Mecha Skirmish / Balance study analysisPrototype field notes / v0.6
Balance study analysis
Phase: cautious-policy. Rows are paired on all eight study keys. Draws score one half. Comeback after first loss is descriptive, not causal.
Comparison manifest: /Users/cody/Dev/mecha-skirmish/docs/tuning/evidence/2026-09-09-multivariant/cautious-policy-comparison.json (adaptive-exploratory, SHA-256 ca5834bb555f878fab692b32dd7811f888340cf7667bff722b1227b10190bae8). Aliases below retain their original control identities and phases.
Candidate summary
| Candidate |
Actual control identity |
Panel |
Spread |
Mean abs dev |
Empty Fire |
Dead before attack |
Physical side |
First side |
First-loss win |
Rounds |
Wipes |
| baseline |
safeapproach (cautious-policy) |
partial-observed-panel |
15.8% |
5.5% |
11.8% |
22.4% |
0.4% |
1.6% |
26.8% |
5.95 |
68.9% |
| meleepoolminus1 |
safeapproach+meleepoolminus1 (cautious-policy) |
partial-observed-panel |
17.9% |
6.7% |
11.7% |
22.7% |
0.4% |
1.8% |
26.4% |
6.16 |
65.0% |
| accuracy2 |
safeapproach+accuracy2 (cautious-policy) |
partial-observed-panel |
16.4% |
5.4% |
7.2% |
26.3% |
0.8% |
2.0% |
27.5% |
5.84 |
75.4% |
| meleepoolminus1+accuracy2 |
safeapproach+meleepoolminus1+accuracy2 (cautious-policy) |
partial-observed-panel |
32.9% |
13.9% |
7.2% |
26.3% |
0.5% |
1.7% |
25.1% |
6.01 |
73.4% |
Paired changes from baseline
Positive guardrail changes are harmful. Worst-map columns are the largest same-map increase in absolute bias. Mean absolute bias remains descriptive.
| Candidate |
Spread |
Empty Fire |
Dead before attack |
Worst added physical |
Worst added first |
Mean physical |
Mean first |
| meleepoolminus1 |
2.0% [-5.6%, 7.7%] |
-0.0% [-0.2%, 0.1%] |
0.3% [0.1%, 0.4%] |
0.4% [-0.0%, 1.3%] |
0.8% [0.3%, 2.5%] |
0.0% |
0.2% |
| accuracy2 |
0.5% [-4.5%, 6.0%] |
-4.5% [-5.3%, -3.9%] |
3.9% [3.4%, 4.3%] |
0.8% [0.3%, 2.2%] |
1.1% [0.4%, 3.7%] |
0.4% |
0.4% |
| meleepoolminus1+accuracy2 |
17.1% [11.8%, 22.2%] |
-4.5% [-5.4%, -3.8%] |
3.9% [3.4%, 4.3%] |
0.5% [0.1%, 1.4%] |
1.0% [0.0%, 3.2%] |
0.1% |
0.1% |
Intervals: 95.0% paired seed-block bootstrap, 1,000 replicates. Exploratory only; final inference requires the predeclared 99% intervals.
Exploratory family-payoff intervals
These 95% bootstrap intervals are descriptive and separate from the five primary claims. Point and interval warnings are reported separately against the 35%–65% review band.
| Candidate |
Family |
Point |
95% interval |
Point outside band |
Interval entirely outside band |
| baseline |
mixed |
41.8% |
[38.7%, 45.8%] |
no |
no |
| baseline |
armored |
46.2% |
[43.4%, 50.0%] |
no |
no |
| baseline |
mobile |
39.6% |
[36.7%, 42.6%] |
no |
no |
| baseline |
melee |
53.0% |
[49.8%, 55.3%] |
no |
no |
| baseline |
gunline |
52.0% |
[49.0%, 56.2%] |
no |
no |
| baseline |
control |
55.4% |
[52.9%, 57.9%] |
no |
no |
| meleepoolminus1 |
mixed |
51.7% |
[49.2%, 54.0%] |
no |
no |
| meleepoolminus1 |
armored |
59.6% |
[55.9%, 62.7%] |
no |
no |
| meleepoolminus1 |
mobile |
55.0% |
[51.9%, 57.7%] |
no |
no |
| meleepoolminus1 |
melee |
43.3% |
[40.8%, 45.7%] |
no |
no |
| meleepoolminus1 |
gunline |
56.0% |
[53.3%, 59.7%] |
no |
no |
| meleepoolminus1 |
control |
61.2% |
[57.9%, 64.1%] |
no |
no |
| accuracy2 |
mixed |
49.2% |
[46.9%, 51.8%] |
no |
no |
