Research · Bench test
Bench test: how accurate is StepSavvy’s manual heel‑angle tool?
On printed angles with known values, readings from the app’s manual heel-angle tool landed within 0.16° of the true angle on average. That checks the tool’s math and photo capture under ideal conditions — it is not a test on real feet, and it is not a health claim.
What was tested
- The manual heel-angle tool (“Measure your heel angle”), where you place 3 dots per leg yourself
- On printed lines at nine known angles, from −20° to +20°
- Photographed with a phone, with dots placed by two podiatry-student raters
What was not tested
- Real feet, skin or legs
- The automatic heel reading, where the app finds the points for you
- Walking (gait) analysis, Form Check or any AI feature
- Any health outcome
01Overview
The short version
StepSavvy has a tool that measures your heel angle from a photo of the back of your legs. You drag three dots onto each leg, and the app works out how far your heel tilts away from the line of your lower leg.
Before anyone tests that on real people, there is a simpler question to answer first: when the dots are in the right place, does the app calculate the angle correctly, even after a real phone camera is involved? To find out, we printed lines at angles we already knew, photographed them, and measured them in the app.
The readings were very close to the true angles — 0.16° off on average, and every goal we set before collecting data was met. For scale, 0.16° is about the tilt of a line that drifts 1 mm sideways over 36 cm (about 14 inches).
What it doesn’t show: how well the tool works on a real heel. On a real leg, the hard part is putting the dots in the right place, and printed lines make that easy. So treat this as a check of the tool’s math and capture, not of its accuracy on your feet.
02Method
What we tested
The tool. In the app’s “Measure your heel angle” screen, you take or pick a photo of the back of your lower legs, then drag three dots onto each leg: one at mid-calf, one in the middle of the ankle, and one at the centre of the heel. The app draws a line through the first two (your lower leg) and a line through the last two (your heel), and reports the angle between them.
The targets. Instead of feet, we used line drawings of a “leg” and “heel” at nine set angles: −20°, −15°, −10°, −5°, 0°, +5°, +10°, +15° and +20°. Each printed target was checked with a digital angle gauge (manufacturer-stated accuracy ±0.2°), and the gauge’s value — not the printed label — was used as the true answer. Three targets measured 0.2° off their label.
The photos. Each printed target was taped flat and photographed five times with a phone at about 45 cm, with the camera fully moved and set up again between shots (45 photos in all).
The raters. Two fourth-year podiatric medicine (DPM) students (at the time of the study) each placed the dots on all 45 photos, separately (90 readings). As a second check, each also measured every printed target three times with a manual goniometer — the hinged protractor clinicians use (54 readings).
Digital targets (no camera). Separately, the same nine angles were drawn digitally and measured five times each in the app by rater 1, with no camera involved (45 readings). A person still placed the dots, so this part checks the math plus dot placement, without the camera.
Study plan
The pass/fail goals and the sample size were written down before any data were collected. Two things changed from the original plan: the targets became printed lines, and the raters were not blinded. The goals and sample size did not change.
03Findings
Results
How close was the app’s average reading at each target?
Each dot is the average of the 10 app readings (both raters) at one printed target, minus its gauge-verified true angle. The shaded band is the 1.2° average-error limit set in advance, drawn either side of zero for scale. Readings above and below the true angle partly cancel in an average; the ±0.4° limits below describe individual readings.
- Average of 10 readings minus true angle
- 1.2° average-error limit set in advance
Accuracy. Across the 90 readings taken from photos, the app was off by 0.16° on average. In statistical terms, the 95% limits of agreement with the true angle were −0.42° to +0.39°. Put simply, 89 of the 90 readings landed within 0.4° of the true angle.
Two raters, close agreement. When the two raters placed dots on the same photo, their readings differed by 0.14° on average (the largest gap was 0.40°).
Every pre-set goal was met. All three goals we fixed before collecting data were met with wide margins (table below). In total the study recorded 198 data points: 9 gauge checks, 45 digital readings, 90 photo readings and 54 goniometer readings.
| Measure | Goal | Result | Met? |
|---|---|---|---|
| Bias (average signed error) | within ±1.0° | −0.01° | Met |
| Average error (MAE) | 1.2° or less | 0.16° | Met |
| 95% limits of agreement | within ±3.0° | −0.42° to +0.39° | Met |
Sorting heels into categories
The app also sorts each reading into one of three bands: varus (heel tilted one way), neutral (within 5° of straight, either way, including exactly 5.0°), or valgus (tilted the other way). The ±5° cut-off is the app’s own convention, not a clinical cut-off, and this test did not check whether it describes real feet.
