Forecast all-down recovery and preserve self-rescue; record Workshop 0.0.133

docs/evidence/recovery-forecast-2026-10-31/README.md

@@ -0,0 +1,32 @@
+# Recovery forecast validation — 2026-10-31
+
+Production policy is evaluated once per second by authority. A known forecast
+beyond 300 seconds uses the existing 30-second defeat grace. Usable self-injection,
+a conscious reserved rescuer, and evacuation preserve their respective safeguards.
+Unsupported/uncertain medical states use the previous recovery checks.
+
+The snapshot simulator accounts for zone multipliers, thresholds, pending regen
+delays, bleeding/tourniquets, and finite medicine duration. Changing blood bands
+and post-treatment passive timing remain conservative: they can fall back to
+unknown rather than force a speculative defeat. This is not a universal precise
+ETA for arbitrary medical mods.
+
+`test_recovery_forecast.py` executes the production model as C++: joint thresholds,
+exact 300-second boundary, delayed regeneration, medicine exhaustion, bleeding,
+self-rescue and unknown-state policy. Existing self-epinephrine, recovery-policy,
+player-medical, spectator-recovery, reconnect-recovery and rescue-recovery checks
+passed. Generated-menu and changelog checks also passed.
+
+The disposable native fixture compiled and completed. A real protected patient
+received the ordinary finite epinephrine effect: predicted wake 19 seconds,
+observed wake 19.0837 seconds. Snapshot inspection left actual blood unchanged.
+The fixture injects the item effect at its native completion boundary; it does not
+claim human Space input, two connected clients, or a multiplayer defeat screen.
+The available dedicated runner is isolated with no clients, so dedicated-server
+multiplayer acceptance remains open.
+
+Earlier iterations caught missing native hit-zone multipliers and over-counting
+passive resilience recovery during treatment. The final run has no script errors.
+The strict runner still exits 1 for the 13 existing localization/world diagnostics
+retained in native-checks.log; feature assertions completed, cleanup succeeded,
+and the tested sources remained unchanged. Validation preceded the Workshop upload and Git push.