| accuracy2 |
armored |
54.4% |
[51.3%, 58.6%] |
no |
no |
| accuracy2 |
mobile |
48.0% |
[44.6%, 51.4%] |
no |
no |
| accuracy2 |
melee |
45.5% |
[43.0%, 47.6%] |
no |
no |
| accuracy2 |
gunline |
58.8% |
[56.8%, 61.4%] |
no |
no |
| accuracy2 |
control |
61.9% |
[59.0%, 65.1%] |
no |
no |
| meleepoolminus1+accuracy2 |
mixed |
58.7% |
[57.0%, 60.2%] |
no |
no |
| meleepoolminus1+accuracy2 |
armored |
68.0% |
[66.5%, 69.6%] |
yes |
yes |
| meleepoolminus1+accuracy2 |
mobile |
61.9% |
[60.3%, 63.6%] |
no |
no |
| meleepoolminus1+accuracy2 |
melee |
36.1% |
[34.6%, 37.4%] |
no |
no |
| meleepoolminus1+accuracy2 |
gunline |
62.1% |
[59.7%, 65.0%] |
no |
no |
| meleepoolminus1+accuracy2 |
control |
69.0% |
[65.4%, 72.8%] |
yes |
yes |
Operational and command metrics
| Candidate |
First attack |
Moves |
Move hexes |
Melees |
Locks |
CP grant |
CP spend |
CP waste |
Policy declines |
Budget denials |
CP left |
Nominal mitigation |
| baseline |
2.56 |
128.63 |
128.63 |
2.58 |
5.68 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
| meleepoolminus1 |
2.59 |
130.20 |
130.20 |
3.14 |
6.07 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
| accuracy2 |
2.49 |
126.13 |
126.13 |
2.36 |
5.56 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
| meleepoolminus1+accuracy2 |
2.52 |
127.13 |
127.13 |
2.86 |
5.90 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
0.00 |
baseline: family payoff (observed panel only)
| Family |
Expected score |
Opponents |
| mixed |
41.8% |
1 |
| armored |
46.2% |
1 |
| mobile |
39.6% |
1 |
| melee |
53.0% |
5 |
| gunline |
52.0% |
1 |
| control |
55.4% |
1 |
baseline: map and pairing points
| Map |
X |
Y |
Games |
X score |
Draws |
| causeway |
armored |
melee |
800 |
35.4% |
0.0% |
| causeway |
melee |
control |
800 |
56.4% |
0.0% |
| causeway |
melee |
gunline |
800 |
61.8% |
0.0% |
| causeway |
mixed |
melee |
800 |
30.5% |
0.0% |
| causeway |
mobile |
melee |
800 |
33.9% |
0.0% |
| flat |
armored |
melee |
800 |
53.2% |
0.0% |
| flat |
melee |
control |
800 |
20.5% |
0.0% |
| flat |
melee |
gunline |
800 |
31.6% |
0.0% |
| flat |
mixed |
melee |
800 |
49.2% |
0.0% |
| flat |
mobile |
melee |
800 |
51.0% |
0.0% |
| foundry |
armored |
melee |
800 |
32.0% |
0.0% |
| foundry |
melee |
control |
800 |
57.5% |
0.0% |
| foundry |
melee |
gunline |
800 |
63.2% |
0.0% |
| foundry |
mixed |
melee |
800 |
33.2% |
0.0% |
| foundry |
mobile |
melee |
800 |
32.2% |
0.0% |
| quarry |
armored |
melee |
800 |
64.4% |
0.0% |
| quarry |
melee |
control |
800 |
43.9% |
0.0% |
| quarry |
melee |
gunline |
800 |
35.2% |
0.0% |
| quarry |
mixed |
melee |
800 |
54.1% |
0.0% |
| quarry |
mobile |
melee |
800 |
41.2% |
0.0% |
meleepoolminus1: family payoff (observed panel only)
| Family |
Expected score |
Opponents |
| mixed |
51.7% |
1 |
| armored |
59.6% |
1 |
| mobile |
55.0% |
1 |
| melee |
43.3% |
5 |
| gunline |
56.0% |
1 |
| control |
61.2% |
1 |
meleepoolminus1: map and pairing points
| Map |
X |
Y |
Games |
X score |
Draws |
| causeway |
armored |
melee |
800 |
49.4% |
0.0% |
| causeway |
melee |
control |
800 |
47.4% |
0.0% |
| causeway |
melee |
gunline |
800 |
57.0% |
0.0% |
| causeway |
mixed |
melee |
800 |
41.5% |
0.0% |
| causeway |
mobile |
melee |
800 |
53.0% |
0.0% |
| flat |
armored |
melee |
800 |
62.0% |
0.0% |
