Devlog: M6

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-15 01:14:10 -05:00
co-authored by Claude Opus 5.5
parent eb239c5424
commit 324bf84dd5
+54
View File
@@ -1568,3 +1568,57 @@ took the deferred bugs, starting with H14 (#81).
This changes the game after the v1.0.0 tag, so it belongs in a 1.0.1 if Eric wants
OpenGL players to have it before the next release.
## 2026-11-15 (small hours): M6, bots that build
The v1.0.0 release is still a draft with no binaries, so R5 waits on Eric. H13 (#80)
hasn't come back, so this run took the deferred M6 (#29). Its commit, eb239c5, is on
`main` after the v1.0.0 tag: the 1.0.0 packages have bots that don't build.
**What bots build.** Every piece goes through the same build command, and the same
`Building.check`, as a player's. The bot checks a stand-in posed where it stands, so it
only presses build when the server would accept.
- **Gathering:** armed and with nothing more pressing, a bot works props with the
boathook, up to 60 of a material (120 late in the match).
- **Cover:** a wall toward a shooter who catches it reloading, from ambush or worn down.
- **Wall and heal:** a badly hurt bot that would break off heals behind a wall instead.
- **Ramps:** a staircase up a cliff, each ramp from the top of the last.
- **Breakwater rings:** for a bot the tide would catch late in the match, and at the
last stand.
**Measuring it.** `make bot-builds` is new. Every other bot builds, and the rest are a
control group.
- **First version:** gathering whenever a prop was in reach. Builders won 15 matches to
25 over 40, and on one seed set lost more fights. The fights they built in went
better, so the cost was gathering: time taken from looting, and standing at a prop
facing away from trouble.
- **Now:** gathering waits 10 s after the last sight of an opponent or hit taken, not
when hurt, and stops at 60.
- **Final 40 matches:**
- builders lost 58% of the fights they built in, against 67% of the rest;
- their mean place was 12.1, against the control's 12.8;
- overall fight losses (65.8% and 65.7%) and wins (20 each) came out level.
- **The gate changed from my first draft.** I'd first required builders to build in a
fifth of all fights. They had the materials in only 23% of fights; most fights are in
the landing. So the gate is now "over a third of the fights they can afford to"
(53% measured). Overall wins aren't gated, because they came out level; that's said
in the tool, BALANCE.md, #29 and #74.
**Gates on eb239c5:**
- `make check`: 261 scripts, 710 tests, smoke ok.
- `make match`: ok.
- `make bot-matches`: 20 with a winner, 3.9% drowned.
- `make balance`: 200 matches, median 13.3 min, sea deaths 3.6%.
- `make bot-builds`: 40 matches, ok.
- Worst p99 tick 3.4 ms.
**Notes:**
- Free memory was around 2 GB all run, so batches ran one process at a time, except
balance with 3 jobs.
- Eric's own terminal sessions running Tidewall on his display were open. I left them
alone. He may have been playtesting; R6 is his.
**Next run:**
1. If the release has its binaries, verify them and close R5.
2. Otherwise M7 (#30), duos and revive, the last open issue in v0.4.
3. Then A12 and A13, tutorial and gamepad (#52, #53).