Model health
Every model behind our figures has a version, release gates and validation — public here.
Energy-gap model (energygap-v1-2026-07-26)
Real vs WLTP consumption per segment (mild, incl. charging losses), calibrated on 20 ADAC Ecotest measurements. Fixed reference scenarios: mild ≈ 23 °C · winter standard −10 °C = real-world consumption + 5.5 kWh/100 km heating penalty (additive, efficiency-sensitive; falls back to ×1.35 without a consumption figure) — country-independent.
| Segment | Mild factor |
|---|---|
| Global | ×1.202 |
| midsize | ×1.22 |
| suv | ×1.213 |
| compact | ×1.173 |
| estate | ×1.218 |
| city | ×1.184 |
| luxury | ×1.204 |
Release gates (published targets)
| Target | published | model | OK | |
|---|---|---|---|---|
| GATE | TNO NL fleet monitoring 2024: real ≈ WLTP × 1.22 (n=45,000 BEVs) | ×1.22 | ×1.202 | ✓ |
| reference | Spritmonitor/Weiss et al. 2024: real ≈ WLTP × 1.07 (n=342 models; onboard-biased logging, gentler mix — different measurement plane than grid-side Ecotest) | ×1.07 | ×1.202 | — |
| GATE | NAF El Prix pair: mild winter −19% vs WLTP (2025) and extreme cold −38% (2026) ⇒ cold/mild uplift ≈ 1.31 (long-distance driving) | ×1.31 | ×1.35 | ✓ |
| reference | ADAC Dauertest: winter vs summer real-world +25–31% (midpoint 1.28; seasonal ~0–5 °C, milder than the −10 °C standard scenario) | ×1.28 | ×1.35 | — |
| reference | Green NCAP −7 °C vs +23 °C: +70% on short WLTC segments (cold-start dominated — short-trip worst case, not combined driving) | ×1.7 | ×1.35 | — |
Offline validation against the EV-Database panel (n=54, never in the product): implied −10 °C uplift ×1.368 (p10 ×1.333 · p90 ×1.4) vs ours ×1.35 — independent agreement.
Residual model (not yet released)
Trains weekly on our own market data (collector live since W28, Mondays 05:43 UTC; training 07:17 UTC). Release only when ALL gates are green: stability · backtest (needs ≥ 8 weeks) · anchors (±5 points vs published residuals) · sanity · interval coverage. Until then the site uses published anchors and the labelled segment approximation.
Latest training run (week 40, 2026)
2026-09-28 · 13 weeks of data · 7705 listings · 20 market buckets · model mixedlm.
| Gate | Verdict | Detail |
|---|---|---|
| stability | PASS | shareWithinTol 0.963 |
| backtest | PASS | mape 0.0982 |
| anchors | FAIL | worstAbsDeviation 0.2944 |
| sanity | PASS | |
| coverage | PASS | coverage 0.799 |
Release: NO — gates holding
Why the anchor gate is holding (week 40, 2026)
The anchor gate compares our model against published 3-year residuals (Bähr & Fess / Schwacke). It is currently failing — not because the model fits badly, but because our own market data sits well above the published forecasts. We would rather show the conflict than loosen a gate. Every figure is a share of today’s list price, and our estimate is evaluated at each bucket’s median listing age so the two sit like for like.
| Market bucket | Our estimate | Median asking | Gap | Anchor (3 yr) | Median age/km |
|---|---|---|---|---|---|
| Tesla Model Y RWD (2025–) | 54 % | 90 % | +36 | 56 % | 3.2 yr · 58k km · n=34 |
| Tesla Model 3 RWD (2025–) | 50 % | 81 % | +31 | 53 % | 3.3 yr · 76k km · n=37 |
| Fiat 500e | 39 % | 68 % | +29 | — | 3.5 yr · 24k km · n=37 |
| Audi Q4 e-tron performance | 44 % | 65 % | +21 | 45 % | 3.1 yr · 36k km · n=33 |
| Skoda Enyaq 85 (2026–) | 47 % | 68 % | +21 | 48 % | 3.1 yr · 38k km · n=32 |
| Renault Mégane E-Tech EV60 | 40 % | 60 % | +21 | — | 4.0 yr · 42k km · n=34 |
| VW ID.4 Pro (2024–2026) | 52 % | 70 % | +19 | 54 % | 3.2 yr · 30k km · n=33 |
| Cupra Born 58 Plus (2026–) | 52 % | 70 % | +18 | 44 % | 2.4 yr · 30k km · n=27 |
| BMW iX1 eDrive20 (2023–2026) | 60 % | 76 % | +16 | — | 3.1 yr · 43k km · n=33 |
| BMW i3 120 Ah (2018–2022) | 43 % | 58 % | +16 | — | 4.2 yr · 52k km · n=29 |
| VW ID.3 Neo 58 (2026–) | 57 % | 72 % | +15 | 49 % | 2.4 yr · 24k km · n=34 |
| Kia EV6 Air 84 RWD | 64 % | 79 % | +15 | — | 2.4 yr · 18k km · n=27 |
| Kia e-Niro 64 kWh (2020–2021) | 55 % | 69 % | +14 | — | 3.2 yr · 36k km · n=29 |
| BMW i4 eDrive40 | 54 % | 68 % | +14 | — | 3.1 yr · 51k km · n=36 |
| Renault Zoe ZE50 R135 (2019–2024) | 26 % | 39 % | +13 | — | 5.2 yr · 44k km · n=31 |
| Opel Corsa Electric 50 kWh (2025–) | 45 % | 56 % | +11 | — | 3.7 yr · 33k km · n=28 |
| Hyundai Kona Electric 65 | 35 % | 46 % | +11 | 48 % | 4.2 yr · 35k km · n=20 |
| Mercedes EQA 250+ (2024–2026) | 59 % | 67 % | +8 | — | 3.2 yr · 27k km · n=34 |
| Hyundai Ioniq 5 N Line 84 (2024–) | 56 % | 52 % | −4 | — | 3.1 yr · 38k km · n=30 |
What these figures do NOT say: median asking prices are what sellers ask, not what cars fetch — real transactions land below them — and they mix trims, while the published anchor forecasts a base trim. The anchor sits at 3 years / 60,000 km; each bucket’s median listing age is shown beside it and differs. And where a maker cut its list price (Tesla 2024–25), any share of TODAY’s list price reads higher than a share of what first owners paid. While this conflict is open the residual model stays unused and the site keeps computing with the published anchors — including where they sit below the market.