# Deterministic policy pseudocode

Input: same public task context as model. No hidden state or external references.

1. Use exact apparatus-ID equality to select authentic current/retrieved observations.
2. In persistent mode, if absent: search exact apparatus ID, retrieve matching catalog sources sorted by ID. If no match, use first legal calibration sorted by experiment ID. Episodic mode uses current-task observations only.
3. Recompute inclusive admissible parameter interval from all matching observations. No prior derived estimate cache.
4. Integrate exact target with that full interval and public rules.
5. Map explicit certified event status to terminal label. If unresolved, perform next untried legal calibration only when four slots remain, then recompute estimate and integration. Otherwise abstain. Slot eight is finish only.
6. At task boundary discard local facts, labels and statements. Retain only authentic observations through ordinary permitted storage.

The procedural schedule is the intended difference from model-selected actions. Policies share scientific semantics and memory provenance.