Methodology

Every figure this site calculates is published as a formula, not asserted as a result. If you disagree with a number here, you can check the working.

What “modelled” means

The most important thing on this page.

Where a manufacturer published a figure, that is what the spec sheet shows. A modelled figure never overwrites a real one. It appears alongside it, labelled as the model's output, which incidentally makes the comparison interesting in its own right. A car that beats its own published 0–100, or badly misses it, is telling you something.

Where an input the physics needs is missing, you get a gap rather than a guess. A car with no published drag coefficient anywhere gets no aerodynamic panel at all.

How accurate is it? Checked against published figures for cars across the range, the acceleration model lands within roughly 20% in both directions, and the top-speed model within about 5% for cars that are not electronically limited. It is useful for comparing cars on a consistent basis. It is not a substitute for a road test.

Units

Everything is stored in SI. Imperial figures are computed for display when you ask for them and are never stored or indexed, so the two cannot drift apart. The schema only accepts SI units, which means a horsepower figure cannot physically be saved into the database by mistake.

Power converts to metric horsepower (PS)by default rather than mechanical bhp, because the overwhelming majority of published figures for European and Japanese cars are PS. The two differ by about 1.4%, and quoting one as the other is a small error repeated endlessly elsewhere.

Formulae

All standard, textbook automotive physics.

Drag force: F_d = ½ · ρ · Cd · A · v²
Rolling resistance: F_r = Crr · m · g
Top speed: solve P_wheel = ½ρ·Cd·A·v³ + Crr·m·g·v for v
Braking distance: d = v² / (2 · μ · g)
Power-to-weight: P / m
Drag coefficient from force: Cd = 2F / (ρ · U² · A)
Frontal area (estimated): A ≈ 0.85 · width · height
Steady-state consumption: (F_d + F_r) · v ÷ (η_drivetrain · η_powertrain)
Depreciation: value(t) = price · (1 − rate)^t · mileageFactor · conditionFactor

Default constants

Every one of these is a reasoned default, not a measurement. Each carries the reasoning that justifies it, and each is a candidate for revision as the catalog grows.

Air density (ρ)1.225 kg/m3: ISA standard atmosphere at sea level, 15 °C. Real density varies with altitude and temperature by several percent; the model does not attempt to correct for either, so drag figures are quoted at standard conditions.
Standard gravity (g)9.80665 m/s2: Standard gravity, by definition.
Rolling resistance (Crr)0.012: Mid-range value for passenger-car tyres on asphalt (typical range 0.010–0.015). Not varied by era: the spread between individual tyre models is wider than the historical trend, so a per-era figure would imply precision the model does not have.
Frontal-area factor0.85: Standard approximation: a car fills roughly 85% of its width × height bounding box when viewed head-on. Published frontal areas are used in preference wherever they exist.
Load-transfer ratio (h/L)0.2: CG height over wheelbase, h/L ≈ 0.55 m / 2.7 m, typical of a passenger car. A per-car figure would need CG height, which is almost never published.
Tyre grip (μ), up to 19690.65: Cross-ply and early radial tyres on dry asphalt.
Tyre grip (μ), up to 19890.8: Period radial tyres on dry asphalt.
Tyre grip (μ), up to 20090.9: Modern radial road tyres on dry asphalt.
Tyre grip (μ), up to present1: Contemporary performance road tyres on dry asphalt.
Drivetrain efficiency, FWD0.9: Transverse front-wheel drive, shortest path from engine to wheels.
Drivetrain efficiency, RWD0.88: Longitudinal rear-wheel drive through a propshaft and final drive.
Drivetrain efficiency, AWD0.85: All-wheel drive: an extra differential and transfer path to drive.
Drivetrain efficiency, 4WD0.83: Part-time four-wheel drive with a transfer case, typically heavier-duty and less efficient.
Powertrain efficiency, petrol0.28: Petrol engine at steady cruise, near its efficiency island. Lower in mixed driving.
Powertrain efficiency, diesel0.34: Diesel engine at steady cruise.
Powertrain efficiency, hybrid0.32: Full hybrid at steady cruise, where the electric path contributes little.
Powertrain efficiency, phev0.32: Plug-in hybrid running its combustion engine at steady cruise.
Powertrain efficiency, bev0.85: Battery to wheels: inverter, motor and reduction gear losses only.
Powertrain efficiency, fcev0.5: Fuel cell to wheels, including stack and drive losses.
Fuel energy, petrol32 MJ/L: Lower heating value of gasoline, approximately 32 MJ/L.
Fuel energy, diesel35.8 MJ/L: Lower heating value of diesel, approximately 35.8 MJ/L.
Power availability in acceleration, ice0.7: Combustion engine through a multi-speed gearbox: the engine sweeps its rev range in each gear rather than sitting at peak power, and each shift costs roughly 0.3–0.5 s. Calibrated so the model lands close to published figures for cars across the range rather than flattering them.
Power availability in acceleration, electric0.92: Electric drive with a single reduction gear: near-peak torque from zero and no shifts, so far more of the rated output is genuinely available. Not 1.0, because motor power still tapers above base speed.

