Abstract
Four backlog items, in order. The team board's front, built every second since LR 24 and read by nothing, now leans the Expand rung's choice of depot (w_front, 0.5): a depot's cost is half its distance to the base and half its distance to the front, and the pick is kept until taken. The meter showed the front known at only 2.5 of Expand's picks a match, since the board counts only hostiles in sight, so the board now remembers hostile tanks for front_memory (30) seconds, for the front alone: 3.2 picks a match with a front, 14% without. A scenario holds with the part and fails at 0; the guard census reads -0.14 +/- 0.42 a seat over 48 seeds. Each scenario family now carries a floor, and make families runs all five at 1,000 seeds in 31 s; each falls under its floor with its part broken. Allies' goals on the coordination screen were found already shown by a hosting window and are now proven by a test; a joining window needs the wire. The crews' hand-out and the hop's bridge choice have tests of their own, and 22 functions got doc comments (121 undocumented to 99).
Index terms: team board, influence front, sighting memory, guard census, scenario floors, fleet command.
I. Introduction
A. Intent
To go on through the backlog after CI-025, each item proven by a test that fails without it, and the testing made faster where it can be.
B. The prompt
read backlog and roadmap. and approparate documenttation to remind yourself of the code goals. continue working on backlg itmes. and continue using testing to verify improvements and additions work as directed, you can also run at faster spaeed.
II. Decision Log
| # | Cause | Decision | Not taken, and why |
|---|---|---|---|
| D1 | "continue working on backlg itmes"; CI-025 named 49 next | Backlog 49 first, Expand's depot only; the guard posts split off as 51 | Both at once: one step at a time (Paul, 18 October) |
| D2 | Paul's rule of 20 October | w_front on at 0.5, the blend's middle; old method at 0 | A sweep first: tuning follows the features |
| D3 | Scenario Commanders had no board | board = true, rungs, [[depot]] at, depot_first in scenario.rs | A Rust test: scenarios are the form for new behaviours (LR 48) |
| D4 | F: the hostile 36 tiles off was never seen, so no front | A teammate placed to see it, as a team shares sight | Putting the hostile in the tank's own sight: the case is the team's picture |
| D5 | F: the tank turned back at 4% of a capture when its arrival erased the front | The pick kept until taken or given up, with w_front on | Hysteresis on the cost: one more number, and the flip is the front's, not the cost's |
| D6 | F: 2.5 picks a match with a front known, 60 matches in | The census stopped; front_memory added | Finishing it: it could only read zero |
| D7 | LR 35: the board forgets each second | Memory for the front only (known), the influence, threat, points, posture and rotation as before | Remembering in the influence: it moves five parts at once |
| D8 | "you can also run at faster spaeed" | Families at 1,000 seeds (6 ms each); the census frozen and in three jobs, other steps worked beside it | More census jobs: four cores |
| D9 | Backlog 47 | runs, floor, make families; floors 99% for the dodge families, 80% for the ready gun | Floors at today's rate: one seed's luck would fail them |
| D10 | Backlog 50: the hosting window already shows FLEET lines | A test of it; the joining half to 52 | A wire change: version 16 is fixed under goal 4 |
| D11 | Backlog 26 | Tests for FleetCommand::think and the hop's choice; doc comments for crew.rs, hop.rs, board.rs | All 121 doc comments: 22 now in the files touched |
| D12 | F: the hop's priced test sailed at 7 tiles on a bare map | Forest put behind the tank; the scarcity rule written into the test's comment | Changing the price: the rule is LR 6.2's, the test was wrong |
III. Methods
Parts explored. board.rs: TeamBoard::rebuild (the influence, the new seen and known), front (now on known), front_dist, sector_centre. commander.rs: Commander::expand (the open depots, the nearest, the kept target, the blended cost), the ladder's holds (auto), defend_base and post (read, for 51). weights.rs: Front, FrontMemory. meter.rs: fronts, front, add_fronts. scenario.rs: run_seeded (boards per team, rungs, depots moved, depot_owned), check (depot_owned_by, depot_first), every_family_holds_its_floor. crew.rs: FleetCommand::think, failed, job_tile, job_for. hop.rs: Hop::send, think (the bridge before a boat), bridging, and facts.rs crossing_prices, bridge_it, scarcity. The client: goal.rs Pilot::screen, session.rs step_host (the FLEET lines).
Scenario. an-expanding-tank-takes-the-depot-at-the-front.toml, run by make scenario S=... [SET="w_front=0"]; in make test with every file of scenarios/.
Families. make families, and SET="dodge_plan=0 duck_plan=0" and make scenario S=... RUNS=1000 SET="dodge_release=0" for the breaks.
Guard. build/lr58/run.sh: a frozen binary; the fair census on naval.map, 900 s, 3 respawns, seeds 0 to 47 in three jobs of 16; trial caps = baseline (the defaults), baseline the same with w_front = 0 and front_memory = 0. The stopped first run (no memory) is in build/lr58/no-memory.
IV. Results
| Step | Measure | Before | After |
|---|---|---|---|
LR 58, w_front | the scenario | nearest depot first | the depot at the front first; fails at 0 |
| fresh picks with a front, a trial match | 2.5 (no memory, 60 matches) | 3.2, 14% with none (memory, 144 matches) | |
| the guard, 48 seeds | -0.14 +/- 0.42 a seat over 48 seeds | ||
| LR 59, floors | five families, 1,000 seeds | run by hand | 31 s, each under its floor broken |
| LR 59, allies' goals | client test | none | the_screen_shows_an_allys_rung_and_target |
| LR 60, tests and docs | brain tests; undocumented | 78; 121 | 81; 99 |
Tests: brain 81 passed and 11 ignored; client 32; net 14; proto 3; sim
- 98 weights, 17 bits; ten scenario files, five families with floors.
V. Discussion
- A front of what is in sight is a flicker. The board reads one member's observation a second; LR 35 found it for the landfall, and the front needed the same memory.
- Count before you census. Expand picks about four depots a match. The meter said in the first minutes of a 45-minute census that the part barely acted.
- Floors make families a regression suite. Five families in 31 s now stand guard over the reflexes the census cannot resolve.
For Paul: none blocking. Next: backlog 51 (the guard posts toward the front), then 20 (island_site where a naval start chooses).