Abstract
On Paul's choice, landfall_memory 30, landfall_shy 1 and landfall_wait 10 are the defaults. As proof, the census was run in reverse: a team on the old defaults among two on the new. It read -0.08 ± 0.45 a seat, where the forward +2.31 ± 0.58 would predict about -2.3. The gain is relative to opponents that play the old way. What the weights do when every team has them is unmeasured: the census reads a team against the others. The defaults stay; backlog 38 takes an absolute reading.
Index terms: promotion, relative evaluation, interaction effects.
I. Introduction
A. Intent
To set the defaults Paul chose and prove the switch.
B. The prompt
yes all three on. commit. push. continue backlog
The answer to: "I recommend turning all three on ... Which do you want?"
II. Decision Log
| # | Cause | Decision | Not taken, and why |
|---|---|---|---|
| D1 | Paul: "yes all three on" | Defaults 30, 1, 10 in weights.rs | |
| D2 | A plain RuleBot has no Settings | It keeps 0, as with the dodge's constants | Changing RuleBot::new: the baseline bot would change too |
| D3 | A promotion needs a proof | The reverse census: old as trial, new as baselines | A null run: it reads 0 whatever the defaults |
| D4 | F: reverse -0.08 ± 0.45 | Report it with the commit; the defaults stay as Paul chose; backlog 38 | Reverting unasked: the forward gain is real, and the choice is Paul's |
| D5 | "look about brain overview" (CI-004) | LR 20 kept current: 90 weights, the landfall in the diagram, the defaults |
III. Methods
The new defaults dumped (CENSUS_DUMP), with the three set to 0 as the trial's CENSUS_WEIGHTS; the baselines on the defaults. 48 seeds in three ranges, naval.map, 900 s, respawns 3, the fair census.
IV. Results
| Trial | Baselines | d a seat | |
|---|---|---|---|
| Forward (CI-013) | three on | old | +2.31 ± 0.58 |
| Reverse | old | three on | -0.08 ± 0.45 |
Tests: brain 73 passed and 8 ignored; client 31; net 14; proto 3; sim 50; no warnings.
V. Discussion
A relative census answers "does a team with this beat teams without it?", and here the answer is yes. It does not answer "is the game better played when everyone has it?" For a default, which every bot takes, the second question is the one that matters, and needs a reading that is not relative. Every earlier promotion, the capped castle included, was read in one direction only.
Next: backlog 38.