Measuring Power Without a Dyno: GPS, Timeslips and Datalogs

People ask for a dyno number when what they actually want to know is whether the change they just made helped. Those are different questions, and the second one is answerable for free.

As of September 2026 the practical position is: a chassis dyno buys you controlled load and a defensible absolute figure, and for everything else — did that revision help, is the car down on power today, did the intake do anything — a datalog comparison and a couple of clean runs answer it more cheaply and can be repeated any evening.

This article is about street-legal measurement on roads or at a track, in that order of caution. Nothing here is worth a ticket or a crash.

Three methods, three different answers

Method	What it really measures	Repeatability	Best used for
GPS / accelerometer estimate	Acceleration, converted to power via assumptions	Good same-session, poor across days	Before/after on the same road
Drag-strip timeslip	Trap speed ≈ power-to-weight; ET includes traction	Good with the same driver	Verifying a substantial change
Datalog comparison	Time through an rpm band at fixed load	Excellent	Every revision, week to week
Chassis dyno	Torque at the roller, corrected	Excellent within one shop	Baselines and tuning under load

The bottom two answer "did it improve" precisely and say nothing absolute. The top two produce a number that feels authoritative and rests on assumptions. Knowing which you are holding is most of the skill.

The physics, and why the inputs decide everything

An accelerometer or GPS app knows one thing: how fast your speed is changing. Power at the wheels follows from mass times acceleration times velocity, plus whatever is being spent on rolling resistance and pushing air. Then drivetrain loss is estimated backwards to reach a crank figure.

Every term after "how fast your speed is changing" is supplied by you or assumed by the app:

Vehicle weight, which must include fuel, driver, tools and anything else in the car. This one is proportional — a weight 8% wrong makes the power 8% wrong, and most people enter a curb weight from a specification sheet and then drive with a full tank and themselves aboard.

Drivetrain loss, usually a flat 15% or so. Reality varies by layout and gear.

Aerodynamic drag and rolling resistance, almost always defaults. At higher speeds drag dominates the correction, so an error there is worst exactly where the interesting part of the pull happens.

Road gradient, which the app cannot see. A 1% incline — invisible from the driver's seat — moves the result by more than most real tuning gains.

That list is why the same car measured on two roads can produce two confident, different numbers. It is also why the comparison use survives: run the same car on the same road on the same evening and every one of those errors appears identically in both runs, so the difference between them is real even though neither number is.

Doing a GPS comparison properly

Same stretch of road, both directions, averaged. This is the single most important rule and it exists to cancel gradient and wind. One-direction results are not comparable to anything.

Same gear, same starting rpm, same throttle. Comparing a third-gear pull to a fourth-gear pull tells you about gearing.

Three runs minimum per configuration, discard the outlier. A single run is noise.

Back to back, same session. Flashing the file and running again within the hour removes fuel level, tyre pressure, air temperature and your own technique from the comparison.

Record the conditions anyway. Temperature, pressure and humidity, so that a later comparison across days has something to correct with.

Report the delta, never the number. "Two tenths quicker through the same band on the same road" is a defensible claim. "Now makes 412 horsepower" is not, and it will be repeated back to you as fact.

Trap speed: the cheapest honest number

At a drag strip, elapsed time and trap speed measure different things. Elapsed time is heavily contaminated by launch, traction, tyres and driver. Trap speed depends mainly on power-to-weight, which is why it is the number people use for comparison.

A genuine power increase shows as trap speed. Same driver, similar weather, similar fuel: a couple of miles per hour is real, a tenth of a mile per hour is not.

The horsepower conversions circulating for trap speed are rough approximations built on assumed drag and weight — useful for a sanity check, not for a claim. Their value is the same as the GPS method's: the change is meaningful even when the absolute is not.

The method almost nobody uses, which is the best one

You already have the instrument. A datalog from the tuning software measures the engine directly and depends on none of the weight, drag or gradient assumptions.

Time through a fixed rpm band, in a fixed gear, at wide-open throttle. Log a pull from 3,000 to 6,000 rpm in third. Note the elapsed time. Change the file, repeat on the same road, note it again. More torque under the curve means less time. That is a direct measurement of the thing you changed, and no app has to guess your weight.

Supporting channels make it interpretable rather than just comparable:

Airflow at the same rpm. Higher measured mass air flow, or higher manifold pressure on a boosted car, at the same engine speed means the engine is genuinely moving more air — a physical explanation for a gain rather than an inference from a stopwatch.

Knock retard, across both pulls. The critical one. A car can be faster and have consumed its safety margin; the timing comparison is what tells the two apart.

Intake air temperature. Establishes whether the two runs are comparable at all. Twenty degrees of difference explains more than most calibration changes.

Commanded versus actual air-fuel ratio. A gain that came with a leaner mixture is not a gain, it is a loan.

Which channels to log and how to set them up is in HP Tuners datalog channels explained, and the shape of a clean baseline run is in reading your first wide-open-throttle pull. The same discipline underpins the VE table correction workflow, where the log is not a proxy for the measurement — it is the measurement.

