Game over is the game's: a seat knocked out no longer ends it for the casters, the viewer and the play page

[earakely-scale]

The bug

A seat's state has its own game_over, which the bridge sets once that seat is defeated (GameOver => human.defeated || turn >= turnLimit || victory != null, per seat). The env took any seat's game_over as the whole game's, in three places:

• The live data (/live/data.json, live.game_over): the casters and the viewer's director read it.
• The play page's status: /play/api/status and the play view.
• The data/get summary's top-level game_over: it was the first seat's, and openciv3_await_game reads it.

What it did:

• In the sol-vs-opus broadcast, Sol lost its last city, and the casters signed off at turn 155 calling it "a win on score". The conquest victory was only recorded at turn 156.
• With three or more agents, the first elimination would bring the outro and the final-summary card mid-game while the match played on.
• openciv3_await_game would stop waiting if the first seat fell.

The fix

One _game_over() serves all three places. The game is over:

• at a victory;
• at the turn limit, which any seat still in the game reports;
• or once every seat is defeated.

A defeated seat's own defeated and its tool replies are unchanged: the game is still over for that agent.

docs/tools.md says data/get's game_over and victory are the game's, not the first seat's, and docs/viewer.md notes that live.game_over means over for every seat.

Tests

• 512 pass; ruff is clean.
• New, on the real bridge (tests/env/test_seats.py): with three seats, Egypt disbands its units and is defeated, while Rome and Greece play on. The live data, _game_over() and data/get say the game goes on, and Rome's brief moves to T1.
  • On main the live data says it is over (assert (True, None, 1) == (False, None, 1)).

🤖 Generated with Claude Code