The 10-second frames reproduce the Pass 1 feature table closely enough to be trusted for pooled work. Fuel matches to the cent; distance and hours run slightly low, which is the expected cost of binning a 2.0 s stream to 10 s and of the moving-row filter. ProPower active energy agrees to 0.09%.
| Quantity | From frames | Feature table | Δ |
|---|---|---|---|
| Fuel | 258.59 gal | 258.59 gal | 0.00% |
| Distance | 3,696.7 mi | 3,719.0 mi | −0.60% |
| ProPower AC (active ≥2.0 kW) | 74.81 kWh | 74.74 kWh | +0.09% |
| Elapsed | 72.4 hr | 71.8 hr | +0.8% |
Day 8 Segment 1 carries no TrackLogger partner and Day 6 only 35% coverage, so grade-dependent analysis below runs on 18,533 of 22,235 moving bins (83.4%). Every Phase A leg except Day 8 Seg 1 retains full altitude coverage.
This is the number LightShip asked for, and Campaign 1 could not produce it one leg at a time. Applying the standing skip rule leg by leg, only three of twelve legs clear the reliability gate, and those three disagree by nearly a factor of two.
| Leg | ProPower on | b (gal/hr per kW) | at 5.8 kW | R² | Verdict |
|---|---|---|---|---|---|
| Day 1 — Prescott → Salida | 25.5% | 0.1193 | 0.692 | 0.592 | clears gate |
| Day 2 — Salida → Broomfield | 99.6% | — | — | — | skipped — no ProPower variation |
| Day 3 Run 1 — State B loop | 96.1% | — | — | — | skipped — no ProPower variation |
| Day 3 Run 2 — State A loop | 0.0% | — | — | — | skipped — no ProPower variation |
| Day 4 Seg 1 | 66.8% | 0.2179 | 1.264 | 0.370 | flagged |
| Day 4 Seg 2 | 43.6% | 0.0860 | 0.499 | 0.165 | flagged |
| Day 4 Seg 3 | 9.8% | 0.1374 | 0.797 | 0.403 | clears gate |
| Day 5 — Phase A | 0.0% | — | — | — | skipped — no ProPower variation |
| Day 6 — Phase A | 0.0% | — | — | — | skipped — no ProPower variation |
| Day 7 — Phase A | 0.0% | — | — | — | skipped — no ProPower variation |
| Day 8 Seg 2 | 0.0% | — | — | — | skipped — no ProPower variation |
| Day 8 Seg 3 — final | 14.6% | 0.0783 | 0.454 | 0.630 | clears gate |
Pooling fixes the problem the individual legs have. Six legs ran ProPower and five were deliberate Phase A pure-tow baselines; four legs cycled ProPower on and off within themselves under BMS control. Together they supply 4,006 ProPower-on and 14,527 ProPower-off moving bins — the matched on/off contrast that no single leg ever achieved.
| Specification | b | gal/hr at 5.8 kW | 95% CI | R² |
|---|---|---|---|---|
| Four-term (standing spec) | 0.1311 | 0.761 | 0.632 – 0.888 | 0.514 |
| + acceleration power | 0.1276 | 0.740 | 0.614 – 0.861 | 0.587 |
| + HV battery power | 0.1224 | 0.710 | 0.580 – 0.848 | 0.622 |
| Leg fixed effects (within-leg only) | 0.1220 | 0.708 | 0.549 – 0.861 | 0.596 |
| BMS-cycling legs only, fixed effects | 0.1235 | 0.716 | 0.579 – 0.853 | 0.626 |
| Cruise band 58–68 mph, near-flat | 0.1238 | 0.718 | 0.621 – 0.816 | 0.234 |
Every specification lands between 0.708 and 0.761 gal/hr at 5.8 kW, implying 7.6–8.2 kWh AC/gal. The KB measured baseline is 0.697–0.753 gal/hr and 7.70 kWh AC/gal, derived by a completely different method (paired elevation-corrected round trip). The regression b is now an independent corroboration of that baseline rather than a restatement of it — on prior legs the yield resolved to 7.70 by construction, because ProPower ran continuously.
The stability across specifications is the substantive result. Dropping to within-leg identification only — where every leg-level confound (canopy, tire pressure, fuel batch, terrain, ambient) is differenced out — moves b by 7%, well inside the bootstrap interval. Restricting further to the five BMS-cycling legs, or to matched-speed near-flat cruise, moves it hardly at all.
| Leg | ProPower off | ProPower on | Δ gal/hr | Mean grade power, off vs on |
|---|---|---|---|---|
| Day 1 | 4.13 | 4.63 | +0.50 | +1.0 vs −2.1 kW |
| Day 4 Seg 1 | 3.69 | 4.77 | +1.08 | −3.0 vs −2.8 kW |
| Day 4 Seg 2 | 3.75 | 4.12 | +0.37 | −2.4 vs −2.5 kW |
| Day 4 Seg 3 | 3.75 | 4.25 | +0.50 | −1.3 vs −1.9 kW |
| Day 8 Seg 3 | 3.97 | 4.43 | +0.46 | +0.1 vs −2.4 kW |
The raw differences scatter 0.37–1.08 gal/hr even where grade is well matched, which is why the unmodelled on/off comparison was rejected earlier and stays rejected. It is shown here only to confirm the regression is not manufacturing a sign: every leg with contrast moves the same direction, and four of five sit close to the pooled estimate.
