feat(SNNBot): adaptive learning dead zone scaled by enemy velocity

Stationary targets get large dead zone (3°) for stable aim. Fast movers
get small dead zone (0.5°) for rapid adaptation. Replaces fixed
threshold and jump-reset hack.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

SNNBot_garage/src/SNNBot.nim

@@ -33,7 +33,6 @@ const
   LEAK     = 0.9         # LIF membrane leak factor
   THRESH   = 0.08        # ponytail: THRESH=0.08 tuned for 80-input layer; raise if firing saturates
   MAX_GUN_TURN = 20.0    # max gun turn per tick (degrees)
-  AIM_TOL  = 2.0         # arrive tolerance (degrees)
   ETA      = 0.1         # SuperSpike learning rate for hidden→output weights (r_0 from paper, bumped 2x)
   ETA_IH   = 0.02        # ponytail: ETA_IH=0.02 bumped 4x for input→hidden (was 0.005, too conservative)
   # ponytail: separate input→hidden rate; add RMaxProp optimizer if convergence still unstable
@@ -215,7 +214,6 @@ type
     velSpeed:        float64            # speed (units/tick) from last scan delta
     lastDecideGunDir: float             # gun heading captured at DECIDE time for EVALUATE
     lastBinInput:    BitVec80           # DECIDE-time binary input, reused in EVALUATE
-    lastError:       float              # last aiming error (for jump detection)
 
 # ── aimTo helper ──────────────────────────────────────────────────────────────
 
@@ -274,7 +272,6 @@ method onRoundStarted*(bot: SNNBot, e: RoundStartedEvent) =
   bot.velSpeed     = 0.0
   bot.phase        = DECIDE
   bot.tick         = 0
-  bot.lastError    = -1.0      # sentinel: no error yet
   setTargetSpeed(0.0)
   setTurnRate(0.0)
 
@@ -385,21 +382,18 @@ method run*(bot: SNNBot) =
       aimTo(bot.targetAngle, gunDir)
       let err = abs(normalizeRelativeAngle(bot.targetAngle - gunDir))
 
-      # Bidirectional fire gate: fire only when error within tolerance
+      # Adaptive dead zone: scale learning threshold by enemy velocity
-      if err < AIM_TOL:
+      # Stationary → 3° (stable), full speed (8 units/tick) → 0.5° (fast tracking)
+      let t = (bot.velSpeed / 8.0).clamp(0.0, 1.0)
+      let adaptiveDeadZone = 3.0 + (0.5 - 3.0) * t  # lerp(3.0, 0.5, t)
+
+      # Fire gate: only fire when error within adaptive dead zone
+      if err < adaptiveDeadZone:
         discard setFire(FIRE_POWER)
         bot.phase = EVALUATE
       else:
         discard setFire(0.0)
 
-      # Fast recovery: detect error jump (regime change) and reset ring buffer
-      # ponytail: >5° jump triggers reset; tune threshold if target moves frequently
-      if bot.lastError >= 0.0 and err - bot.lastError > 5.0:
-        bot.res.count = 0
-        bot.res.nextSlot = 0
-
-      bot.lastError = err
-
     of EVALUATE:
       # Predictive error signal: extrapolate enemy position at bullet impact time.
       if bot.hasLastPos:
@@ -409,12 +403,17 @@ method run*(bot: SNNBot) =
         let futureY = bot.lastEnemyY + sin(velRad) * bot.velSpeed * travelTime
         # correctAngle: absolute world-frame bearing [0,360) to predicted enemy position
         let correctAngle = directionTo(myX, myY, futureX, futureY)
+
+        # Adaptive learning dead zone: scale by enemy velocity (same as fire gate)
+        let t = (bot.velSpeed / 8.0).clamp(0.0, 1.0)
+        let adaptiveDeadZone = 3.0 + (0.5 - 3.0) * t  # lerp(3.0, 0.5, t)
+
         when USE_RESERVOIR:
           let aimRaw = bot.res.forward(bot.lastBinInput)
-          # Dead zone: only learn if aim error > 1.5° (prevents ring buffer churn)
+          # Only learn if aim error exceeds adaptive dead zone
           if aimRaw >= 0.0:
             let aimErr = abs(normalizeRelativeAngle(aimRaw - correctAngle))
-            if aimErr > 1.5:
+            if aimErr > adaptiveDeadZone:
               bot.res.learn(bot.lastBinInput, correctAngle)
           else:
             # Cold start: always learn