Your phone is the dyno.

The FXR Archive Phone Dyno is a web app for your phone. It estimates rear-wheel horsepower and torque, corrected to SAE J1349 standard conditions (SAE J1349), from GPS speed and a physics model of the loads on the bike.

A roller dyno is still the best way to measure power, and this is not meant to replace one. The idea is a repeatable, low-effort way to see a power curve on the road, compare before and after a change, and learn something about your bike between dyno sessions.

It is not an app-store app. You open it in Safari on an iPhone, tap Share, then tap Add to Home Screen. After that it launches from its own icon, runs full screen, keeps the screen awake, and updates itself when you reopen it. Once it is installed it works without a signal. Only the weather lookup needs a connection, and the app tells you when it has to do without one.

Each bike preset carries the gearing (primary, transmission and final drive), the rear tire size, an estimated drag area (CdA), a rolling-resistance base (Crr) and an estimated share of weight on the front wheel. You can edit total weight and tire radius in Settings; drag area and rolling resistance come from the preset. The beta ships with a small set of presets, including two Evo FXRs and a few modern touring and sport-touring bikes. There is no custom-bike option yet.

GPS only

The app asks iOS for high-accuracy location with no cached fixes, reads the Doppler speed the GPS chip reports, and falls back to distance between fixes divided by time when a fix has no speed. Every fix is placed at its own timestamp, not at the moment it happened to arrive in the app. That matters, because acceleration comes from speed changes over time, and arrival times can be uneven.

Earlier test versions also read the phone's motion sensors and had a coastdown mode for measuring drag. Both were removed. There is no accelerometer and no calibration step: you mount the phone and ride. Drag and rolling numbers always come from the bike preset, so every pull on a given bike is computed on the same basis.

A drum-dyno basis

The number most riders know is the one on a sheet from a drum (inertia) dyno such as a Dynojet, so the app is built to produce a number on that same footing.

On a drum dyno the rear tire's rolling loss happens before the power reaches the drum, so the drum never sees it, and the front wheel does not turn at all. On the road both tires roll. Counting both would report more power than a drum dyno shows for the same bike, so the app counts only the front tire's rolling drag. In the words of the source code: to match a drum dyno, the road-to-wheel force counts ONLY the front tire's rolling resistance.

Semi-automatic

You tap Record pull once. Recording never starts from motion alone. After that tap the app does the rest:

1. It watches for the roll-on and arms when it sees one.
2. It ends the pull by itself when you roll off.
3. It saves the pull with every raw GPS fix.
4. When you open the pull, it trims the data to the roll-on to roll-off window, looks up the weather and fits the curve.

A 30 second limit and a Stop button are always there as backstops. The point is safety and consistency: you should not be looking at or tapping a phone during a full-throttle run, and a detector that uses the same rules every time gives more repeatable pulls than a thumb on a screen.

Power at the rear wheel is the force needed to push the bike down the road times the road speed. The app adds that force up from four parts:

F = m × a  +  aero drag  +  front-tire rolling drag  +  grade
power = F × speed

• Mass times acceleration. Mass is the bike plus rider and cargo (default 200 lb, adjustable). Acceleration comes from GPS speed changes.
• Aerodynamic drag. ½ × air density × CdA × speed². Saved pulls use the actual air density from the weather at the time of the pull.
• Front tire rolling drag. Crr × front weight share × total weight, with Crr rising with speed (below). The front share is a per-bike preset value, 0.40 unless the preset says otherwise.
• Grade. Road slope, entered by hand or estimated from GPS altitude, on the live screen only. The saved-pull analysis assumes level ground, which is one reason a flat road matters.

Crr(mph) = base Crr × (0.01632 + 0.0000018625 × mph²) / 0.020

That speed shape was fitted to tire research values of 0.018 at 30 mph, 0.029 at 82.5 mph and 0.035 at 101.5 mph. The research was on bias-ply tires, so for radial-tired bikes the shape is an estimate. There is no allowance for rotating inertia (wheels, driveline) in the model.

RWHP = force (lbf) × mph / 375        (same as N × m/s ÷ 745.7)

Engine rpm from speed, tire and gearing

GPS knows road speed, not engine speed, so the app works rpm out from the tire and the gearing:

wheel rpm  = speed ÷ (2π × effective tire radius) × 60
overall    = primary × gear × final drive
engine rpm = wheel rpm × overall

The effective radius is 0.97 × the unloaded radius from the tire size, to allow for squat under load. You pick the gear for each pull in the analysis (4th by default on five-speeds). On a dual-clutch bike you have to hold it in manual mode, because an automatic upshift mid-pull breaks the rpm math.