Splitting the grade term into climb and descent halves yields something the campaign was not designed to produce: an empirical marginal conversion between fuel rate and true mechanical power at the wheels, anchored on measured gravitational power rather than on the Engine Power PID.
| Term | Coefficient | Reading |
|---|---|---|
| Climb power | 0.0732 gal/hr per hp | 10.19 kW per gal/hr — 30.2% fuel LHV to wheels |
| Descent power | 0.0382 gal/hr per hp | 52% of the climb slope — regen and braking absorb the rest |
| ProPower | 0.1226 gal/hr per kW | 0.711 gal/hr at 5.8 kW — unchanged by the split |
The ProPower paper's elevation correction used 10.22 kW per gal/hr, taken from the median Engine Power / Fuel Rate ratio — and flagged as suspect there, because Engine Power appeared to be computed from fuel flow rather than measured independently. This campaign confirms the suspicion (Engine Power and Fuel Rate correlate at r = 1.000 across 22,235 moving bins) and then independently reproduces the constant at 10.19 kW per gal/hr from gravitational power alone. The paper's Equation 15 survives; its stated uncertainty about that step can be closed.
The 80% ratio between ProPower's 8.16 kWh AC/gal and the 10.19 kWh wheel/gal figure is the honest way to price in-motion generation against propulsion: exporting a kilowatt-hour as AC costs about a quarter more fuel than putting one through the driveline.
| Leg | Hours armed-idle | Mean kW | kWh | Fuel-equivalent |
|---|---|---|---|---|
| Day 1 | 9.26 | 0.99 | 9.13 | 1.19 gal |
| Day 4 Seg 1–3 | 5.58 | 1.03 | 5.63 | 0.73 gal |
| Day 8 Seg 3 | 1.47 | 0.98 | 1.44 | 0.19 gal |
| Day 2, Day 3 Run 1 | 0.18 | 0.97 | 0.17 | 0.02 gal |
| Campaign | 16.5 | 0.99 | 16.39 | 2.13 gal — 0.8% of campaign fuel |
Day 1 alone accounts for 56% of the campaign's armed-idle waste. By the final leg the converter was being switched off between charging windows, and the overhead falls to 0.19 gal across 400 miles. Cheap to eliminate, and worth stating in the pitch as an operating discipline rather than a system limitation.
The rev-2 observation that the worst thermal event happened at 78.8°F ambient now generalises across the whole campaign, and it survives conditioning.
| Altitude | Load | OAT <70 | 70–80 | 80–90 | >90°F |
|---|---|---|---|---|---|
| <2,500 ft | 40–60% | 199.5 | 205.6 | 203.6 | 201.6 |
| <2,500 ft | >80% | 204.2 | 208.7 | 207.5 | 204.8 |
| 5,000–7,500 ft | 40–60% | 209.0 | 208.3 | 206.3 | 200.5 |
| 5,000–7,500 ft | >80% | 216.8 | 219.5 | 216.6 | 211.3 |
| >7,500 ft | 40–60% | 204.9 | 219.3 | 208.1 | 215.0 |
| >7,500 ft | >80% | 223.9 | 225.4 | 222.1 | 220.5 |
Read across any row and ambient temperature does essentially nothing. Read down the table and altitude adds roughly 20°F; read within an altitude band and load adds roughly another 20°F. The multivariate fit says the same thing:
ECT = 184.8 + 1.20 × (altitude, kft) + 0.187 × load% + 0.144 × mph − 0.052 × OAT (R² = 0.325, N = 19,589).
+1,000 ft is worth +1.20°F. +10% load is worth +1.87°F. +10°F of ambient is worth −0.5°F — not a cooling effect, but a fair statement that ambient carries no independent signal over the range this campaign sampled. Bins above 230°F averaged 8,264 ft and 112% load at 79.8°F ambient; bins below the thermostat spec averaged 3,856 ft and 59% load at 80.8°F ambient. Same ambient, opposite thermal outcome.
