What happened after arrival?

Inspect saved ordinary-reset games. Actual actions, unused allowance, life losses and resource changes stay separate.

Complete acquisition cohorts

Each row is one trained agent, with two 32-game batches sharing novelty within each batch. These are descriptive outcomes; games are not training replications.
AcquisitionTraining transitionsGamesMean returnMean living roomsExitsCompletionsLife losses

Choose a game and interval

Loading saved games…

Dots: recorded room contexts. Red: life loss. Green: net inventory change. Dashed line: observed game end. Grey region: allowed actions that this game did not execute. Click an event point to start an interval there.

Recorded events in the interval

Completed actionPost-action room / positionRecorded eventInventory [key, sword, torch, amulet]Life state at this event

What the data establish—and what is unobserved

No new simulator replay or policy evaluation is performed by this viewer. Fine movement, intentions and planner predictions are absent. An action-end position need not be the interaction location. Net changes do not identify individual items.

Sources and exact input hashes