Torque

Torque follows the usual dyno-sheet convention, torque (ft-lb) = RWHP × 5252 ÷ rpm. When the curve is drawn, horsepower is re-derived from the fitted torque curve, so the two curves always agree.

SAE J1349 correction and the weather

Engines make less power in hot, humid or thin air. To compare pulls from different days, every pull is corrected to the J1349 reference day of 77 °F, 29.23 inHg of dry-air pressure and 0% humidity:

CF = 1.18 × (29.23 / Pd) × √((T°F + 459.4) / 536.4) − 0.18
Pd = station pressure − water vapor pressure (inHg)
corrected HP = raw HP × CF;  corrected torque = raw torque × CF

When you open a pull, the app asks Open-Meteo for temperature, humidity and station pressure at the pull's start location and hour. If station pressure is missing, it derives it from sea-level pressure and an elevation chosen automatically. Offline, or if the lookup fails, it uses the J1349 standard day (CF 1.00) and says so. All the weather fields stay editable, and the analysis also shows density altitude.

The live screen gets its own correction factor from current conditions, refreshed every 30 minutes. Without one, the big tile says Unadjusted Raw HP, and the MAX HP tile resets whenever the basis changes, so corrected and raw numbers are never mixed.

After you tap Record, every raw fix goes into a small state machine. Acceleration is the speed change since the previous fix divided by the time between their own timestamps. The pull arms on the roll-on, keeps track of the highest acceleration so far, and watches for a low stretch. The pull ends only when it has been armed for at least 3 s, has reached a real peak and speed gain, and then either passes all three end tests or stays low for 3.8 s.

Gear shifts and partial lifts

A quick shift makes acceleration dip for about a second, but speed is rising again by the next fix, so the low stretch resets before it can pass the not rising and clearly decelerating tests. As the code comment puts it: A gear-shift dip (~1 s) recovers in the next interval, so it doesn't meet all three; a roll-off does. A shift still breaks the rpm calculation, because the analysis assumes one gear, so shifting mid-pull isn't recommended; the detector just won't cut the recording short because of it.

If you only roll part of the way off, the bike may not slow hard enough for the −0.08 g test. The 3.8 s rule covers that case. If the pull never arms, it records to the 30 s limit, and Stop always works. Each saved pull records why it stopped: roll-off, the 30 s limit, or the rider.

A recording includes some cruising before the roll-on and some slowing after the roll-off. Those parts would bend the ends of the curve, so the analysis trims them automatically:

1. It smooths the speed trace with a local fit over about ±2.5 s. This smoothed trace is used only to find the window, never to build the curve.
2. It finds the peak acceleration, which must be at least 0.3 m/s² or there is no clear pull.
3. Start: it walks back from the peak while acceleration stays above 60% of the peak. That marks the roll-on.
4. End: it walks forward and stops at the first sustained drop: below 30% of the peak for 2 samples in a row, negative acceleration, or speed no longer rising.

The window needs at least 5 samples. If there is no clear run, the app says so (need about 4+ seconds of steady throttle) and offers two sliders to pick the window by hand. A manual trim is saved with the pull, and Reset to auto goes back to the automatic one.

Building the curve

Inside the window, each pair of consecutive fixes gives one point: the speed change over the time between them, placed at the average of the two speeds. At one fix per second that is 1 s of averaging, with no separate speed smoother. A point is kept only if its rpm is higher than every point before it, so the curve can never fold back, and at least 4 points are needed.

Corrected torque is then fitted against rpm with a local weighted fit (LOESS): a window of about ±200 rpm, widened where needed to take in at least 4 points, nearby points weighted most, a lean toward a straight line at the one-sided ends so they don't hook on one noisy point, values held within the points' range plus 10%, and nothing drawn past the first or last point. The individual points are drawn as faint dots behind the curves, so you can see the scatter. Horsepower and torque always share one y-scale.

Phone GPS gives about one speed reading a second, so a single pull has only a few points, often 150 to 300 rpm apart. Stacking combines several pulls of the same bike in the same gear into one result, so the points from one pull fill the gaps between the points of another.

1. On Saved pulls, tap Stack pulls, tick 2 or more pulls of the same bike in the same gear, then tap Stack. Pulls of another bike or gear are grayed out. A pull too short to graph on its own is marked as one that can still be stacked.
2. Each pull is worked out exactly like a single pull: its raw GPS fixes, its own weather and SAE J1349 correction factor, its own trim.
3. The points of all the pulls are lined up by engine rpm, not time, and pooled. One smooth corrected-horsepower curve is fitted through them (a local weighted fit), and torque = HP × 5252 ÷ rpm.
4. Every point is checked against the curve of the other pulls. Points far off it are dropped, and a pull whose points are mostly off is dropped whole, with the reason shown. Telling which pull is off needs 3 or more pulls.
5. The curve covers only the rpm range where at least 2 pulls overlap. The shaded band shows ±1 standard deviation between pulls at each rpm.

