Methodology
What the labels mean, how metrics are defined, and what we do not claim.
Last updated 2026-10-08
Value classes
- Manufacturer stated
- Published by the brand (product page, press kit, spec sheet). Recorded verbatim with the source URL and access date. Brands differ in whether stack height includes the insole; we record the brand’s figure and note the convention when known.
- Externally measured
- Measured by a third party with a stated method (for example a review lab with calipers or a flex rig). The source and method are recorded; sizes vary between labs.
- AI estimated
- Estimated by our image-analysis pipeline from public product photographs (side profile for rocker and heel geometry, top and sole views for the toe box shape). Photos are analysed, not republished. Always carries an algorithm version, a confidence score and the input image’s provenance. Values are measured relative to shoe length; the few shown in millimetres are converted for a men’s US 9 and marked approximate.
- Derived
- Computed from other recorded values (for example a composite rocker index). The formula is versioned.
When several values exist for the same metric, the page shows one by precedence: measured › manufacturer stated › derived › AI estimated, then higher confidence, then the most recent. Stack heights and drop are the exception: the brand’s figure is shown first and a lab’s reading (taken at 12 % and 75 % of length on every shoe) beside it, because runners and shops use the brand’s figure and labs often measure at other points. All approved observations remain visible on the shoe page, with a dated change history; nothing is deleted when a better method appears. Why brands, labs and photos disagree.
Three different kinds of certainty
Three things are easy to confuse, and the site keeps them apart:
- Extraction confidence (the meter below): a score from the extractor or algorithm saying how cleanly it read the source or the photo. It is a heuristic, not a probability that the number is right.
- Photo disagreement: for photo estimates, how far apart the different photos of the same shoe are. Shoe pages show the median and the number of photos, and mark a value whose photos disagree more than usual with * as uncertain.
- Validated accuracy: how close an estimate is to an independent measurement of the shoe. Agreement between photos shows consistency, not accuracy: photos taken the same way can share the same error. We have compared the photo estimates with one independent lab (below); we have not measured shoes ourselves.
How the photo estimates compare with lab measurements
In October 2026 we compared the photo estimates with the rocker measurements published by RTINGS for the roughly 110 catalogue shoes their lab has tested. Their lab is a reference, not ground truth, and the two use slightly different definitions, so we report agreement:
| Estimate | Agreement with the lab | What it means |
|---|---|---|
| Toe tip over ground | Ranks shoes well (correlation 0.79). Reads about 8 mm lower, because we measure just behind the tip; allowing for that, typically within about 3 mm. | Differences of 5 mm or more between shoes are meaningful. |
| Rocker rise angle | Ranks shoes well (0.74). Reads about 3–4° steeper; allowing for that, typically within about 2°. | Differences of 3° or more are meaningful. |
| Toe rocker | Moderate (0.65). Typically within about 3 points, while most shoes sit within a few points of each other. | Only clear differences count, so filters and guides compare toe rocker by thirds (shortest, typical, longest), never by a split at the median. |
| Heel bevel height | Moderate (0.64) since we read it at the heel corner (below); it was weak (0.40) when read at a fixed point near the back. Typically within about 5 mm after a constant offset. | Differences of 7 mm or more are meaningful. |
| Heel bevel length | Weak (about 0.4). None of the alternative definitions we tested did clearly better. | A rough indication only; it is not used to rank or filter shoes in the guides. |
Correlations are for the values as displayed (the median of each shoe’s photos for the toe measures). Stack heights from published specs agree closely with the same lab’s measurements (typically within 1.3 mm at the heel and 1.7 mm at the forefoot).
Extraction confidence
Each observation has a confidence between 0 and 1, set by the extractor or algorithm that produced it and shown as High, Medium or Low. Confidence describes how certain we are that the number faithfully represents the source, not how accurate the source itself is. When a reviewer confirms a value against the source text, its confidence is raised to at least 0.85; AI-estimated and derived values are never raised this way, because a reviewer checking an overlay cannot make an estimate certain. Values below the auto-approval threshold are held for human review before they appear.
