Prompt 24 — Physics invention candidate miner

Inputs

FROZEN_SESSION_EPISODES: {{FROZEN_SESSION_EPISODES_JSON}}
VISIBLE_LIBRARY_INDEX: {{VISIBLE_LIBRARY_INDEX_JSON}}
FAILED_ATTEMPTS_AND_NEGATIVE_RESULTS: {{FAILED_ATTEMPTS_AND_NEGATIVE_RESULTS_JSON}}
HELD_OUT_TASK_SUMMARY: {{HELD_OUT_TASK_SUMMARY_JSON}}
WORLD_INTERFACES: {{WORLD_INTERFACES_JSON}}
CLEAN_ROOM_POLICY: {{CLEAN_ROOM_POLICY_JSON}}

Task

Mine recurring, decision-relevant structure from verified or explicitly bounded episodes and propose candidate inventions.

Candidate classes include:

• new quantity or operational definition;
• dimensionless group;
• recurring relation or empirical law;
• state variable or hidden parameter;
• conserved or approximately conserved expression;
• symmetry or transformation;
• representation or coordinate system;
• model family, closure, approximation, or regime boundary;
• experiment strategy or measurement operation;
• reusable simulation or inference algorithm.

For each candidate:

1. cite the episodes and observations that create the pressure;
2. distinguish discovery from naming or rebranding;
3. give positive, negative, borderline, and adversarial cases;
4. define quantities, units, assumptions, regime, and operational tests;
5. propose alternative formulations and the nearest visible-library concepts;
6. state predictions, falsification tests, and independent validation routes;
7. estimate compression, predictive, transfer, and experiment-efficiency utility;
8. preserve failed evidence and avoid cherry-picking;
9. mark all candidates discovery_staging and no stronger than supported evidence;
10. never claim equivalence to hidden modern concepts or novelty to humanity.

Return a physics-library-batch object.