A–W Methods & Evidence

• pp (percentage points): If baseline wins 45% and A–W wins 70%, Δ = +25 pp.
• Win rate (WR): Share of rounds that Blue wins in the surrogate (non-kinetic).
• Paired A/B design: Same scenario, same random seed, run twice: once with baseline, once with A–W.
• 95% confidence interval: A statistical range around a percentage; we use Wilson intervals.

We evaluate in a laser-tag–style training surrogate (no weapons, no kinetic modeling). For each scenario (terrain, weather, EW conditions, force ratios):

1. Run Baseline.
2. Run A–W under the same seed and conditions.
3. Record WIN/LOSS and round-level attributes to CSV.
4. Repeat thousands of times; compute WR for each arm and Δ in pp with 95% CIs.

Because both arms use identical seeds, differences isolate the contribution of A–W.

• Δ at parity: Pass if Δ ≥ +20 pp at equal force size, with 95% CIs reported.
• Safety: ≤ 1% leakage across ≥ 1,000 probes with deny-first policies on.
• Reproducibility: Your rerun on air-gapped hardware matches within the reported CI.

• ISR triage: surfaces the right items first; fewer misses, faster human review.
• Fusion & handoff: keeps tracks coherent; fewer drops/duplicates across teams.
• COA generation & self-critique: proposes courses of action and auto-red-teams before a human decides.
• Tasking & deconfliction: schedules who does what, avoiding friendly conflicts.
• Resilient comms choices: prefers robust routes/frequencies under interference.
• Logistics & routing: keeps people, fuel, and parts synchronized.

We provide a turnkey kit so your lab can re-run the study offline:

• Scripts + fixed seeds for parity/2×/3× blocks
• CSV/XLSX logs (one row per round) + summary.json
• Governance policies (deny-first, kill-switch, purge, trace logs)
• “How to Grade” insert (acceptance gates, where to look)

Outcome: either the surrogate deltas replicate within CI on your hardware, or they don’t. The kit is designed to make that determination quickly and cleanly.

A–W runs as a sidecar next to your existing C2/mission tools:

• Inputs: mirrors existing feeds (sensor summaries, tracks, chat, reports)
• Tools: calls only whitelisted local tools (routing, link planning, reporting)
• Outputs: pushes traceable recommendations back into your displays for human approval
• Security: default offline; no external network calls