| Metric | Campaign | Worst leg |
|---|---|---|
| ECT peak | 241.2°F | Day 7 |
| Minutes above 217.4°F | 586 | Day 1 (149.5) |
| Minutes above 230°F | 15.7 | Day 1 (5.8), Day 7 (5.0) |
| Minutes above 240°F | 0.2 | Day 7 |
| IAT2 peak / minutes above 140°F | 197.6°F | 27 min campaign-wide |
The TurboAssist trigger currently lowers its State-A threshold above 86°F ambient. Across 19,589 moving bins that variable predicts engine coolant temperature with the wrong sign and negligible magnitude. An altitude term, or ECT itself, is what the trigger needs. Stated plainly for the pitch: the assist should fire on the climb, not on the forecast.
| Leg | Bins (10 s) | Mean λ | Peak IAT2 | Mean altitude | Mean OAT | Mean load |
|---|---|---|---|---|---|---|
| Day 1 | 6 | 0.90 | 143.6°F | 9,596 ft | 80.7°F | 121% |
| Day 2 | 6 | 0.90 | 154.4°F | 8,423 ft | 87.8°F | 118% |
| Day 7 | 11 | 0.80 | 197.6°F | 7,011 ft | 81.3°F | 120% |
Total qualifying time is 3.8 minutes and roughly 0.20 gal of excess fuel across 3,719 miles. Mean enrichment fraction over those bins is 18.3%. Pass 1 had found only the Day 7 episode; pooling surfaces two more, and all three share the same signature — 7,000 to 9,600 ft, sustained load above 115%, ambient in the low 80s.
Two of the three episodes occurred below the 86°F ambient threshold the trigger uses. The mechanism is real and measurable on this platform, but it is a high-altitude grade phenomenon, not the hot-plains phenomenon the protocol was written to catch. In absolute fuel terms it remains small; its value to the pitch is as an engine-protection argument, not a fuel-saving one, and that framing should be made explicit rather than left for a reader to infer.
The rev-2 rule required that any future drift claim condition on a consistent RPM/load band before being reported. Conditioning on RPM 1,500–2,000 and load 40–60% (2,441 bins) confirms the retirement: the B2−B1 gap scatters 0.19 to 2.57 with a slope of +0.019 per leg and a correlation of 0.080 against leg order. There is no bank-asymmetry trend.
Conditioning also surfaces something the unconditioned means were hiding. Both banks move together, and they move with altitude:
| Leg | Mean altitude | Common-mode LTFT | B2 − B1 gap |
|---|---|---|---|
| Day 1 | 6,752 ft | +2.71 | 0.80 |
| Day 2 | 6,644 ft | +0.51 | 0.19 |
| Day 4 Seg 1 | 4,063 ft | −0.29 | 1.39 |
| Day 4 Seg 3 | 1,283 ft | −2.22 | 2.57 |
| Day 5 | 1,008 ft | −5.60 | 2.47 |
| Day 6 | 936 ft | −3.13 | 0.54 |
| Day 7 | 5,584 ft | −1.67 | 0.95 |
| Day 8 Seg 3 | 5,570 ft | +0.44 | 1.50 |
Common-mode long-term trim tracks mean leg altitude at r = 0.833, against r = 0.346 for ambient temperature. The nine-point excursion that looked like drift is an altitude effect: the trim table is indexed by RPM and load, and load is a normalised airflow term, so the same table cell is reached at very different barometric pressures on a mountain leg and a plains leg.
Altitude and leg order are partly confounded in this campaign — mountains at both ends, plains in the middle. What separates the two hypotheses is the recovery: a genuine sensor drift would not return to its Day 1 value, and common-mode trim came back from −5.60 on Day 5 to +0.44 on the final leg as the rig climbed back into the Rockies. That is the altitude pattern, not a drift pattern. Recorded as an explanation, not a new watch item.
Pass 2 does not reopen it. The pooled fuel-side model does produce a positive, physically-signed aerodynamic term where the power-balance regression produced a negative one, but its magnitude is not credible and it is not a measurement.
| Route | Result | Status |
|---|---|---|
| Power balance, Pwheels/v against v² (Pass 1) | −0.136 ± 0.303 m² | Interval contains zero |
| Fuel-side v³ term × measured climb slope (Pass 2) | 1.57–1.83 m² depending on air density | Positive but ~45% of expected |
| Expected for this rig | ~3.9 m² | — |
The reason is unchanged and is not fixable in analysis. Across near-flat moving bins the speed interquartile range is 59.4 to 63.3 mph and 55.9% of the data sits inside a single five-mph window. A v³ term and a constant cannot be separated over that range, so the fitted coefficient absorbs the rolling-resistance term and whatever else varies weakly with speed. It is reported here so that the number is not rediscovered later and mistaken for progress.
The Method-2 speed-hold protocol remains the only path: 50/55/60/65/70 mph, at least 90 s each, first 15 s trimmed, on a flat low-traffic segment. Half an hour of deliberate speed variation carries more leverage than the 504 minutes of incidental flat cruising Campaign 1 collected. Schedule it in the first two days, not on a convenient leg.