| flat |
melee |
control |
800 |
19.4% |
0.0% |
| flat |
melee |
gunline |
800 |
28.7% |
0.0% |
| flat |
mixed |
melee |
800 |
56.9% |
0.0% |
| flat |
mobile |
melee |
800 |
64.4% |
0.0% |
| foundry |
armored |
melee |
800 |
51.2% |
0.0% |
| foundry |
melee |
control |
800 |
51.2% |
0.0% |
| foundry |
melee |
gunline |
800 |
58.0% |
0.0% |
| foundry |
mixed |
melee |
800 |
45.1% |
0.0% |
| foundry |
mobile |
melee |
800 |
48.8% |
0.0% |
| quarry |
armored |
melee |
800 |
75.6% |
0.0% |
| quarry |
melee |
control |
800 |
37.2% |
0.0% |
| quarry |
melee |
gunline |
800 |
32.4% |
0.0% |
| quarry |
mixed |
melee |
800 |
63.1% |
0.0% |
| quarry |
mobile |
melee |
800 |
53.8% |
0.0% |
accuracy2: family payoff (observed panel only)
| Family |
Expected score |
Opponents |
| mixed |
49.2% |
1 |
| armored |
54.4% |
1 |
| mobile |
48.0% |
1 |
| melee |
45.5% |
5 |
| gunline |
58.8% |
1 |
| control |
61.9% |
1 |
accuracy2: map and pairing points
| Map |
X |
Y |
Games |
X score |
Draws |
| causeway |
armored |
melee |
800 |
45.5% |
0.0% |
| causeway |
melee |
control |
800 |
45.6% |
0.0% |
| causeway |
melee |
gunline |
800 |
53.6% |
0.0% |
| causeway |
mixed |
melee |
800 |
40.5% |
0.0% |
| causeway |
mobile |
melee |
800 |
44.6% |
0.0% |
| flat |
armored |
melee |
800 |
61.6% |
0.0% |
| flat |
melee |
control |
800 |
18.0% |
0.0% |
| flat |
melee |
gunline |
800 |
24.8% |
0.0% |
| flat |
mixed |
melee |
800 |
59.1% |
0.0% |
| flat |
mobile |
melee |
800 |
63.6% |
0.0% |
| foundry |
armored |
melee |
800 |
42.5% |
0.0% |
| foundry |
melee |
control |
800 |
50.2% |
0.0% |
| foundry |
melee |
gunline |
800 |
56.0% |
0.0% |
| foundry |
mixed |
melee |
800 |
40.1% |
0.0% |
| foundry |
mobile |
melee |
800 |
36.6% |
0.0% |
| quarry |
armored |
melee |
800 |
68.1% |
0.0% |
| quarry |
melee |
control |
800 |
38.6% |
0.0% |
| quarry |
melee |
gunline |
800 |
30.4% |
0.0% |
| quarry |
mixed |
melee |
800 |
57.2% |
0.0% |
| quarry |
mobile |
melee |
800 |
47.2% |
0.0% |
meleepoolminus1+accuracy2: family payoff (observed panel only)
| Family |
Expected score |
Opponents |
| mixed |
58.7% |
1 |
| armored |
68.0% |
1 |
| mobile |
61.9% |
1 |
| melee |
36.1% |
5 |
| gunline |
62.1% |
1 |
| control |
69.0% |
1 |
meleepoolminus1+accuracy2: map and pairing points
| Map |
X |
Y |
Games |
X score |
Draws |
| causeway |
armored |
melee |
800 |
60.5% |
0.0% |
| causeway |
melee |
control |
800 |
33.9% |
0.0% |
| causeway |
melee |
gunline |
800 |
48.4% |
0.0% |
| causeway |
mixed |
melee |
800 |
50.2% |
0.0% |
| causeway |
mobile |
melee |
800 |
61.1% |
0.0% |
| flat |
armored |
melee |
800 |
72.8% |
0.0% |
| flat |
melee |
control |
800 |
15.8% |
0.0% |
| flat |
melee |
gunline |
800 |
22.4% |
0.0% |
| flat |
mixed |
melee |
800 |
64.0% |
0.0% |
| flat |
mobile |
melee |
800 |
69.4% |
0.1% |
| foundry |
armored |
melee |
800 |
61.1% |
0.0% |
| foundry |
melee |
control |
800 |
42.8% |
0.0% |
| foundry |
melee |
gunline |
800 |
53.2% |
0.0% |
| foundry |
mixed |
melee |
800 |
51.0% |
0.0% |
| foundry |
mobile |
melee |
800 |
56.1% |
0.0% |
| quarry |
armored |
melee |
800 |
77.8% |
0.0% |
| quarry |
melee |
control |
800 |
31.6% |
0.0% |
| quarry |
melee |
gunline |
800 |
27.6% |
0.0% |
| quarry |
mixed |
melee |
800 |
69.4% |
0.0% |
| quarry |
mobile |
melee |
800 |
60.9% |
0.0% |