Applying the app’s ±5° rule to the recorded readings, 83 of 90 (92%) fell in the same band as the true angle. On the seven targets well away from the 5° line, every reading fell in the same band (70 of 70). The other two targets sat right on the line or within 0.2° of it (true values −5.0° and +5.2°); there, an error smaller than 0.3° is enough to cross into the next band, and 13 of 20 readings matched. Five of the seven misses were readings of exactly 5.0°, so the result at the line depends on which side the rule counts 5.0° as: counting exactly 5.0° as outside the neutral band instead, 87 of 90 match.
The study recorded the number the app showed, rounded to 0.1°, not the label the app displayed. The bands above were applied to those recorded numbers afterwards. The app labels from the unrounded angle, so a reading shown as 5.0° could carry either label in the app.
What that means for you: if a reading is close to 5°, the label (neutral or not) is less meaningful than the number itself. On real feet, errors will be larger than on printed lines, so labels near the line deserve even more caution.
The goniometer check
Both methods landed far inside the 1.2° limit we set in advance for the app: average error was 0.16° for the app and 0.10° for the goniometer. No equivalence test was planned or done, so these numbers should not be read as showing the two methods give the same results. A check done afterwards (not planned in advance) found the goniometer’s average error was lower than the app’s on these printed targets, by more than chance would explain (Welch’s t-test, p ≈ 0.02).
The goniometer’s lower error most likely comes from the printed targets, not from a better instrument. Printed lines remove the goniometer’s main real-world difficulty — finding landmarks on a real heel by feel — and because both raters knew which target they were measuring, 49 of the 54 goniometer readings simply equalled the printed label.
04Scope
What this test did not cover
These results apply only to the manual dot tool, under the conditions above. They do not apply to:
- Real feet. No skin, soft tissue, hair or bony landmarks were involved. Finding the right spots on a real leg and heel is the hardest part of this measurement, and this test left it out.
- Everyday users. Both raters were podiatry students who knew what they were looking for. People placing dots on their own photos were not tested.
- The automatic heel reading. In Foot Photo Analysis, the app can find the leg and heel points for you. That works differently and was not part of this test, so these numbers don’t apply to it.
- Walking (gait) analysis, Form Check, the AI foot type analysis, or shoe suggestions. None of these were tested.
- Any health outcome. This test measured how closely the tool reads an angle. It did not look at pain, injury, recovery, or whether knowing your heel angle helps with anything.
05Caveats
Limitations
- Printed lines are not feet. There was no skin, soft tissue, hair or bony landmark to find.
- Not independent. StepSavvy’s founder made the targets, set the reference (“true”) values, took all the photos, and was rater 1.
- Raters were not blinded. Both raters knew which target they were measuring.
- One setup. One phone, one app version and one lighting setup were used, and the camera was always square-on to the target at about the same distance. Tilt, viewing angle, distance and lighting were not varied, so their effect is unknown.
- The reference has limits too. Each true value comes from a single reading of a gauge rated to ±0.2°, so errors as small as the app’s are near the limit of what the reference can resolve.
- Small and short. Nine targets, two raters, collected in July 2026. No comparison between the app and the goniometer was planned. A check done afterwards found the goniometer’s average error was lower on these printed targets (see the goniometer check above).
App version. Testing used the July 2026 version of the app. Since then the dot tool has gained a magnifier and placement tips, and the left/right foot labels were corrected; the angle it calculates for each heel in a photo is unchanged.
06Next steps
What’s next
Technical detail for researchers
ShowHide design, analysis and accuracy tables
Design. Two-part bench (phantom) study of nine known rearfoot (calcaneal–tibial) angles from −20° to +20° in 5° steps. Non-biological targets only; not human-subjects research.
Part 1 — measurement engine (digital). Each angle was rendered digitally and measured five times in the app by rater 1 (45 readings), isolating the computation and landmark placement from photographic capture. Readings reported to 0.1°.
Part 2 — real capture (printed, photographed). The nine angles were printed and each verified once with a digital angle gauge (manufacturer-stated ±0.2°); the gauge value served as ground truth. Each target was photographed five times at ~45 cm with the camera fully re-placed between shots (45 images). Two raters independently placed landmarks on all 45 images (90 app readings), and each measured every printed target three times with a manual goniometer read to a 0.5° scale (54 readings).
