The honest kilowatt-hour.

The set draws 5 amps a leg. The generator won’t go below 42.

A 70 kVA-class spot generator is committed through the night, measured the way sets measure everything — in film amps, each 120 V leg counted on its own.

• THE RIGA 70 kVA-class towable diesel on 120/208 V three-phase — healthy anywhere up to ~155 A a leg — with an integrated load bank.
• THE RULERun it at 70%, 50%, even 32% of rating and both maths agree. But diesels held under roughly 30% risk wet-stacking — the engine must stay loaded.
• THE TRICKSo below a ~42 A floor the load bank burns real fuel to fake the missing demand. The energy does nothing.

ILLUSTRATIVE SCENARIO — GENERIC 70 kVA-CLASS TOWABLE DIESEL WITH INTEGRATED LOAD BANK; FILM-AMP CONVENTION (PER-LEG 120 V AMPERES, NOT TRUE THREE-PHASE LOADS); TYPICAL PUBLISHED FUEL CURVE (≈2.2 L/h IDLE + 0.26 L/kWh) AND STANDARD DIESEL FACTOR 2.68 kg CO₂/L. THE STRUCTURAL EFFECT IS THE PAPER'S DEMONSTRATION D2b.

The battery remembers. The books don't.

Two puzzles live inside every battery on a hybrid set. Both are known to the field — and both are worth feeling firsthand.

Kilowatt-hours are not interchangeable — the pack remembers where its energy came from:

PRESS & HOLD TO POUR — CHARGE FROM A SOURCE, THEN DELIVER TO THE SET

Your carbon report covers a window — a week, a show, a quarter. Batteries don't respect it:

• ARRIVE FULL, LEAVE EMPTYThis window ran on energy made with last window's diesel. The report reads too clean.
• ARRIVE EMPTY, LEAVE FULLThis window paid carbon forward for the next one. The report reads too dirty.

Now the pack arrives on a truck. You never saw the charger. Set the level it arrived at, the level it left at — and the books answer:

Same set, different battery schedules — the report can miss by −65.6% to +61.0%. The curve is exact: e = s / (1 − s).

BOTH PUZZLES HAVE ESTABLISHED CONVENTIONS IN THE FIELD — THE PAPER ADOPTS THEM. ITS CONTRIBUTION IS THE DENOMINATOR: THE USEFUL KILOWATT-HOUR.

SOURCE: PAPER §7 (D1) — EXACT RELATION e = s/(1−s), VERIFIED ON EVERY SIMULATED TRACE; MISS RANGE FROM THE PAPER'S FULL BIAS STUDY. PACK-CONSOLE INTENSITIES ILLUSTRATIVE (GENSET AT ITS FLOOR ≈1.09, FROM PANEL 01; GRID AND SOLAR TYPICAL PUBLISHED LIFECYCLE VALUES — SOLAR IS NEVER QUITE ZERO; PANELS ARE BUILT SOMEWHERE); AVERAGE-OF-STOCK CONVENTION; WINDOW DEMO ASSUMES A 100 kWh PACK, 250 kWh USED, AND THAT STOCK LEAVES AT THE INTENSITY IT ARRIVED WITH.

Measure honestly, and the fix appears.

This one is real: a 21-day production power deployment in Hawai‘i, 2026.

• THE LEGACYTwo committed 70 kVA-class diesels, every production hour.
• THE SWAPTwo battery packs and one genset — recharging at 3 AM, before crew call.
• THE TESTSame set, same schedule, directly measured.

The data had said this site didn't need a microgrid. Measured honestly, it did — emphatically.

FIELD DATA — 21-DAY PRODUCTION POWER DEPLOYMENT, HAWAI‘I, 2026 (CLIENT CONFIDENTIAL; PRODUCTION UNNAMED). ENERGY AND RUNTIME DIRECTLY MEASURED VIA THE STORAGE PLATFORM'S MONITORING; FUEL FROM THE MANUFACTURER FUEL CURVE INTERPOLATED TO OBSERVED LOAD × ALARM-VERIFIED RUNTIME; CO₂ PER EPA AP-42 §3.4 (22.4 LBS/GAL DIESEL); LEGACY BASELINE MATCHED TO THE ACTUAL SHOOTING SCHEDULE. THE PAPER'S PAIRED SIMULATION (§7, D0) REPRODUCES THE SAME STRUCTURAL EFFECT ON GENERIC HARDWARE.