diff --git a/spark-physics.txt b/spark-physics.txt
index 194b9b5..2b255b1 100644
--- a/spark-physics.txt
+++ b/spark-physics.txt
@@ -268,8 +268,8 @@ Requires calibration per coil. Starting values:
 **QCW-style growth:**
 - ε ≈ 5-15 J/m
 - Long ramp times (5-20 ms)
-- Leader-dominated channels
-- Energy efficiently extends length
+- Leader-dominated channels: low resistance, thermally persistent
+- Lowest ε because leaders don't require re-ionization — energy goes to forward propagation
 
 **High duty cycle DRSSTC:**
 - ε ≈ 20-40 J/m
@@ -279,9 +279,9 @@ Requires calibration per coil. Starting values:
 
 **Hard-pulsed DRSSTC (burst mode):**
 - ε ≈ 30-100+ J/m (single-shot)
-- Short pulses, mostly streamers
-- Much energy → brightening/branching
-- Poor length efficiency
+- Short pulses, mostly streamers: high resistance, short-lived
+- Much energy → re-ionization overhead, brightening, branching
+- Highest ε because channels cool between bursts and must be rebuilt each pulse
 
 **Advanced refinement:** ε decreases during heating due to thermal accumulation:
 ```
@@ -355,7 +355,10 @@ QCW uses voltage ramps of 10-22 ms at 300-600 kHz to grow thermally persistent l
 - High peak current → bright, thick but short
 - Voltage collapse limits length before leader formation
 - Growth saturates at ~80 μs ON time (burst ceiling)
-- Short bursts of high peak power outperform long bursts of low peak power at the same total energy (Steve Conner)
+
+**Within burst mode**, short bursts of high peak power outperform long bursts of low peak power at the same total energy (Steve Conner). A 100 μs burst works better than 150 μs at the same energy because higher peak power pushes the initial streamer further before the 80 μs thermal ceiling hits. Beyond one thermal time constant, additional drive just reheats decaying channels rather than extending the spark.
+
+**Note:** This is a within-mode optimization (burst vs burst). It does NOT mean burst mode is more efficient than QCW. Across modes, QCW leaders (ε ≈ 5-15 J/m) are far more efficient per meter than burst-mode streamers (ε ≈ 30-100+ J/m) because leaders are thermally persistent, low-resistance channels that don't require re-ionization. The key distinction: burst mode peak power helps overcome the propagation field threshold within a single shot, while QCW's sustained drive enables a fundamentally different (leader) plasma type with lower energy cost per meter.
 
 ### 5.5 Streamers vs Leaders