Known limitations of the dynamics model

Stated plainly, because a model whose weaknesses are hidden is worse than no model.

It does not know about gearing. Top speed is where power balances drag. A car geared out below that point will never reach it.
Electronic limiters are invisible unless recorded. Many EVs and most German cars are limited well below the speed the physics allows. Where a limiter is known it is used; where it is not, the figure shown is what the car could do unrestricted, and it will exceed the published number, sometimes by a lot.
It has no torque curve. Acceleration assumes a fixed fraction of peak power is available throughout the run rather than modelling an engine sweeping its rev range and losing time to shifts. That fraction is the model's largest single approximation and it is listed in the constants above.
Consumption is a steady cruise, not a drive cycle. It models holding a constant speed on level ground, so it is not comparable with a WLTP or EPA figure and must not be read as one.
Braking is the physical floor. It assumes every tyre is at peak grip for the whole stop, and ignores fade, weight transfer and whether the brakes could actually reach the tyres' limit, which for many older cars they could not.
Air density is standard. No correction for altitude or temperature, both of which move real drag by several percent.

The Markey Score

The Markey Score blends four sub-factors on equal weights. Those weights are provisional: they will be tuned once there is enough of a catalog to tune against, and equal weighting is what we use in the meantime rather than inventing a split we cannot justify. A score is a comparison between cars on these factors. It is not a percentage, and it is not a verdict.

efficiency: 25%
performance: 25%
practicality: 25%
value Retention: 25%

A car is never scored on data it does not have. Missing sub-factors are excluded and the remaining weights renormalised, and where fewer than half the factors have data behind them the blended number is withheld entirely rather than shown with false confidence. Scoring a missing figure as zero would rank the least-documented cars last, which measures our research rather than the car.

Depreciation

A projection from a published formula. Not market data, and not a valuation.

economy15% / yr: Low purchase price, strong used demand, cheap to run and repair.
mainstream17% / yr: The volume market, and the reference point for the other tiers.
premium20% / yr: Higher option content that does not survive resale, and dearer out-of-warranty maintenance.
luxury24% / yr: A thin used market and running costs that deter second owners.
ultra luxury18% / yr: Very low volume, where collectibility begins to counteract depreciation. The least predictable tier by a wide margin: individual cars deviate enormously and this figure should be read as barely more than a placeholder.

Decay is exponential (a percentage of remaining value each year) because that is how cars actually behave. A linear model would reach zero on a specific date, which no car does. If you log real valuations against a car in your garage, they are plotted on the same chart as points, and the curve is deliberately not refitted to them: the divergence is the interesting part.

Insurance risk

A relative comparison between cars, not a quote. What you would actually pay depends far more on you (your age, licence history, address and claims record) than on the car. Nothing here is an insurance estimate.

power To Weight: 40%
acceleration: 25%
value: 25%
body Style: 10%

The wind tunnel, and what its numbers are worth

A real lattice-Boltzmann solver, and an honest account of the one place the obvious validation turned out to be meaningless.

The solver is a D3Q19 lattice-Boltzmann method with two-relaxation-time collision, halfway bounce-back walls, and drag measured by momentum exchange across the boundary links. It exists twice: a CPU reference that runs in the test suite, and a GPU version for interactive use. Both read their velocity set from the same file, so they cannot disagree about the physics.

The obvious validation is a category error, and we nearly shipped it. The canonical shapes have published drag coefficients: a sphere at 0.47, a cube at 1.05, a flat plate at 1.17, a streamlined teardrop at 0.04–0.05. Those figures describe flow at Reynolds numbers around 10⁴–10⁶. The CPU reference runs on grids small enough to finish inside a continuous-integration job, which puts it atRe ≈ 50. At Re 50 a sphere's real drag coefficient is about 1.5, not 0.47. The published number is not wrong, it simply describes a different flow regime.

Comparing the two would have failed in both directions: it would reject a correct solver, and it could be made to "pass" by adjusting the grid until the numbers happened to meet. So the sphere is measured against the Clift–Gauvin correlation evaluated at the Reynolds number actually run, which is a function of Re rather than a single figure.

Sphere, CPU reference: Cd 2.25 measured, against 1.52 from Clift–Gauvin at Re 51, a ratio of 1.48.
Is that bias a bug?: It does not appear to be. Staircase bounce-back walls over-predict drag at low resolution: the modelled wall is rougher than the shape it stands for. Measured across two grid sizes the ratio held at 1.37 and 1.41, and it is resolution-insensitive. A bug would move with the grid; a discretisation bias does not.
What the gate actually asserts: That the solver stays finite; that drag opposes the flow; that force scales with the square of velocity, as the definition of Cd requires; that a streamlined body has less drag than a bluff one at matched conditions; and that the sphere sits inside a stated band of the Re-appropriate reference. These run on every build.

Not yet claimed: absolute agreement with the four published high-Reynolds figures. That needs the GPU solver at full grid resolution, and the numbers will be published here, whatever they turn out to be, when it can be measured. Until then no drag coefficient from this solver appears anywhere on the site as though it were a property of a real car. A car's published Cd remains the authoritative figure on its spec sheet.