Model-guided action

Current approved assessment

Evidence position

ESTABLISHED / STRONGLY EVIDENCED

Review depth

BASELINE SURVEY

Evidence movement

Initial observation — no prior comparable baseline linked

Evidence-bearing for local model-guided action; strong capacity evidence and limited programme distinctiveness can coexist.

Why it sits here

Model-based reinforcement learning and associative inference provide operational frameworks. Instrumental choice and Pavlovian inferred-value responding supply distinct behavioural assays; neither alone licenses all meanings of an internal model [S80, S81, S82].

Principal uncertainty

The remaining uncertainty concerns which representations and computations particular assays identify, their causal location, and transfer across tasks, rather than whether the modest capacity exists.

What would change the view?

Reassess if a proposed application depends on a particular forward-model architecture, or matched alternatives explain both transfer and intervention effects. Define that comparison before commissioning further work.

Launch-v1 depth

Baseline-only public wrapper. No deeper evidence artefacts are fabricated.

Approved history