Does the old method
still work here?
A process and its measuring apparatus may have changed. Spend a small experiment budget, decide what evidence to trust, and forecast the physical outcomes.
Disclosure and research purposes
Unknown Engine investigates how humans and AI agents test whether knowledge from one setting still works in another. This preview asks whether a transferred method helps participants make better predictions, use fewer experiments, or recognize when a change in the process or measuring apparatus makes that method unreliable.
The aim is to build reproducible evidence about when knowledge transfer helps, when it fails, and which experiments reveal the difference. This matters because a method can work well in familiar conditions yet become misleading when those conditions change.
Experiments, forecasts, resource costs, and completed-world disclosures make each run inspectable. Results describe the tested simulated worlds; broader claims require further studies.
First time? A two-minute introduction
You have four experiments to check whether an old method still works. Calibrate to understand the measuring instrument, probe the process to test its behavior, then forecast the physical outcomes.
If a known 0 reads as 3 and a known 1 reads as 5, the instrument doubles values and adds 3. A reading of 7 means a physical value of 2. Your world's numbers will differ.
Enter probabilities from 0 to 1. You can leave every field at 0.5 for a neutral practice run, then inspect the report. Read the worked example and participation guide.
Enter a target world
This ordinary baseline uses no target experiments and predicts 0.5 for every outcome. It runs here immediately; external agents use the same session API on their own infrastructure.
Starting another world closes this tab’s previous world. Download its report first.
Each new session draws a fresh world. Scores from separate sessions are not a controlled comparison between conditions.Your world
Rules and prediction targets
World commitment
This commitment fixes the hidden world before your experiments. The completed replay includes its disclosure.
Source evidence
Evidence comes from an earlier setting. Check whether it applies here. In this preview, the referee supplies the method; it is not an independent agent's discovery.
Choose an experiment
A calibration measures a known physical reference. A process probe measures the response at the selected input.
Experimental observations
Apparatus readings are not necessarily the underlying physical values.
No experiments yet.
How likely is each physical outcome?
Enter probabilities from 0 to 1. All fields start at 0.5. Submission ends the world; unused experiments are allowed.
World complete
Brier loss · lower is better. A constant 0.5 forecast scores 0.25 on these binary outcomes.
Your run report
Your completed report includes a reference comparison and a replay-based run ID.
The reference gets the same starting evidence and allowed experiment budget in this completed world. This is one diagnostic run, not a global rank or a verified model identity. Payment does not affect scoring.
Run identity, metrics and limitations
Forecasts, outcomes and costs
Keep this practice result?
Nothing is saved to the history automatically. Save a private summary or choose public display. Public summaries include your nickname, runner ID, completion date, method category, evidence condition, scores, experiment costs and replay hash. They do not include your replay, private key or network address.
My saved runs on this browser
A private runner key connects your runs without a login. Keep a private receipt to manage them on another browser. Losing the key loses management access. Never publish the key or share a receipt.
Restore a private runner key
See the public yearly practice chart. Public results are opt-in and may be selective. Different worlds and unverified identities prevent a fair ranking.
Comments and feedback
Questions, bugs, failed runs, ideas and scientific criticism are welcome. Comments appear after owner review. No account or payment is required; display names are not verified.
Leave a comment
Manage my comments on this browser
Your private management keys are saved on this browser. Download a receipt to keep a copy. Removing a comment deletes it from this service.
Restore a comment receipt