Two-minute path
Five clicks on this site that show what GAVEL claims, what is proven and what is not.
If the host has judge rooms switched on (JUDGE_ROOMS=true), you can also start a key-less sandbox room from the web app; the page /try lists the steps. The help button on the bidder screens opens the hold explainer and a six-step guided tour. See the demo guide.
Read the promise and the status
Open Start here and then Real, simulated, recorded.
Expected: one sentence (every bid is a PayPal authorization hold, only the winning price is captured, every other hold is voided) and a table that labels each area as real sandbox, fake, simulated, recorded or not available.
Open the gates
Open Gates G0 to G5.
Expected: a table with G0, G1, G2 and G4 as PASS, G5 as PARTIAL and G3 as IN PROGRESS. The table is generated from docs/GATES.md, so it cannot differ from the repository.
See why voice goes through a confirm prompt
Open Voice and the confirm gate.
Expected: the speech accuracy numbers, both providers above the 0.05 wrong-accept limit, and the decision that every spoken bid is confirmed before it counts.
See the money path
Open Settlement walkthrough or Money flow.
Expected: hatched means held, solid means charged, hollow means released. One winner is captured at the winning price and the other holds are released.
Check what is simulated
Open The 30-day simulation.
Expected: eight scenarios, all PASS, each labelled simulated. They run our code against a model of PayPal on a fake clock, so they do not replace gate G3.
What you have seen
- Real on the PayPal sandbox: gates G0, G1, G2 and G4.
- Partial, with the fallback in place: gate G5 on synthetic speech (not a pass; fixes chosen on the same clips, no real-room test).
- In progress: gate G3, final reading about 12 Nov 06:17 UTC.
- Simulated: the multi-day run and whole-auction settlement on the fake PayPal, plus one lot of three holds on the real sandbox.
Next: the rubric map points each judging criterion at its evidence.
Last updated on