The rig under test: Ford F-150 PowerBoost (3.5 L EcoBoost + MGU, ProPower Onboard) towing the LightShip AE.1 — 15,320 lb / 6,949 kg combined, CAT-scale basis. NACS umbilical runs from the truck bed to the AeroHub receptacle on the trailer nose.

Prescott → Oshkosh → Prescott

The route ran Prescott AZ → Salida CO → Broomfield CO (the LightShip Energy pitch, plus two back-to-back calibration loops) → North Platte NE → Aurora NE → Adel IA → Oshkosh WI for EAA AirVenture → Grand Island NE → western Colorado → Moab UT → Prescott AZ. The planned Fayetteville AR waypoint was dropped in-trip. Planned distance was roughly 3,000 miles; actual was 3,719.0.

The rate consistency deserves emphasis because it survived the campaign's hottest conditions. On Day 4, the three consecutive segments ran at 5.83 → 5.82 → 6.06 kW while OAT max climbed 89.6 → 93.2 → 96.8°F. Duty cycle fell hard across those same segments — but the rate did not move. Duty is governed by the trailer's battery-management setpoints and by driver command. The rate is not modulated by heat.

The three converter states

Why the threshold matters: setting the "active" cutoff at 0.5 kW instead of 2.0 kW merges the two services and overstates bulk charge by up to 50% on a low-duty leg. Day 1 reads 27.68 kWh combined against 18.48 kWh of genuine bulk charge. The 2.0 kW threshold is now a locked convention.

The sign is stable across the entire plausible range. But the sign depends on one assumption that is worth stating out loud every single time: valuing trailer kWh at the ProPower recharge chain (7.70 kWh AC/gal) is only correct when the truck is the sole charging source. With destination charging, that energy costs grid rates and the net flips positive. This is the Case 1 / Case 2 split that was already in the energy-balance model — the calibration pair is the first measured evidence for it.

Enrichment — one clean episode

The campaign produced exactly one qualifying sustained rich-λ episode under the filter (λ < 0.95, load > 60%, IAT2 > 140°F, single-sample DFSO transients excluded): 33 rows / 1.1 minutes, mean λ 0.824, enrichment fraction 21.4%, peak IAT2 197.6°F at 7,362 ft, Day 7. Small in absolute fuel terms, but it demonstrates the mechanism is real and measurable on this platform — and, again, it landed on a mountain grade rather than the hot plains the protocol was aimed at.

Everything else held. No frame drops, no PID dropouts, no logger crashes on the OBD side, no charge faults elsewhere, no thermal derate.

One thing did survive. The regression intercept gives Crr = 0.0197 ± 0.0028 — physically plausible for a heavy trailer at highway speed. Rolling resistance was never the blocked term; only the aero split was.

The consequence propagates: the trailer road-load figure (~14.8 kW at 62 mph) stays modeled everywhere it appears, including inside the TurboAssist energy balance. Campaign 2's Method-2 speed-hold protocol — 50/55/60/65/70 mph, ≥90 s each, first 15 s trimmed — is the only path. One deliberate 30-minute session beats seven hours of incidental cruising. Campaign 1 deferred it to a convenient leg and got nothing.

A corrected-coastdown alternative was considered and rejected on the arithmetic: MGU regen force (−10 to −14 kW routinely, up to −38 kW) is the same magnitude as or larger than the trailer road load being sought, so the method subtracts one large noisy number from another to recover a small difference. Error amplification, not sample size, is the killer.

Clocks — the most common defect in the campaign

Four of the ten legs with a GPS partner file carried a one-hour clock error, in both directions. The pre-trip rule — infer the timezone from the first GPS fix and convert the filename stamp — fails whenever a device carries a stale zone across a state line. On Day 4 the OBD logger held MDT all day while physically in CDT Nebraska from Segment 2 onward; the old rule would have produced a one-hour error in the opposite direction from the real one.

Other data defects found and handled

• OAT ceiling saturation. The outside-air sensor pins at exactly 419.0°F. 685 such rows on Day 1 alone, 761 campaign-wide. A floor-only dropout filter passes these straight into any air-density or thermal calculation. Both ends now filtered: OAT ≤ 1 and OAT ≥ 120.
• TrackLogger died silently on Day 6. Clean 1 Hz with no gaps until it simply ended at 09:33, 4.16 hours in — 214.4 of 660.1 miles covered, 67.5% of the leg lost. Nothing in the field checklist catches a logger that stops looking healthy. Now a check at every fuel stop.
• OBD gap between files. Day 1 has a 2.3-minute gap with 1.58 unrecorded miles between files 1 and 2. Confirm the recording is running after every stop.
• OBD GPS is a usable fallback. On the Day 6 stretch with no TrackLogger, OBD-GPS-derived speed tracked the speed PID at +0.44 mph mean / 1.31 mph std over 12,978 rows. Campaign 1 files carry true 6-decimal position. Legacy pre-trip files truncate to 2 decimals and remain unusable for heading.
• Frame rate held. All 15 files: 43 columns, median Δt 2.000 s, std 0.0028 s, 100% of intervals within [1.9, 2.1] s.
• One TrackLogger file unaccounted for — 13 expected, 12 received. Still open.