Rocker geometry: toe rocker, apex, radius, angles and onset
With the shoe resting on a flat surface, viewed from the side, we fit a ground line to the sole’s contact zone and trace how far the outsole lifts off that line along the length. From that profile we report the following. All are read as fractions of shoe length, so an uncalibrated photo is enough; one is then converted to approximate millimetres:
- Toe rocker: how much of the shoe curves up at the front, the length from the toe tip back to the rocker apex as a percentage of shoe length (100 % − apex). It is what the site shows, because it reads the same way as the heel bevel at the other end: bigger means more rocker. A shoe whose curve leaves the ground 64 % of the way from the heel has a 36 % toe rocker.
- Rocker apex: where the forefoot curve leaves the ground, from heel (0 %) to toe (100 %), found as the tangent point of a circle fitted to the forefoot sole. The research literature defines apex as the last point of full ground contact under load; we measure an unloaded shoe in a photograph, so the numbers are close in meaning but not interchangeable.
- Rocker radius: the radius of that circle. A smaller radius is a tighter curve. Reported only when a circle fits the sole well.
- Rocker rise angle: the angle, against the ground, of a straight line from the rocker apex to the toe tip, computed as arctan(toe-tip height ÷ (100 % − apex)) from the same photograph. It summarises how steeply the forefoot rises overall without depending on the curve’s exact shape, so it is a useful check on the apex: an early apex with a modest toe height gives a shallow angle (a long, gentle rocker), a late apex with a high toe a steep one.
- Rocker angle at toe: the local angle of the sole over the last 3.5 % of length, against the ground; steeper than the rise angle on a curved sole.
- Visible rocker lift onset: the position from which the sole stays more than 1.5 % of shoe length above the ground. This is a visibility threshold, so it reads later than the apex; it is kept for continuity and for comparing with earlier figures.
- Toe tip over ground: the height of the underside of the sole above the ground at the front of the outsole, 97 % of shoe length from the heel, just behind the toe bumper. It is read as a fraction of length and converted to millimetres for a men’s US 9, assumed to be 300 mm long. Real lengths of that size vary by a few percent between models, so the millimetre value inherits that uncertainty on top of the photo’s own. Unlike the rise angle, it does not depend on where the rocker starts: a short, late rocker and a long, early one can rise at the same angle but end at very different heights. Lab toe-spring figures measured at the very tip read about 8 mm higher.
- Toe spring (relative) is the height of the very front of the shoe above the ground line, which includes the toe bumper of the upper.
- Heel bevel length and height: how far forward from the back of the heel the sole is visibly off the ground (lifted more than 1.5 % of shoe length), and how high the underside of the sole sits where it turns into the back of the heel (the heel corner: the first point, followed back from the ground, that is steeper than 60°; ≈ mm for a men’s US 9). Reading it at the corner rather than at a fixed point means a flared or overhanging heel is not mistaken for a high bevel. A long, high bevel is a rounded or cut-away heel that rolls into the stride; a short, low one is a square heel. The side-profile drawings use these two numbers to draw the heel; where they are not measured, a generic rounding is drawn.
There is no agreed definition or threshold for a “rocker” in the research literature, and brands rarely publish geometry, so figures from different sources may not match ours.
How photo values are checked
- Several photos per shoe: side views from retailer galleries (several colourways, outer and inner side) and brand product images, each measured independently. Photos that are not a clean, single-shoe side view on a plain background are rejected before measuring.
- Camera correction: each photo source has a small systematic offset (for example, one retailer’s inner-side photos read the apex about 1.3 points later than its outer-side photos). Offsets are estimated from shoes photographed by several sources and removed before comparing.
- Two photos must agree: a value is published only when two different photos agree within a tolerance (apex 3 points, rise angle and toe angle 3–5°, radius 15 %). Radius and angles also require the apex to agree; heel bevel length agrees within 3 points, and its height (within 4 mm) only counts where the length agrees.
- What is shown: for the toe measures (toe rocker, toe tip over ground, rise angle, toe spring) the shown value is the median of all the shoe’s photos, which agreed better with lab measurements than any single pair; heel bevel height is shown the same way. Heel bevel length and toe box shape show the average of the closest agreeing pair. The closest pair agrees well by construction and can hide wider disagreement: across the catalogue, the closest pair of photos differs by about half a point of toe rocker on average, while all photos of the same shoe span about 6 points. That is why shoe pages mark a value with * as uncertain when all its photos span more than about twice the change we treat as noise (toe rocker and rise angle 6, toe tip and heel height 10 mm, heel bevel 4.5 points, toe box shape 4 points): roughly the most uncertain quarter of shoes for each measure. The full range of every photo is in the values file of the open dataset.