Analysis. Accuracy: bias, mean absolute error (MAE), 95% Bland–Altman limits of agreement, and calibration regression. Repeatability: within-target SD, standard error of measurement (SEM) and minimal detectable change (MDC₉₅ = 2.77 × SEM). Classification used the app’s ±5° neutral band (neutral = −5.0° ≤ angle ≤ +5.0°). Sample size (9 targets × 5 photographs × 2 raters = 90) was fixed in advance. No hypothesis or equivalence test was planned. Post hoc, a Welch t-test on absolute errors (app n = 90 vs. goniometer n = 54; MAE 0.156° vs. 0.098°) gave t = 2.40, p ≈ 0.018, favouring the goniometer; it treats repeated readings of the same target as independent and was not pre-specified.
Accuracy by arm (degrees)
| Arm | n | Bias | MAE | 95% limits | Max |
|---|---|---|---|---|---|
| Part 1 engine (digital) | 45 | −0.05 | 0.25 | −0.63 to +0.53 | 0.8 |
| Part 2 app, rater 1 | 45 | +0.02 | 0.16 | −0.40 to +0.43 | 0.4 |
| Part 2 app, rater 2 | 45 | −0.04 | 0.15 | −0.43 to +0.36 | 0.5 |
| Part 2 app, pooled | 90 | −0.01 | 0.16 | −0.42 to +0.39 | 0.5 |
Part 1 is measured against exact digital construction values; Part 2 against gauge-verified truth. Pooled SEM 0.18°, MDC₉₅ 0.49°. Calibration slope 1.006 (95% CI 1.002–1.009). Part 1 and Part 2 differ in reading resolution, rater count and session, so comparisons between them are descriptive only.
| True angle (gauge) | App mean (SD) |
|---|---|
| −20.0 | −20.20 (0.13) |
| −15.2 | −15.18 (0.18) |
| −10.0 | −10.17 (0.20) |
| −5.0 | −4.95 (0.21) |
| 0.0 | +0.00 (0.21) |
| +5.2 | +5.14 (0.16) |
| +10.0 | +10.18 (0.18) |
| +15.2 | +15.16 (0.16) |
| +20.0 | +20.12 (0.17) |
Camera re-placement. With the camera fully re-set between every photo, readings of the same target varied with an SD of 0.18° (this includes the raters’ dot-placement variation). There was no fixed-camera comparison. Perspective angle, tilt, distance and lighting were held constant and remain untested.
How the tool computes the angle. The manual tool calculates the angle directly from the on-screen positions of the dots you place. The automatic reading in Foot Photo Analysis locates the points with on-device pose detection instead; it was not evaluated here.
07Disclosures
Study team, funding and conflicts of interest
Study team. Mahin Acharya (StepSavvy founder; fourth-year DPM student at the time of the study; rater 1) and a second fourth-year DPM student rater (rater 2).
Funding. No external funding.
Conflicts of interest. Mahin Acharya is the founder of, and holds an equity interest in, StepSavvy LLC, and served as rater 1: he produced and photographed all targets, fixed all ground-truth gauge values, and contributed all 45 Part 1 readings and 45 of the 90 Part 2 app readings. The second rater declares no competing interest. Neither rater was independent of the developer and both were unblinded; ground truth was fixed by an external digital angle gauge (manufacturer-stated ±0.2°) to limit, but not eliminate, this bias.
08Citation
Poster and citation
The study was also written up as a poster for a clinical and research audience. A version of the full poster for a general audience may be posted here later.
How to cite
StepSavvy LLC. Bench test of StepSavvy’s manual heel-angle tool on printed known angles. Study conducted July 2026. https://stepsavvy.app/research/heel-angle-bench-test/
Wellness disclaimer
StepSavvy’s heel-angle tool is a general-wellness measurement aid. It does not diagnose, predict, or stage any disease or medical condition, and it is not a substitute for an examination by a qualified clinician. StepSavvy has not been evaluated by the U.S. Food and Drug Administration.
This bench test measured how closely the tool reads printed angles. It says nothing about your health, and the readings you get on your own feet may be less precise. If you have a concern about your feet, talk with your physician, podiatrist or physical therapist. See our Terms of Service for more.