The reserve, provable: energy-feasibility certificates for eVTOL missions
Battery-electric air taxis fly under a hard reserve rule — the FAA's powered-lift SFAR (14 CFR Part 194, 89 FR 92296 (2024-11-21, eff. 2025-01-21)) allows a VFR reserve of 20 minutes at normal cruise only where the flight manual shows a vertical landing is possible along the whole route, and 30 minutes by day otherwise — and industry is arguing the number because reserves gate route economics. Every feasibility figure in that argument today is simulation. This page holds the other kind of statement: for a mission whose durations, powers, capacity and efficiency are known only as BOXES, the verdict "every parameter point lands at or above the reserve floor" is decided — CERTIFIED, REFUTED with an exact falsifying witness, or honestly REFUSED — by interval arithmetic with outward rounding. Mathematically certified enclosures; no airworthiness meaning, and the page says so.
The same mission, decided at every cruise duration under each rule. Both strip boundaries are located to the second by bisection, with a certified verdict on each side; a sweep every 15 seconds checks that the verdict only ever moves from CERTIFIED to REFUSED to REFUTED as the cruise grows, and the build refuses the page if it does not.
The FAA's 2024 powered-lift SFAR sets energy reserves for eVTOL operations — under VFR, 20 minutes at normal cruise where the flight manual shows a vertical landing is possible along the whole route, otherwise 30 minutes by day and 45 by night (the rule pack apps/skyaudit/scenario/rules/faa-sfar-vfr.json, pinned to the Federal Register text) — and the industry pushback is public and quantitative: reserves determine which routes close. EASA sets a 5-minute final reserve inside a wider energy plan that does not reduce to one number (apps/skyaudit/scenario/rules/easa-final-reserve.json). Yet every number in that argument — mission feasibility, deconfliction energy overhead, reserve adequacy — is produced by simulation: Monte Carlo percentiles over sampled parameters, floating-point optimizers, no statement that survives an adversarial reading of "for every case in the stated envelope".
The gap is not modeling sophistication — the field's power models are fine. The gap is the QUANTIFIER. "Feasible in 10,000 samples" and "feasible for every point in the box" are different claims, and only the second is the shape a reserve rule actually asserts. Interval arithmetic decides the second shape directly. We have not found such a certificate in the eVTOL energy literature; no search record is pinned here, so that is a statement about our reading, not a survey.
A mission is segments (hover-climb, cruise, hover-land, …) with duration and electrical-power BOXES; the battery contributes usable-energy and efficiency boxes; the reserve rule contributes a reserve time at a reserve power. Used and required energy are enclosed over the whole box with outward rounding, and the universal verdict follows from two one-sided comparisons: worst case clears — CERTIFIED for every point; best case fails — REFUTED for every point, with the most favorable corner re-proved negative in exact BigInt rationals as the falsifying witness; anything between — REFUSED, with both margins printed. A refusal means the boxes genuinely contain passing and failing aircraft, and no honest instrument can say more until the boxes tighten.
The corner-witness logic is checked, not assumed: the battery sweeps all 256 corners of three two-segment instances (feasible, marginal, doomed) in exact rationals and requires the interval bounds to be sound AND corner-achieving (tight to rounding) on each. Scope, honestly: this is energy accounting — Σ duration·power/efficiency against usable capacity — not electrochemistry. Voltage sag, temperature and aging live INSIDE the boxes a user states, which is exactly why the inputs are boxes: widen them to cover what the model does not resolve, and every verdict above stays sound.
Every verdict is exact about the boxes it was given. What it says about an aircraft rests on premises the arithmetic cannot reach. In the order a verdict travels through them, each with the rung its checking reached — validated (against the thing itself, held out), compared (against a model that keeps more), verified (the code against an exact solution of its own equations) or assumed (adopted; nothing here checked it):
The arithmetic is verified; the weakest premise is assumed. A verdict here is exact given the boxes and the rule: about a real aircraft it says what those boxes say, and no more.
| mission | 20-min reserve | margin (kWh) | 30-min reserve | margin (kWh) |
|---|---|---|---|---|
| short hop (10 min cruise) | CERTIFIED | 27.5 | CERTIFIED | 6.7 |
| design mission (20 min cruise) | REFUSED | -0.9 … 61.2 | REFUSED | -21.8 … 45.2 |
| stretch (35 min cruise) | REFUSED | -43.5 … 31.9 | REFUSED | -64.4 … 16.0 |
| deep stretch (55 min cruise) | REFUTED | -7.1 | REFUTED | -23.0 |
Margins are worst-case for CERTIFIED rows, best-case for REFUTED rows, and the worst…best pair for REFUSED rows — the two numbers a refusal honestly has — each printed rounded outward, a worst case down and a best case up. The parameter boxes are a representative vectored-thrust class (hover 420–520 kW, cruise 110–150 kW, usable 130–145 kWh, efficiency 0.88–0.94, reserve power 90–110 kW), stated in the page source as inputs — they are NOT any manufacturer's aircraft, and tightening or widening them re-decides every row by the same arithmetic.
Bisecting on cruise duration with a certified verdict at every probe puts the end of the mathematically certified envelope at 18 min 41 s of cruise under the 20-minute reserve and 11 min 21 s under the 30-minute one, for the stated boxes; each is the low end of a two-minute cruise box, so the cruise it certifies runs up to two minutes longer. That difference — 7 min 20 s of provable endurance — is the quantity the reserve debate is actually about, computed here as a theorem about the boxes rather than a fleet average. The same bisection run against an operator's own measured boxes is the instrument's intended use.
This was the first instrument of this site's aerospace front: the cheapest object where a certified universal statement replaces a simulation percentile in a live regulatory argument. The same verdict was then run on a real day: SkyAudit re-flies a pinned day of New York helicopter traffic with four eVTOLs' own published boxes under the same SFAR rule pack, and the glide band draws an engine-out reach as a proved inner ring, a refuted outer one and the undecided annulus between.