These are estimates of the unloaded shoe, labelled AI estimated; see how they compare with lab measurements. Every photo’s measurement stays in the record, including the ones not shown.
Toe box shape
From photographs taken from above and from below, we trace the outline of the shoe and measure its width at 90 % of the length from the heel, as a percentage of the widest point of the forefoot. Around 80 % or more is a roomy, foot-shaped front (Altra and Topo sit here); around 70–73 % is a tapered, pointed front, typical of racing shoes. It describes the outside outline of the shoe, including any midsole flare, not the room inside the upper; for that, see the lab-measured toe box width where available. Several colourways and both views are measured, and a value is published only when two photos agree within 2.5 points (their average is shown).
Stiffness and firmness scores
Two 0–100 scores are derived from RunRepeat’s lab tests. The longitudinal stiffness score comes from the force needed to bend the shoe 30°, rescaled so that about 9 N maps to 5 and 26 N to 95. The midsole firmness score comes from a durometer reading of the midsole foam (Asker C), with about 22 mapping to 5 and 45 to 95; higher is firmer. Firmness matters for rockers: the shape is measured on an unloaded shoe, and under body weight a softer forefoot compresses more and partly flattens the curve. Both scores rank shoes against each other; neither is a physical quantity, and they exist only for shoes that lab has tested. Each shoe page shows the lab’s original reading next to the score (for example “25.3 N, force to bend the shoe 30°” or “35.1 Asker C”), so the source value is never replaced by our scale. The test size the lab used is not recorded here; check the lab page if size matters.
How the best lists are ranked
Each best list ranks the current shoes matching a stated rule (a category, a characteristic, or having a measured value) on four to six things its page names, some counting double: measured values such as heel stack, weight, toe rocker or lab-measured widths, how widely each shoe is bought (Running Warehouse best-seller flags and review counts, plus visits here, refreshed weekly) and how those customers rate it. A rating is pulled toward the average of all rated shoes as if it had ten more reviews at that average, so a 5.0 from two buyers counts for less than a 4.6 from fifty, and it is quoted on a pick only from ten reviews. Every value is compared with the median of the shoes ranked and divided by their interquartile range, or by twice the measurement noise if that is larger, then capped at two such steps either way, so differences within noise count for little and one extreme number cannot carry a shoe. Photo estimates whose photos disagree count half; a missing value counts as typical. A list shows about half the shoes it ranks, at most seven and at most three per brand, and is published only when it ranks six or more. “Stands out” and “Trade-off” quote a pick’s value against that median only when the difference is beyond noise (3 mm for stack heights, because brands measure them differently). Nothing is tested on foot.
Metric definitions
| Metric | Unit | Definition |
|---|---|---|
| Heel-to-toe drop drop_mm | mm | Difference between heel and forefoot stack height. |
| Heel stack heel_stack_mm | mm | Total sole thickness under the heel (midsole + outsole, sometimes insole, per source convention). |
| Forefoot stack forefoot_stack_mm | mm | Total sole thickness under the forefoot. |
| Weight weight_g | g | Weight of one shoe. Size varies by source; see size_variant on the observation. |
| Toe rocker toe_rocker_pct | % | Length of the curved front of the sole, from the toe tip back to where the forefoot curve leaves the ground (the rocker apex), as a percentage of shoe length: 100 − apex. Higher = longer, earlier-starting toe rocker. Estimated from photos. |
| Rocker radius rocker_radius_pct | % | Radius of curvature of the forefoot rocker as % of shoe length (circle fit from apex to toe). Smaller means a tighter curve. Only reported when a circle fits the sole well. Estimated from photographs. |
| Rocker rise angle rocker_chord_angle_deg | deg | Angle, against the ground, of a straight line from the rocker apex to the toe tip: how steeply the forefoot rises overall. Computed from the apex and the toe-tip height of the same photograph, without curve fitting. Estimated from photographs. |
| Rocker angle at toe rocker_angle_deg | deg | Angle of the sole near the toe tip against the ground. Larger means the toe rises more steeply. Estimated from photographs. |
