Abstract
Backlog 54. The kite was the one part left leaning below zero (LR 64, -0.94 +/- 0.77 at 48 seeds), and the meter counted no kiting. A new meter counts each live seat's seconds by kite step. Over 12 seeds of the fair census the trial seat kites 4.7% of its seconds alive, about 114 s a match. Of that, 62% is spent holding past the target's reach, 18% massing, 11% stepping out and 9% stepping in. The kite acts, and mostly waits. The meter changes no play (the per-seed differences are LR 63's). The kite stays on; its hold goes to a duel family, backlog 55.
Index terms: kite, meter, census, tuning.
I. Introduction
A. Intent
To learn whether the kite acts on naval.map, and what it does when it does, before any number of it is tuned.
B. The prompt
continue
II. Decision Log
| # | Cause | Decision | Not taken, and why |
|---|---|---|---|
| D1 | The prompt; CI-030 named 54 next | Backlog 54 | — |
| D2 | Act::kite() already names the step for all three kites (Kite, WingKite, RallyKite) | Count the act once a second after the think (think_released) | Counting inside Tactician::kite: it is also asked by plain RuleBots and in what-ifs, which do not move the tank |
| D3 | The question is "how much of a life" | Seconds by step, out of seconds alive, by group | Kills won and lost while kiting: needs kill attribution the meter lacks; the time answers the hypothesis's first branch |
| D4 | Steps are 8 | Five words in the line (stepping in and holding summed) | Eight: the reasons inside a hold are for 55 |
| D5 | F: 4.7% of the life, 62% holding | The kite stays on; its hold to a duel family (55) | Tuning kite_spare on the census: a number moved by noise; a family scores a duel |
III. Methods
Code. meter.rs: Meter::kites: [[u64; 9]; 2], by group the 8 steps in KiteStep::ALL's order and seconds alive; kite(step); summed in add_fronts; the census line. commander.rs: think_released, for a live seat on a whole second, meter::kite(self.act.kite()). Test: a_second_alive_counts_its_kite_step.
Read. build/lr65/run.sh: a frozen binary; the fair census, naval.map, 900 s, 3 respawns, seeds 0 to 11 in three jobs of 4; trial the defaults, baseline caps = baseline, -kite. The meter totals from each job's census line, summed.
IV. Results
| Measure | Result |
|---|---|
| Trial seconds alive | 88,122 |
| Kiting | 4,111 (4.7%), about 114 a trial match |
| Holding / massing / stepping out / stepping in / closing | 2,549 / 746 / 433 / 356 / 27 |
| The rest kiting (kite off) | 0 of 199,437 |
| Difference a seat, 12 seeds | -1.76 +/- 1.04 (LR 63's row) |
Tests: brain 84 passed and 11 ignored; client 32; net 14; proto 3; sim 50.
V. Discussion
- The kite acts, and waits. Two minutes a match of a part is not seldom; six tenths of it is a tank past its target's reach, waiting for a window.
- Time is not cost. Whether those minutes are lost depends on what the tank would do instead, which a census cannot isolate and a duel can.
For Paul: none blocking. Backlog 55, the kite's hold in a duel, is next. Commits are local; nothing was pushed.