The result also says whether the pulls ran in both directions. Pulls run both ways on the same road cancel most of the wind and grade; if every pull ran the same way, a steady wind or a slope is baked into the result. A stack can be named and saved on the phone (it keeps its own copy of the pulls) and shared as a PDF or Excel file.

Automatic start and stop is not new in itself, and it's fair to say so. Other phone apps have it too: True Power starts a timed run automatically when its trigger detects movement (True Power (Google Play)), and Clockwatts has an automatic measurement mode that can end a run without input (Clockwatts). What the Phone Dyno does better is the combination:

• Roll-off detection that ignores shifts. The end rules look for a sustained low stretch with no speed gain and real deceleration, so a one-second shift dip doesn't end the pull, and a lazy partial roll-off still does.
• Auto-trim. Every saved pull is cut to the roll-on to roll-off window by the same fixed rules, so the cruise lead-in and the slowdown never bend the curve, and every pull is trimmed the same way.
• A drum-dyno basis. Counting only the front tire's rolling drag, with rolling resistance that rises with speed, puts the numbers on the same footing as a Dynojet-style sheet.
• Built for real gearing. Live J1349 correction, rpm from each preset's full gearing, and a stock Evo reference curve for the Evo FXR presets.

Two pulls on Evo FXRs, shown as the app computed them. Bike details are kept general on purpose.

The dashed stock Evo reference is Cycle World's rear-wheel dyno run of a bone-stock 1994 Road King with the 1340 Evo, digitized from the published chart (Cycle World, May 1994). It is a reference to eyeball against, not a calibration, and it came from a CV-carb engine.

Simulated pulls

The roll-off rules were tuned on a physics simulator that uses the same road-load model as the app: about one GPS fix per second with timestamp jitter, Doppler speed noise of 0.08 to 0.2 m/s, optional receiver lag of 0.6 s, a cruise lead-in, gear shifts and partial lifts, across three different bike models.

The curve method was tested the same way. With 0.12 m/s of simulated GPS noise it recovered 95.8% of a known torque dip when points were about 160 rpm apart, but only 45.2% when they were about 280 rpm apart, and it read peak horsepower about 3.8% high (median). Those are simulation results, and a fair picture of what 1 Hz GPS can and can't resolve.

What 1 Hz GPS can and can't do

• Phone GPS reports about one fix per second, so over a typical pull the points are often 150 to 300 rpm apart. Sharp, narrow dips and bumps get softened.
• A short pull gives very few points. The analysis wants about 4 or more seconds of steady throttle. Stacking several pulls of the same bike and gear pools their points into one denser curve.
• The model depends on inputs you supply or the preset estimates: total weight, tire radius, drag area, rolling resistance and front weight share.
• There is no wind term and no grade term in the saved-pull analysis, and no rotating-inertia allowance.
• GPS elevation on the live screen is for reference only. Web apps can't read the iPhone's barometer.

• Ride safely and legally. Use a closed course or a place where the speeds involved are legal, with clear sight lines and no traffic. Use a solid, lockable phone mount, and don't touch the phone in traffic.
• Pick a flat, straight road and a calm day. The analysis assumes level ground and has no wind correction.
• Stay in one gear for the whole pull, and choose that gear in the analysis afterward. On a dual-clutch bike, use manual mode.
• Set your real total weight. Enter rider and cargo weight, and check the bike weight and tire radius. They feed straight into the force and rpm math.
• Tap Record, then roll on firmly and hold it for at least 4 seconds. The status line should switch to armed.
• Roll off cleanly. The pull ends by itself about 2 seconds later and saves.
• Stay under open sky. Buildings and trees make GPS worse.
• Check the weather fields when you open the pull, and look at the dots behind the curve. A tight cluster means a cleaner pull.

There is no account, no login and no server that stores your pulls. Settings live in the phone's local storage and pulls in its built-in database. Up to 20 pulls are kept; when a new pull would make 21, the oldest pull that isn't starred is deleted. Starred pulls are never deleted automatically.

Apart from loading and updating its own files, the only network call the app makes is the weather lookup. To get the weather it sends Open-Meteo the location and time of the pull. Nothing else leaves the phone unless you share an export.

Each pull can be shared through the iPhone share sheet as a PDF dyno sheet or an Excel workbook (summary, raw GPS data, calculated points and chart). Privacy tip: exports include location. The PDF shows the pull's start location and the workbook has every fix's latitude and longitude, so think before posting a raw export in public.