EXPERIMENTS IN TRUST
Another report.
A better reason?
Run a fresh world with our trained confirmation selectors. Follow the observations they buy, see what changes their answer and inspect the price of checking.
The workbench checks the frozen release and executes all three checkpoints before accepting a trial.
What should we trust?
Awaiting your trial
Choose a world and run a trial. No sample outcomes are prefilled.
Observation replay
Reveal what the policy saw, one report at a time. This replays your completed trial; moving the slider does not rerun inference.
| Step | Source / charge | Group | Observed label | Predictive probability | Resources left | P(bias=0) |
|---|
Recorded prediction endpoints
Initial means after the four inherited reports. Final means after all paid actions. Intermediate group predictions are not reconstructed.
Inspect exact execution receipt
This unsigned record binds the request, result, original source and checkpoint through hashes. It establishes internal consistency, not independent authenticity or factual correctness.
A receipt you can keep.
Your receipt is available here for at most 24 hours, until the 128-receipt capacity evicts it, or until this workbench restarts. Download it to retain your own copy. Receipts from other visitors are not listed.
Receipts use ephemeral process memory. Hosting does not provide permanent trial storage.
Useful evidence.
An unfinished result.
The overall registered gate remains FAILED. New exploratory trials do not amend it.
Checking the source-bound evaluation.
Explore all recorded findings ↗