The wind pipeline, retired

The Open-Meteo HRRR anchor-link round-trip — generate URLs, download in browser, upload back for merge — was specified pre-trip and failed in practice on most legs. No usable wind data was collected. This is a methodology finding, not housekeeping: a per-leg manual download round-trip is not a viable field workflow. If wind is reinstated, it must be either an onboard measurement or a bulk post-hoc archive pull. The Section 03 wind cross-check is removed from the report template, and wind_residual_pct stays blank — not zero — on every Campaign 1 ledger row.

The open question: does the truck's "Hybrid/EV Battery Power" channel measure total high-voltage bus draw, or only the traction motor? If traction-only, the traction-load residual formula would double-subtract the accessory branches and be invalid — and the wheel-power term that Cd·A depends on would be untrustworthy.

The field intuition, built over years of driving, said traction-only: the meter swings wildly with acceleration and braking and appears never to flinch when a several-kilowatt ProPower load lands on the bus. That observation was real but non-discriminating — traction swings ±20–35 kW routinely, so a few-kW accessory step is simply lost in the visual noise.

Consequences: the traction-load residual formula stands, validated. The latent double-subtraction flag is logged as raised and resolved. The power-balance wheel-power term is sound — which clears the exact prerequisite that Cd·A's second run was deliberately waiting on. The method generalises, and is now written up as a reusable protocol: one continuous engine-in-the-loop file, pre-charge the pack high, apply a known load in the engine-off window, difference against the measured quiet baseline rather than against zero.

Two smaller corrections came out of the same file: the A/C compressor is a high-voltage device on the bus, not a 12V load off the DC/DC converter (the converter acts as a pass-through junction, which is what caused the earlier misstatement), and the 44-PID configuration passed its static frame-rate verification — median Δt 2.000 s, std 0.0031 s, 100% in band. That verification still has to be repeated under towing load before a full Campaign 2 leg is trusted.

The bus, drawn

Confirming that the meter reads the whole bus only matters if you know what is hanging on the bus. The topology below is factory-documented (414-03A / 414-05 / 303-01G), with the rev-4 compressor correction applied. Everything on it is ultimately crankshaft work: the pack is a buffer, not a source, so on a tow leg every kilowatt-hour that leaves the bus is a kilowatt-hour the engine put there.

What actually pulls on the bus

Five branches, and Campaign 1 could resolve exactly one of them. Magnitudes below come from the reduced campaign frames except where noted as bench observations, which are single parked measurements and carry no leg-scale weight.

The asymmetry is the point. ProPower is the one branch Campaign 1 metered directly, and it is also the one branch that is not a truck-side inefficiency — it is deliverable product. The two branches that consume fuel purely to run the truck, DCDC and ACCM, were invisible for all 3,719 miles. That is why the campaign can state ProPower's cost to four significant figures and cannot state the accessory cost at all.

Why this is the hot-weather mileage question

The cooling burden loads the bus twice, and the second one is not optional. The compressor serves cabin cooling and, via a BECM-controlled coolant diverter valve, the battery chiller — the factory description has the valve routing coolant through the chiller when extra cooling is needed under high ambient temperature and/or high current demand. So the compressor cannot be switched off the way a cabin load can: turning cabin cooling down still leaves it cycling for the pack, and the battery circuit's documented upper limit is 131°F.

That "and/or high current demand" clause carries a consequence for the pitch that Campaign 1 could not test. Bulk-charging the trailer at 5.8 kW pushes sustained current through the pack; sustained pack current is one of the two conditions that calls for chiller duty; chiller duty is compressor draw; compressor draw is more bus load and more engine work. If that loop is real, V2V charging carries a second-order cooling cost in hot weather on top of its measured 7.70 kWh AC/gal, and the measured figure is a floor rather than a full cost. This is a hypothesis with a documented mechanism behind it, not a finding — it should be stated that way until 44-PID data exists.

What Campaign 1 can and cannot say about it

It cannot say much, and the reason is worth stating plainly rather than glossing, because the obvious analysis gives the wrong answer with the wrong sign.

The ProPower-off legs remove that confound but leave terrain in its place. Across the four usable Phase A highway legs, mpg spans 14.03 to 16.53 with no ordering by ambient at all — the lowest is Day 7 at 89.6°F, which is the high-altitude mountain leg, and the highest is Day 8 Seg 1 at 84.2°F on flat ground. The Day 5 / Day 6 pair (15.43 mpg at 95.0°F, 15.27 at 80.6°F) is the cleanest comparison in the campaign and agrees within 1% across a 15°F spread.

