diff --git a/SNNBot_garage/src/SNNBot.nim b/SNNBot_garage/src/SNNBot.nim
index 085e58c..da90e45 100644
--- a/SNNBot_garage/src/SNNBot.nim
+++ b/SNNBot_garage/src/SNNBot.nim
@@ -27,11 +27,12 @@ const
   SPEED_BAND_WIDTH = 1.0      # units/tick per band
   MAX_SPEED       = 8.0
   LEAK     = 0.9         # LIF membrane leak factor
-  THRESH   = 0.2         # ponytail: THRESH=0.2 — unitless system, must match weight scale; raise if neurons fire too much
+  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      = 0.05        # SuperSpike learning rate (r_0 from paper)
-  # ponytail: single learning rate, add RMaxProp optimizer if convergence unstable
+  ETA      = 0.05        # SuperSpike learning rate for hidden→output weights (r_0 from paper)
+  ETA_IH   = 0.005       # ponytail: ETA_IH=0.005 scaled down for 80 inputs; raise if input→hidden converges too slowly
+  # ponytail: separate input→hidden rate; add RMaxProp optimizer if convergence still unstable
   N_INFER  = 10          # inference window ticks per DECIDE
   # ponytail: N_INFER=10, increase if output still noisy; decrease if too slow per tick
   TRACE_DECAY = 0.9      # pre-synaptic trace decay (exponential low-pass)
@@ -184,7 +185,7 @@ proc superSpikeUpdate(snn: var SNN,
     let sg     = surrogateDerivative(vSnap[h])
     let errHid = snn.bFb[h * 2 + 0] * errSin + snn.bFb[h * 2 + 1] * errCos
     for i in 0 ..< N_IN:
-      snn.wih[i * N_HID + h] += ETA * preTrace[i] * sg * errHid
+      snn.wih[i * N_HID + h] += ETA_IH * preTrace[i] * sg * errHid
       snn.wih[i * N_HID + h] = snn.wih[i * N_HID + h].clamp(-W_CLAMP, W_CLAMP)
 
 # ── Bot state machine ─────────────────────────────────────────────────────────