Skip to content

Real-world Bridge

ChemWorld’s relationship to reality should not be reduced to whether one simulator is “realistic enough.”

The Bridge question is whether virtual training reduces the experiments, risk, and cost needed to adapt to independent models, real datasets, and narrow physical systems.

Current status

This is a validation roadmap, not an operational physical-lab product.

Validity ladder

Level Question
Contract validity Are state, actions, conservation, and replay correct?
Decision validity Does the world create meaningful experimental trade-offs?
Causal validity Do interventions change rules and effective strategies?
Behavioral validity Do agent rankings persist across independent backends?
Transfer validity Does virtual training reduce target-system experiments?
Numerical validity Are values accurate for a specific real system?

Core primarily targets the first three. Bridge experiments test the fourth and fifth. Universal numerical prediction is not the current goal.

What may transfer

Measurement strategy, exploration order, uncertainty handling, failure recovery, safety habits, change detection, and few-shot adaptation may transfer even when a virtual optimum does not.

What cannot be copied directly

Anonymous catalyst doses, virtual risk scores, uncalibrated yields, virtual equipment settings, and rankings from a single simulator cannot be treated as real recommendations.

Bridge path

Causal Core → independent backend → real dataset
→ shadow-mode physical lab → approved narrow closed loop

A Bridge Pack must bind material identity, action and observation mappings, units, equipment limits, calibration and independent test data, uncertainty, safety approval, and replay provenance.

The primary metric is not zero-shot replication of a virtual recipe. It is transfer advantage at the same target-system budget and the number of real experiments saved relative to learning from scratch.

Partition is a lower-risk first candidate; flow offers stronger control relevance but requires shadow-mode safety and equipment validation. Crystallization and distillation should follow only after narrower bridges are established.