Read honestly, that says: whatever the hot-weather accessory penalty is, on flat plains legs it is smaller than the leg-to-leg terrain spread, and Campaign 1 cannot separate it from that spread. It does not refute the mechanism. It does bound the size of what we are hunting, and it means the number cannot be recovered by any reanalysis of this dataset — the same conclusion Cd·A reached, for the same structural reason.

What the new PIDs settle, and what still blocks them

Campaign 2's two added channels — DCDC HV-side current and DCACA HV-side current — convert this from an argument into an accounting. The DCDC channel is documented as covering the entire downstream 12V draw as seen from the bus, so it captures the low-voltage half of the cooling burden (blower, controls) without further inference. With both channels populated, every branch except traction is directly metered, traction falls out as the validated residual, and the six-term nightly regression can carry e (DCDC) and f (HVAC) as real coefficients instead of omitted terms.

Two things still gate it, and neither is fixed by driving more.

Both blockers land in the same parked session, and that session is free — no fuel, no route, no weather window. It should run before Campaign 2 departs, not during it. The lesson Campaign 1 taught about deliberate experiments is that they do not happen opportunistically.

The Phase A baseline is the most robust number in the campaign. Day 5 (15.43 mpg) and Day 6 (15.27 mpg) cover 1,080 miles across a 15°F ambient spread and agree to within 1%. Everything else in the fuel accounting is measured against that.

Day 8 Seg 2 (8.94 mpg over 14 miles) is a 17-minute repositioning segment, not a representative leg — short, low-speed, and dominated by a cold-ish start. Included for completeness; excluded from any economy comparison.

• Missing TrackLogger file — 13 expected, 12 received. Day 1 files 1–2 and Day 8 Seg 1 lack partners; at most one of those gaps is fillable.
• Restart-lockout event count — rev 2 logged the lockout as "Durango, 2026-07-11"; the receptacle thermal stop is Cortez → Durango with resumption on the climb. Recorded as a single event. If they were two separate occurrences, the campaign has two confirmed thermal events rather than one, and the EVSE ceiling assessment needs revisiting.
• EVSE thermal ceiling — permanently uncharacterised from Campaign 1 data, because the operator correctly protected the engine instead of running the experiment. Say so; stop implying otherwise in any artifact.
• Section 04 net — not reportable on any leg until a LightShip draw-kW or Atlas SOC channel exists. Truck-side logging cannot observe the trailer.
• KB docs still to update against Pass 1 findings: Trip Plan (route and mileage both stale), V2V baseline doc (three converter states), TurboAssist energy balance (47% → 27.8%, negative net), report template (Section 03 wind table removal, Section 02 load-support line), HV System Reference (compressor-is-HV-bus correction).
• Instruction Set rev 5 required — the converter-states section still reads "9.26 kWh over 9.73 non-bulk-charging hours — 1.20 gal, 2.6%". Per the adjudication above that becomes 9.20 kWh / 9.73 hr / 1.19 gal / 2.6%, with the superseded figures recorded rather than deleted. The rev stamp should also log the ledger promotion.

Note that the first three pair naturally into a single session. The speed-hold protocol, the load-support SOC comparison, and a measured HVAC isolate can all run on the same flat, low-traffic stretch on the same afternoon — which is exactly the kind of consolidation that makes a deliberate experiment survive contact with a travel schedule.

Field-checklist additions earned by this campaign

• Verify both device clocks at every state-line crossing. Four of ten legs carried a one-hour error; a ten-second check at each border would have prevented all of them.
• Check TrackLogger liveness at every fuel stop. A logger that dies quietly looks identical to one that's working.
• Confirm the OBD recording restarted after every stop.
• Log every ProPower and TrekDrive state change with its reason, at the time it happens. Five words at the time would have prevented a misattribution that instead survived two knowledge-base revisions.

• V2V charging works and works consistently. 74.74 kWh delivered at 5.7–6.1 kW for 3.8% of campaign fuel, with no rate derate under any condition encountered. This is the pitch's strongest measured asset.
• The charging hardware has three states, not two, and the middle one is a product benefit rather than a standing cost — though proving that requires trailer-side instrumentation we don't yet have.
• TrekDrive's headline number was inflated by the interlock. 27.8% mechanical, not 47%, and net-negative on a truck-only charging basis. Lead the pitch with battery neutrality instead.
• Altitude, not ambient temperature, drives the worst thermal loading. The TurboAssist trigger design needs an altitude or ECT term; the current OAT-above-86°F correction would have missed the campaign's worst event entirely.
• Cruise control destroyed the aerodynamic measurement. The fix is deliberate speed variation, scheduled early, not more miles.
• Everything the truck can't see about the trailer is now the binding constraint on the analysis — Section 04 net, the load-support attribution, and the TurboAssist economics all bottleneck on one missing channel.
• Four published conclusions were withdrawn. Three of the four traced back to inferring cause from an OBD pattern that had more than one possible cause. The mitigation is a driver note, not a better algorithm.