| Toe tip over ground toe_tip_height_mm | mm | Height of the underside of the sole above the ground at the front of the outsole (97 % of shoe length), with the shoe standing unloaded. Measured from photographs as a fraction of shoe length and converted for a men's US 9 (assumed 300 mm long), so it is approximate (about ±3 mm). Estimated from photographs. |
| Visible rocker lift onset rocker_onset_pct | % | Position along the shoe (heel = 0 %, toe = 100 %) from which the sole stays more than 1.5 % of shoe length above the ground. A visibility threshold, so it reads later than where the curve begins (see rocker apex). Estimated from photographs. |
| Rocker curvature (legacy score) rocker_curvature_score | score | Deprecated 0-100 score that saturates near 100 and is no longer published; superseded by rocker radius. |
| Toe spring toe_spring_mm | mm | Height of the toe tip above the ground plane, in millimetres (only from calibrated sources). |
| Toe spring (relative) toe_spring_relative | % | Toe-tip height above the ground plane as a percentage of shoe length (image-derived, uncalibrated). |
| Heel bevel length heel_bevel_pct | % | How far forward from the back of the heel the sole is visibly lifted off the ground (more than 1.5 % of shoe length), as a percentage of shoe length. Longer = a more rounded or bevelled heel. Estimated from photographs. |
| Heel bevel height heel_bevel_height_mm | mm | Height of the underside of the sole above the ground where it turns into the back of the heel (the heel corner), with the shoe standing unloaded: how strongly the heel is rounded or cut away. A rounded heel turns late and high; a flat underside with a flared or overhanging heel turns early and low. Measured as a fraction of shoe length and converted for a men's US 9 (assumed 300 mm), so approximate. Median of the shoe's photos. Estimated from photos. |
| Heel bevel angle heel_bevel_angle_deg | deg | Angle of the heel bevel relative to the ground plane. |
| Midsole width (forefoot) midsole_width_forefoot_mm | mm | Width of the midsole at its widest forefoot point, as measured by a lab. The platform the foot lands on; says nothing about interior room. |
| Midsole width (heel) midsole_width_heel_mm | mm | Width of the midsole at the heel, as measured by a lab. |
| Toe box shape toe_width_ratio_pct | % | How much of the shoe's width is left near the tip: the outline width at 90 % of length as a percentage of the widest forefoot width, from top and sole photographs. Higher = a roomier, more foot-shaped front; lower = a more tapered, pointed front. It describes the outline of the shoe, not the room inside. Estimated from photographs. |
| Forefoot platform width forefoot_width_pct | % | Widest sole width in the forefoot as a percentage of shoe length, from a sole photograph (image-derived, uncalibrated, approximate). |
| Heel platform width heel_width_pct | % | Sole width near the heel as a percentage of shoe length, from a sole photograph (image-derived, uncalibrated, approximate). |
| Interior forefoot width interior_forefoot_width_mm | mm | Internal width at the widest forefoot point as measured by a lab ("width / fit"). A fit characteristic, not a platform measure. |
| Toebox width toebox_width_mm | mm | Internal width at the big-toe area as measured by a lab. A fit characteristic. |
| Longitudinal stiffness longitudinal_stiffness_score | score | Resistance to bending along the shoe length, 0–100 relative scale. Method-specific. |
| Forefoot flex stiffness forefoot_stiffness_score | score | Resistance to bending at the metatarsophalangeal flex zone, 0–100 relative scale. |
| Torsional stiffness torsional_stiffness_score | score | Resistance to twisting about the long axis, 0–100 relative scale. |
| Midsole firmness midsole_firmness_score | score | How firm the midsole foam is, 0-100 (higher = firmer), derived from a lab durometer reading. Under body weight a softer forefoot compresses more and partly flattens the rocker. A ranking, not a hardness value. |
| Rocker Index rocker_index | score | Composite 0–100 score combining rocker onset, curvature and toe spring. Derived; definition versioned via algorithm_version. |
What we do not claim
ShoeGeometry describes products. It does not assess individuals, diagnose conditions or recommend treatment. Image-derived values are estimates and are labelled as such. Sources are credited and linked; we store brief factual extracts and our own derived numbers, not copies of copyrighted articles or photographs.