fire-modes: Ultra and Rotary rates priced against the jam roll #

48 matches a side against Princess, --jobs 6, default 8 GiB container cap, on the representative scenarios where both sides carry an Ultra or Rotary autocannon. Same suite, same seeds, current main against this branch. The clean 8 GiB baseline (24/45, 53.3%, no bot exits, 99.8% answered) is the harness reference, not the scenario subset.

Predicted before the run #

1. Jams fall by at least half.
2. Rapid mounts still present and unjammed at the end rise by at least 3 points.
3. Damage and win rate do not fall, and should rise: mean BV left +1-3 against the clean baseline, win rate at or above 53%.

Measured #

	before	after
matches / decided	48 / 45	48 / 46
sds wins	26 (57.8%)	26 (56.5%)
mean BV left, sds / princess	44.2% / 42.6%	43.2% / 42.5%
mean damage inflicted by sds per match	563	615 (+9.3%)
all rapid mounts jammed at end	31 (8.6%)	19 (5.0%)
sds rapid mounts jammed	17 of 190 (8.9%)	4 of 184 (2.2%)
princess rapid mounts jammed	14 of 171 (8.2%)	15 of 199 (7.5%)
sds rapid mounts available (not jammed, not destroyed)	139 (73.2%)	136 (73.9%)
answered / defaulted / illegal / failed	99.91% / 3 / 1 / 0	99.92% / 4 / 0 / 0

Jams on the sds side fell 77%, and pooled 40%; Princess's rate is unchanged, as it must be because its policy did not move. Damage inflicted rose 9.3%, but the predicted win-rate gain did not appear: the win count is identical and the BV differential moved by less than a point, both well inside the interval this sample can resolve. Rapid-mount availability was flat rather than up 3 points.

No harness regression: zero bot exits or failures on either arm, decisions answered above 99.9%, and the after win rate is above the clean 8 GiB baseline.

Why the rate changes #

Ultra still wins most of the time because a double rate costs 1/36 and buys a shot every turn; the model declines it when the magazine is long and the mount is most of the unit's firepower. Rotaries move between 3-shot and 5-shot when the extra two shots clear the higher jam odds, and to Single when they do not. The declared modes in the sds decision logs changed from {Ultra: 181, 3-shot: 113} to {Ultra: 173, 3-shot: 71, 5-shot: 25, Single: 4, 6-shot: 1}.