Post-adoption balance and item-pricing checks — 2026-09-09

The owner requested balance checks after adopting ranged Accuracy +2 and paid elevation movement, plus lessons for designing the real parts and weapons. This is a bounded exploratory study. It changes no playable data or rules and remains on the local review branch. TASK-093 tracks delivery. Historical evidence and the unused 600000-range confirmation data stay untouched.

Full-game adoption comparison

Compare four explicit conditions on the same repaired maps and current engine: current (Accuracy +2, paid movement), accuracy only (new catalog, legacy movement), movement only (archived pre-accuracy catalog, paid movement), and pre-adoption (archived catalog, legacy movement). Other rules, pilots, rosters and controller options must be identical. In particular, preserve the selected permanent offensive Enraged and live-default AI routing flags; do not silently enable the experimental safe-approach controller.

Each condition covers all 15 distinct pairs of the six existing legal roster families, four maps, eight shared seeds 1100000–1100007, and roster at A/B crossed with first side A/B. That is 1,920 games per condition/profile. Run two separate declared profiles: fixed .75/.75 as the reference, and the prior study's locked control-selected profile as transferred-policy sensitivity. The latter was trained on a different controller and movement rules, with only 20 cells trained and 40 .75/.75 fallbacks. It is not optimized current counterplay. Do not average the two profiles into a single balance score. Total: 15,360 game executions.

Run current-rules mirrors separately: six families × four maps ×16 fresh seeds1110000–1110015 ×two first sides, fixed .75, for768 games. Report physical-side and first-side results separately, with draw and engagement rates so non-engaging draws cannot masquerade as fairness.

Retain exact configurations, source hashes, all rows, commands and completion counts. Validate complete factorial keys and before/after source hashes. Payoffs count draws as half a win. Equal-weight maps/opponents for family summaries; give per-map matchups too. Retain empty proactive Fires, deaths before first proactive attack, first-death timing including no-death games, wipe rate, game length, movement, melee exchanges and objective activity. First-loss outcomes are descriptive associations, not causal estimates of losing a mech.

Exploratory paired intervals use 1,000 whole-seed bootstrap replicates at95%, retaining every linked condition/map/pair/assignment within each seed. Eight shared blocks do not support acceptance claims merely because the total execution count is large. Calculate factorial interactions on matched payoff/outcome cells, never by adding changes in nonlinear family spread. No automatic candidate promotion or extra tuning ladder follows this stage.

Weapon and part marginals

A separate read-only audit measures eight standard ranged profiles on a declared legal shared platform across five stock targets, three cover states and standing/moving contexts. Six profiles per cell are base,+1Accuracy,+1Pen,+1damage,+1projectile and−1Heat:1,440 exact printed-damage rows. State exclusions of critical extras, finite target durability, Heat, loads and Overdrive wherever the exact helper omits them. Heat should be an exact negative control in this immediate calculation, with its value measured separately.

Persistent engine sequences cover Rifle/Cannon/Pulse ×Courier/Foreman/Bastion ×clear/hard cover ×base/−1Heat ×spare-Quick-Vent/no-Vent ×128 shared seeds:9,216 trials of at most eight activations. These are isolated attacks on passive targets, not games or duel results. Retain kill/censoring, actual damage, Heat, firing and Vent counts; stop on death and preserve sequence semantics. Pin exact seeds and scripts in the audit manifest before outcomes.

Exact melee budgets cover dual Blade/Spike, printed pool, arm Handling and Piloting changes against stock targets. Distinguish initial resources from complete exchanges: the same die cannot simultaneously strike and parry. Preserve initiative and Piloting breakpoint effects. Conservative part/weapon dominance checks must retain categorical abilities, mounting constraints, capacity, weight and survivability rather than reducing everything to damage per weight.

Mobility and practical pricing

A separate deterministic geometry audit compares paid ordinary and Jump costs to every complete objective footprint from all legal deployment cells on both physical ends. Inspect fixed, unbuffed individual movement allowances of2–6 and optimistic one-to-three-activation cost bounds. Enraged movement bonuses are excluded; a printed Speed2 mech with an active +2 ordinary Move bonus has a larger allowance and can cross edges that this allowance2 model cannot. These are optimistic route bounds without enemies, reactions or competing actions. Flat has no objective rows and is explicitly excluded from zone-access denominators. Cross-check shortest fields against actual reachable paths and rotational symmetry. A saved activation or enabled shot is a threshold effect; one movement point has no universal price.

The deliverable is a current baseline report plus a reusable pricing method: offense/reliability, survival/disable, mobility/access, Heat/action costs and team utility. Keep context matrices, legal whole-build substitutions and breakpoint flags. Capacity, chassis/team caps, ammunition limits, two-handed occupancy and pilot/support interactions constrain useful power. A low damage-per-weight ratio alone is not proof a part is weak. Independent review must check both the measurements and the generalizations before delivery.