Correcting for weather, or not

Within one session, don't bother. Conditions are effectively identical; correcting introduces error rather than removing it.

Across days or seasons, you must. A naturally aspirated engine loses roughly three percent of its power per thousand feet of density altitude, and hot air costs real power at sea level too. A car that is "down 15 horsepower" in August compared with a February baseline is usually just breathing worse air — the mechanism is in heat soak and density altitude, and whether that justifies a separate calibration is in seasonal and altitude retunes.

Uncorrected cross-season comparisons are where most phantom power losses come from.

What the dyno still buys

Three things the roadside methods cannot provide.

Controlled, sustained load. A dyno holds an engine at a chosen speed and load for as long as needed, which is how tables get filled properly and how a problem gets provoked deliberately instead of discovered on a motorway.

Repeatability within one shop. Same machine, same operator, same corrections — differences between runs are the car.

A number others accept. For a sale, a class, or a build record, the sheet is the currency. What that sheet actually says, and how correction and smoothing change its appearance, is in how to read a dyno sheet; what the session itself involves and costs is in dyno session cost.

Dyno figures are not comparable between shops. Different machines and correction standards produce different numbers for the same car, which is why the same engine can be "410" at one shop and "380" at another with neither lying.

What you need to do any of this

Everything above except the timeslip needs the ability to record a datalog, which means the interface and licensing for your vehicle. If you are choosing hardware, the interface comparison is in MPVI3 vs MPVI2 and the platform-coverage question — which tool even supports your car — is in HP Tuners vs COBB vs EFILive. A GPS app costs nothing and measures acceleration regardless of platform, but it cannot see inside the engine, which is the limitation this whole article keeps running into.

The thing none of these measure

Not one method here tells you whether the car is safe.

A faster car with no knock margin left measures as an improvement on every one of them. A leaner mixture at wide-open throttle makes more power and shortens the engine's life. A timeslip cannot see intake air temperature and a GPS app cannot see fuel trims.

Power measurement and safety verification are separate jobs, and the second one matters more. The channels that answer it are the same ones in every log you take — knock retard, air-fuel ratio, intake air temperature, injector duty cycle. Read those first, and treat the speed comparison as the part that tells you whether the work was worth doing.

The summary

If the question is "did that change help", log the same gear on the same road before and after and compare the time through a fixed rpm band. It is free, repeatable and depends on no assumptions about your car's weight.

If the question is "how much power does it make", a GPS estimate gives a defensible delta and an indefensible number, a trap speed gives a good comparison and a rough conversion, and a dyno gives a figure that is only comparable to other runs on that same dyno.

And whichever you use, read the safety channels in the same log. Faster is not the same as fine.

Frequently asked questions

How accurate are GPS horsepower apps?

As an absolute figure, not accurate enough to quote — the result depends on vehicle weight, drivetrain loss, rolling resistance and aerodynamic assumptions you are supplying or accepting as defaults, and a wrong weight alone moves the answer proportionally. As a before-and-after comparison on the same road, same day, same car, they are genuinely useful because every one of those errors is present in both runs and cancels.

Can I tell if a tune worked without a dyno?

Yes, and more reliably than most people expect — log the same gear, same throttle and same road before and after, and compare how long the engine takes to pull through a fixed rpm band. That number improves when torque improves and is not affected by the weight and drag assumptions a virtual dyno depends on. As of September 2026 it remains the cheapest honest comparison available.

Does trap speed tell you horsepower?

Trap speed is the best cheap proxy there is, because it depends mainly on power-to-weight while elapsed time is heavily contaminated by traction and launch. Two runs a week apart with the same driver and similar air will show a real power change as a trap speed change. The absolute horsepower conversions floating around are rough approximations, not measurements.

Why do my runs give different numbers every time?

Almost always road gradient, wind or air temperature rather than the car. A slight incline you cannot feel changes the result more than a real tuning gain does, which is why runs in opposite directions on the same stretch — averaged — are the standard method. Air temperature and density altitude explain most of the rest.

Do I need to correct for weather?

For comparisons within the same session, no — conditions are effectively identical. Across days or seasons, yes, because a naturally aspirated engine loses roughly three percent per thousand feet of density altitude and hot air costs measurable power. Without correction you will credit or blame a calibration change for weather.

Is a chassis dyno still worth paying for?

For tuning under controlled, repeatable load and for a defensible number, yes — that is what you are buying, not the graph. For answering whether last night's change helped, the methods here are free and you can repeat them any evening. The two are complements: the dyno sets a baseline, the cheap methods track everything afterwards.

What can none of these methods do?

None of them verifies safety. A car can be faster and simultaneously running with no knock margin left, and every method here would call that an improvement. Power measurement and safety verification are separate jobs — the second one is the datalog channels for knock retard, air-fuel ratio and intake temperature, and it is the one that matters more.