Endurance Data Tools: Plausibility, GPX Audit & Power

Each instrument here takes something you already have and reports what it implies. Two need nothing but figures visible on a public activity. One needs a file you hold yourself. None needs an account, an upload or a connection to any tracking platform whatsoever. The arithmetic happens in your own browser, on your own machine, every time.

Which instrument answers which question

Use the plausibility check when the question is about somebody else's result and all you have is the summary. Use the audit when the question is about your own recording and you hold the file. The two calculators serve narrower jobs where one figure is known. Between them they cover what can be settled without access to the athlete.

All four print their assumptions. That is the point of the design: a number produced by an unexamined model settles no argument, whereas a number whose inputs are visible can be challenged on those inputs, which is how a disagreement about data is supposed to proceed. Hidden coefficients turn a disagreement into an appeal to authority instead.

What none of them will tell you

No instrument here identifies cheating. Each reports what a set of numbers requires and where that sits against recorded human performance. A result outside the range has several possible explanations, and arithmetic cannot distinguish between them from the outside. The instrument reports the reading and stops there, deliberately, because that is where its competence ends.

That restraint is a deliberate choice, not a cautious one. Recording faults, mis-entered weights and mis-measured segments produce impossible figures far more often than deliberate manipulation does, and an instrument that announced a motive would be wrong most of the times it spoke. Restraint costs nothing here and buys the reading its credibility.

Why the assumptions are printed on the page

Every physical estimate rests on quantities nobody observed. This one assumes a rolling resistance coefficient of 0.004, a drag area of 0.32 square metres, air density of 1.225 kilograms per cubic metre and drivetrain efficiency of 97.6 percent. Each is a road-bike default, not a measurement, and changing one moves every result on the page.

Hiding those numbers would make every result unarguable, which sounds like confidence and is actually the opposite. A figure whose inputs are visible can be challenged on its inputs. A figure presented bare cannot be examined at all, and an unexaminable number settles nothing. Printing the coefficients is the cheapest honesty available to a calculator.

The same reasoning governs the duration curve. Sustainable relative power falls as an effort lengthens, so each reading is compared against a ceiling that falls with it, anchored near 7.5 watts per kilogram at five minutes and 6.4 at twenty. An hour sits a little under six, and the curve keeps falling gently beyond that.

Where each instrument is strongest and weakest

Gradient decides almost everything about estimate quality. On a climb steeper than about eight percent the gravitational term dominates so heavily that the aerodynamic guess barely matters, and the reconstruction lands within a few percent of a power meter. Gravity is the one term in the model that requires no assumption at all about the rider.

On flat ground the same arithmetic becomes a statement about the assumed drag area instead of about the athlete, because position, clothing and shelter move it by twenty percent or more. The file audit has no such dependency, since it checks internal consistency, not physics. A file either contradicts itself or it does not.

Estimate quality by terrain
Terrain	Dominant term	Plausible error
Climb above 8 percent	Gravity	A few percent
Climb of 4 to 8 percent	Gravity, some aerodynamics	Five to ten percent
Rolling terrain	Mixed	Ten to twenty percent
Flat road	Aerodynamics	Twenty percent or more

Queries

Questions about the instruments

Do any of these tools need an account or an upload?

None of them. Two work from figures a public activity already displays, and the file audit parses your activity inside the browser without transmitting it. There is no server doing the work, which is why no account exists and nothing can be stored. Closing the tab is all it takes to discard everything you entered.

How accurate is the implied power figure?

It depends almost entirely on gradient. On a steep sustained climb the estimate lands within a few percent of a power meter, because gravity accounts for nearly all the work. On flat ground it becomes a statement about the assumed drag area and should be read as a wide band.

Why does the ceiling change with duration?

Because sustainable power falls steeply as an effort lengthens, and the energy systems supplying it change. Comparing a five-minute result against an hour-long one without adjusting for that would make almost every short effort look superhuman and every long one look ordinary. Duration has to be fixed before two efforts can be compared at all.

Can these instruments prove somebody cheated?

No, and none of them claims to. They establish whether a set of figures sits inside or outside recorded human performance. A recording fault, a mis-entered weight and a mis-measured segment all produce impossible numbers, and none can be separated from the outside. Establishing that a figure is impossible is a narrower claim than naming a cause.

Which tool should I use if I have the file?

The GPX audit, first. It checks whether the recording itself is sound before any conclusion is drawn from it, and a file with elevation spikes or position jumps will feed wrong figures into any calculation you run afterwards. Check the recording first, then reason about what it says, never the other way round.

Background on what the numbers mean: the Strava data reference.