[ptescher]

Step 2 of plans/actors.md: Captain is split into a commander for each team and a unit for each drone, and a game is played from a roster. Based on actors-queue.

What changed

• recreation_towers::actors (no I/O, for the arena and later the server):
  • trait Command { fn decide(&mut self, &TeamView) -> Result<Vec<Directive>, String> }, Directive { drone, order, queue }. Commander implements it.
  • trait Unit { fn fly(&mut self, &OnBoard, &mut Instruction) -> Option<Order>; fn restart(&mut self) }. OnBoard has private fields: team, number, flying, clock, silent/in_contact, battery, link timeout and bingo, obs() (its EdgeObs), done(order). A unit cannot reach another drone or the team's map.
  • Autopilot, the default unit: Instruction::tick_with.
• cognition_towers: Match::with_units (a unit for each of ten drones) and step_with(&mut dyn Command, &mut dyn Command); Roster (new, of_captains, with_unit) and play_roster. Captain is kept: Directing(captain) is its Command; Captain::unit(team, k) replaces Captain::edge. Edged makes an EdgeUnit (autopilot, then the policy once a second when out of contact) with its own clone of the policy. No existing edge read captain state. compare_edge takes an Rc<dyn Edge>.
• Seat protocol: a reply is a list of [drone, order] or [drone, order, queue] (Reply); old replies read as before. serve_with serves any Command; SeatProcess hands the queues to the match.
• Glossary: Commander, Unit. plans/actors.md results, domains/cognition/README.md.

Not done / differs from the brief

• A unit runs on every arena step (10 a second), not once a second. The queue (a Silent(after) time, the Done test) is worked at that rate today; running the default unit at 1 Hz moves the numbers. The edge policies still run once a second, inside EdgeUnit.
• No advisor trait (step 4). No server changes.
• cognition_league tests/acceptance.rs fails with "The suite's 120 s ran out before this test started", as on the base branch; its unit tests pass.

Gates (before = base actors-queue, after = this branch)

denial, ruleset v2 200, edge 30 and cognition_league judge --game captain --champion timid --challenger default --seeds 24 are byte-identical before and after (cmp). Judge: kept out (mean difference +0.175, lower bound -0.132); champion 24 won / 24 lost, 512 points; challenger 26 won / 22 lost, 547 against 493.
Tests: -p recreation_towers -p cognition_towers -p recreation_towers_ground pass (36 + 140 + 7 acceptance + 7); cognition_league unit tests pass; recreation_towers_server and teleoperation_command build; fmt and clippy clean; expectation check expectations: 0 errors.
New tests: a counting unit runs on every step for every drone in and out of contact; teams on different units, and one drone on a different unit, are deterministic and change the game; old and new reply; a served and a spawned seat with a queue.

🤖 Generated with Claude Code