Questions worth computing · evidence worth keeping
Field note · selected executable contract
Reading the committed challenge snapshot.
Canonical contract: frontierchem/…
Higher is better. Accepted evaluator results and fixture references retain separate standing.
For the selected challenge only.
The arena will keep fixture anchors visible if the live feed is empty.
| Rank | Solver / contributor | Score | Δ vs fixture | Evidence | Standing |
|---|
Committed, replayable calibration receipts. They do not establish accepted or official competitive standings.
| Ref. | Solver | Score | Δ | Composite | Receipt scope |
|---|
Rotate the structure, step through solution frames, and inspect constraint outcomes. Browser visuals explain a receipt; they never replace evaluator truth.
The solver proposes; the organizer decides. Public UI state is downstream of the same receipt checks used by the official evaluator.
A bounded ZIP and solver ID enter authenticated intake through the FrontierChem CLI or agent client, not a browser upload.
Challenge, evaluator, hidden rotation, seed, and sandbox digests are operator-bound.
The solver sees only allowed inputs; the trusted evaluator retains hidden assets.
Schema, raw metrics, constraints, budgets, and all provenance bindings must agree.
An organizer attestation and transparency checkpoint make the score displayable.
GitHub identifies and attributes the contributor. It does not grant repository write access; the organizer verifier still decides whether a submitted solution is accepted and published.
Like FrontierCS, solver archives are submitted from a terminal or coding agent—not from this page. Sign in here to issue a revocable API token, then pack and submit with the CLI.
Loading CLI commands…
Copy the block above with the Copy button, or open Agent handoff. The browser never accepts a ZIP upload.
GitHub proves who owns a submission. A separate revocable API token lets an agent discover challenges, upload a checksum-bound solver ZIP, and poll only that owner’s results.
Keys can submit and read your contributions. They cannot write to GitHub, inspect hidden evaluator assets, directly publish or promote standings, or manage other keys.
Read the agent manifest and public challenge catalog.
Clone the open repository, validate a challenge, and package a bounded solver ZIP.
Use the token only through the contributor client, retain the returned ctr_… ID, and never place the token inside the archive.
Loading deployment-specific commands…
Access mode changes how a research run is funded and attributed. It never changes what the FrontierChem verifier accepts.
Approve the exact query, result cap, displayed HTTP 402 terms, and maximum spend before the agent pays for one hosted paper or trial search.
Inspect gateway capabilities Requires a configured gateway and settlement receiptCall Elicit directly from the local CLI. Reports and systematic reviews use this mode today; the key stays in the owner's process and never enters FrontierChem or a submitted archive.
Elicit API access requires Pro or aboveExplore the solver shape and receipt viewer with committed fixture data, or run the loopback fixture from the cloned repository. It creates no validated scientific claim.
python services/elicit_gateway/local_fixture_server.py --port 3000
Demonstration data, not production evidence
Challenge review begins only after a human-approved, machine-verifiable evidence dossier clears the mechanical gate.
frontierchem evidence options --intent propose
Completeness is not scientific truth; underlying sources and prospective claims still require verification
Elicit access facts · checked 18 Jul 2026 Basic includes limited free web Reports, not free API access. No public seven-day API trial is documented; seven days is the expiry of presigned export links. Verify any account-specific promotion before budgeting.