A Choice to Eat

From urgency to option

In the earlier berry fixture, native self-care waited until urgent hunger. Saying that a pawn would eat did not dispatch a job. Now, a pawn answering a core question has a separate eating choice when eligible food is available. Saying something and staying silent remain valid, non-executable alternatives. The eat choice names one supplied nearby food stack—berries or simple meals for now. The core cannot order or propose consumption directly.

The game computes a portion capped at 25 units, with no extra stacks collected. Diet, forbidden status, reachability, reservations, availability and a deadline still constrain execution. A deliberate choice can select berries before native self-care would, without changing the native hunger thresholds or diet. Existing commitments and pending work offers block this first eating path.

Three different outcomes

Scripted tests began at 20% Food with 225 berries. Speech alone left all berries untouched and Food at 18.4%. The explicit eat choice consumed 16 berries, leaving 209 and raising Food to about 97.8%. Withdrawal before consumption recorded an interruption, consumed zero berries and left all 225 intact. Each case received twenty seconds of native gameplay; no character model chose these responses.

The consumption receipt comes from the native ingestion step, not from a promise or a job disappearing. The crew log distinguishes choosing, starting and consuming. Review also caught a relay that rejected the new response and an old answered counter that kept options hidden. Both are corrected, alongside shared trial cleanup; consumed food is not counted as hauled supplies.

The release build also stopped eating cleanly through shared operator shutdown, with that intervention labelled separately from pawn withdrawal. All four cases survived paired restore and a full restart. The final review is clean; 324 Node and eleven Python checks pass. Earlier trials and allowances remain unchanged.

What remains open

This proves mechanics with scripted choices, not live model-driven eating or campfire-to-meal planning. Simple meals are implemented, but these native consumption checks use berries. The entry point is answering a core question, not autonomous reflection. Paired checkpoints still require active eating to finish or stop. Whether a live pawn chooses this action appropriately remains the next question.