AUTONOMOUS PROCESS DISCOVERY / PROOF 01
Find the useful process window before the sample—or the budget—runs out.
MatterLoop closes the loop between a simulated materials reactor and the next experiment. It interpolates results from sparse trials and balances promising recipes against unexplored conditions.
Choose a useful recipe with fewer trials.
Change how the planner chooses temperature and processing time. It tries to recover more oxygen while using less energy in a simulated reactor.
How to try it in 60 seconds
- Choose a feedstock and a search policy. Read the next recipe before running it.
- Press Run 18-experiment campaign. The surface updates as the planner gathers synthetic measurements.
- Compare oxygen recovery and energy use, then read the recipe log. Reset before comparing a different policy from the beginning.
The reactor and measurements are synthetic. The adaptive policy uses an exploration heuristic, and its distance score is not a statistical confidence estimate.
Yield / energy frontier
CAMPAIGN LEDGER
Every proposal leaves a trace.
Observed values include seeded measurement noise. Repeat a campaign and the same policy produces the same evidence chain.
| RUN | PROPOSED BY | TEMP. | TIME | O₂ RECOVERY | ENERGY | UTILITY | DECISION |
|---|---|---|---|---|---|---|---|
| No experiments recorded / run the first recipe above. | |||||||
PROTOTYPE BOUNDARY
The loop is real. The reactor is simulated.
This browser proof runs a sequential experiment policy against a deterministic synthetic process. The adaptive policy uses inverse-distance interpolation and an exploration bonus, not Gaussian-process expected improvement. Its distance score is not calibrated uncertainty. A field pilot requires reactor telemetry, calibration and safety interlocks.