v1/use-cases/: Every use case in the release, with the protocol its leaderboard is ranked on, if any.
  • GET /api/v1/use-cases/{slug}/board/: The ranking the use case's page shows: one protocol, one metric, never a composite. protocol.relevance says whether the protocol is directly relevant or a proxy. When nothing is ranked, ranked is false and empty_reason says why.
  • GET /api/v1/contested/: Findings in which studies disagree about the same models, in the release's order: unresolved before resolved, then most severe first. A finding is an automatic lead, not a finding of error, and never withholds a result. As on the site's contested page, the list holds every finding, including those the release lists for curation only (listing review) and resolved ones; filter with listing and status, and read each finding's listing and resolved fields. Query parameters: limit, cursor, status, severity, system, listing.
  • GET /api/v1/contested/{id}/: A finding with the studies it compares, their numbers, its follow-up actions, and links to the sources and results on each side.

Every endpoint also takes release_id. A request pinned to a release other than the one served is answered with 404, never from another release. An unknown query parameter is answered with 400.

What every answer carries

  • release: The data release every value in the answer was read from: its id, date, the repository and commit it was built from, and the release manifest with checksums.
  • licence: The data licence position. statement gives it in one sentence. spdx_id is null because no single licence covers an answer: values taken from sources stay under their sources' terms. compilation names the licence for what Rewire made, with what it covers and excludes.
  • citation: How to cite the release.
  • links: The answer's own address, this page, the OpenAPI description and the page of this site that shows the same content.
  • page: On lists: the total, and the cursor and address of the next page.
  • data: The answer itself.

Data licence: The compilation and annotations made by Rewire in this release are offered under CC BY 4.0; values and text taken from cited sources stay under the terms of those sources and are not covered by this licence. The limits are set out on the data licence page.

Citation for this release: Rewire biological model benchmark database, release 2026-11-21-adc15377fab2. https://benchmarks.rewirebio.io/

How to read the answers

  • A leaderboard ranks one use case on one protocol and one metric. Ranks are never combined across protocols and there is no composite score.
  • Boards, a model's rankings and its results name things the same way. numeric_value is the value as the result record holds it, in the unit given beside it. display_value is the text the site prints, which may show a fraction as a percentage. Every url is absolute.
  • protocol.relevance is direct when the use case's reviewers judged the protocol directly relevant, and proxy when a related protocol is used because no direct one ranks two tools.
  • review_status is reproduced only when the result or its evaluation cites a run receipt in the release; the receipt field then links it. source_checked means the value was checked against the source it cites. It does not mean the result was rerun. superseded means the release has replaced the result and keeps it for reference.
  • record_status is the record's own review state in the release, with one exception: a record the release marks reproduced is given as source_checked, because this API says reproduced only in review_status, and only with a cited run receipt.
  • origin says who produced a result. independent is true only for a study whose authors did not develop the tool. A Rewire run has origin rewire_run and is not counted as independent.
  • Every result links each source it cites: the source's page here, and its own address when the record gives one.
  • A contested finding is an automatic lead that studies disagree. It is not a finding of error and never withholds a result.
  • Each finding, and each link to one, carries listing: public for a finding the release lists as a lead for readers, review for one it lists for curation only. The contested list holds both unless you filter by listing.
  • A filter value the release cannot match is answered with 400, not with an empty list: an unknown parameter, a metric the model has no result for, or a system that is not a record.

Releases and caching

Every value in an answer is read from the release named in release.id, and a new release replaces the old one at the same addresses. Ranks, labels and counts are worked out by this site from the release, so they can also change when the site's code changes; the X-Rewire-Frontend response header names the site version. Answers are sent with Cache-Control: public, max-age=300, so a client may reuse one for five minutes; the release it came from is always in its body. To keep answers longer, key them by both the release and the site version. Cursors are valid only for the release and filters they came from. The API holds no keys, accounts or quotas of its own.

The whole release is also available as files, with checksums, from the downloads and the release manifest.

Other ways for an agent to read the database are on Connect your agent.