Abstract
LR 82 left open what a bridge on atlantic costs in the builders' time. The meter now counts, each second a seat is alive, whether its squad is out and on what task: each build kind (a road over water is a bridge), harvest, capture, repair, lift, the walk home, or out with no task, with the trial's seats apart. The count decides nothing: the same atlantic census before and after prints the same lines to the digit. On a first four seeds the squad is out 37% of the time alive, and most of that is the walk home (39%) and captures (28%); bridges are 1.5%.
Index terms: meter, instrumentation, squads, building.
I. Introduction
A. Intent
To commit and push LR 82, then add the build-time meter it called for.
B. The prompt
commit and push. then add the build time meter
II. Decision Log
| # | Cause | Decision | Not taken, and why |
|---|---|---|---|
| D1 | "commit and push" | LR 82 committed and pushed on main (a6c71f3) | — |
| D2 | A census sees the bot's observation, and obs.me carries the squad's task | Count in Commander::think_released, a second a seat alive, beside the kiting count | Count in the sim: the meter is the brain's, and per seat and group there |
| D3 | BuildKind::Road builds a road on land and a bridge on water | A road whose site is water or bridge on the map is a bridge | One "road" kind: the bridge was the question |
| D4 | F: on the first look "other" was two-thirds of the time out | Split it: capture, repair, lift, home, idle | Attribute the walk home to the job before it: needs memory a seat, and a count that reads as what the squad is doing is simpler |
| D5 | A meter must not change play | The same census at a6c71f3 and after, 4 seeds on atlantic: the same to the digit | Trust it: a counter cannot steer, but the check was cheap |
| D6 | cargo fmt --check fails | Left as is: 972 diffs before the change, the file's long-line style kept | Format the crate: a thousand-line change in a meter step |
| D7 | The machine plays the pushed tree; this step is uncommitted | The atlantic reading waits for the commit | Copy the files to the machine: its tree would not be a commit |
III. Methods
meter::SquadTask (13 tasks) and meter::squad, Meter::squads ([[u64; 14]; 2], the last seconds alive), summed by add_fronts and printed by Meter::line. The check: SEEDS=4 SEED0=500 SECS=300 MAP=maps/world/atlantic.map JOBS=4 make census in a worktree at a6c71f3 and in the working tree; diff of the sorted seed and team lines. cargo test -p treadcore-brain --release.
IV. Results
In LR 83: the same lines to the digit; 88 tests passed; the first look (the squad out 37% of seconds alive; home 39%, capture 28%, pillbox 11%, wall 9%, harvest 5%, bridge 1.5%).
V. Discussion
The question LR 82 asked was about bridges, but the first look says the walking matters more: each job sends the squad out and back. If a census on atlantic holds that, the posture saves its team the round trips more than the building itself. That census is the next step.