[ptescher]

Step 1 of plans/actors.md: a unit holds a queue of instructions, not one order and one standing order.

What changed

• recreation_towers: Instruction { order, queue: Vec<Queued>, taken }, Queued { when: When, order }, When::{Silent(f32), Done, Battery(f32)}, MAX_QUEUE = 4 (a longer queue is cut, tested). Standing stays, and is the queue of one (From<Standing> for Queued, Instruction::give, Instruction::standing()). Entries are taken only out of contact, in order. The failsafe (home on a lost link or at bingo) holds for a drone with no instruction that waits for the silence and none started.
• "Done" comes from the drone's own knowledge (edge.rs: order_done, Game::order_done): Take when its camera sees its team holding the tower, Goto within 35 m, Hold/Circle/Home never.
• Seat protocol: ToServer::Order has queue: Vec<Queued> beside the old standing (both serde(default), an old seat still works; a queue wins over standing; tested).
• Fast arena: Match flies the queue (Captain::queue, default from standing; Autonomy::queue). New policies Chain (persist, the nearest tower the team does not hold when done, home) and Timed (the same by the clock). denial measures them.
• Server: World::command_queue, queue flown for drones with no wing. Commander app: only compiles (queue: vec![]).
• Spec: expectations/09-towers/rules/17-queue.expect, glossary term Queue, vocabulary in 09-towers/section.expect, driver support. expectation check has 0 errors.

Not done

• The flight controller still has one standing slot. The ground uploads the first Silent instruction as the standing plan, as today. cognition_plan, cognition_executor, the MAVLink encoding and ardupilot/ are untouched. The on-board queue is a later step.
• No UI for a queue in the commander app.
• Docs not regenerated (needs a window).
• The suite's coverage mode runs only some expectations of the section, so acceptance.rs also has a queue module that plays each queue scenario by hand.

Results (denial, 100 seeds, both sides)

Existing rows identical before and after:

against failsafe	margin	error	won	lost	dark	dark(f)
persist	+0.060	0.095	103	97	157s	153s
Hold after 5 s	-0.441	0.095	69	131	341s	156s
Home after 5 s	-0.807	0.083	48	152	175s	179s
Home after 30 s	-0.272	0.095	81	119	206s	169s
chain (new)	+0.060	0.095	103	97	157s	153s
timed 20 s, 60 s (new)	-0.666	0.085	55	145	197s	189s
timed 30 s, 90 s (new)	-0.226	0.096	85	115	211s	168s

against persist	margin	error	won	lost
chain	-0.000	0.094	100	100
timed 20 s, 60 s	-0.730	0.082	50	150
timed 30 s, 90 s	-0.350	0.093	75	125

A longer queue does not beat persisting in the valley. chain plays exactly as persist: a drone that takes a tower ends the shadow, is in contact again, and a drone in contact takes no instruction, so "done" never starts the next one. timed pulls drones off their towers and loses.

Tests

cargo test --release -p recreation_towers -p cognition_towers -p recreation_towers_ground -p recreation_towers_server pass. cognition_league: unit tests pass; its acceptance test fails with "The suite's 120 s ran out" (the nested builds of its challenger tests); not shown to be caused by this change. cargo clippy --workspace --all-targets and cargo fmt: clean. cargo build --release -p recreation_towers_server -p teleoperation_command: OK.

🤖 Generated with Claude Code