The inside of a shoe is not a number
A size is one figure printed on a box. The inside of the shoe is a surface, shaped by the last it was built on, the material it was built from, the way it closes, the sole it stands on and the way its sizes were derived. Fit happens between that surface and your foot. This is what it takes to know the surface before the shoe exists.
When two people with the same foot length buy the same shoe in the same size and one of them sends it back, the size chart has already failed. It failed because it compares one number on the foot with one number on the box, and neither of those numbers describes the thing that actually touches the foot.
We spent the better part of two years learning what does describe it. Most of what we learned is not a secret. It is simply hard, and it is hard in ways that are easy to underestimate from the outside. This article lays out the problems. It does not lay out our solutions.
Three things that must never be confused
The first lesson was about vocabulary. Foot, shoe and last are measured in three different worlds, and almost every fit error we have seen begins when a number from one world is used as if it belonged to another.
The foot
anatomy · millimetres
Length, ball width, ball girth, instep height, heel width and some twenty further parameters, each defined at an anatomical landmark. A foot has no size. It has a shape.
The shoe
commerce · a size label
What the shopper sees: a size, a width if you are lucky, a material name, a closure. A shoe has no foot length in millimetres. It has a label that maps, loosely, to a last.
The last
geometry · a 3D form
The solid shape the shoe was built around. This is where the designer's intent lives. It is upstream of every pair, every size and every colourway that will ever carry that model name.
A foot is anatomy. A shoe is a commercial object. A last is geometry. Fit is a relationship between the first and the third, mediated by the second. The moment a system starts subtracting a foot length from a shoe size, or reading a last dimension as if it were an interior dimension, it is guessing.
Why the last is the right place to start
Everything else in footwear is downstream of the last. A single last typically sits behind a dozen or more articles across seasons and colours. Measure it once, correctly, and you know something about every one of them before they are made, before they are photographed, before anyone has bought and returned one.
That is the opposite of the approach most sizing tools take, which is to wait for enough purchase and return data to accumulate around a finished product and then infer what the shoe must be like from how people reacted to it. That approach can only describe shoes that already exist and have already been returned. It cannot say anything about a new model, and it cannot tell a designer what to change.
But the last is only the beginning. The naive assumption, and the one we started with, is that the inside of the shoe is the outside of the last. It is not.
What the last alone does not tell you
Take a last, measure its outer surface at the ball, the instep and the heel, and compare those figures with a foot. Every single zone comes out wrong, and the errors are large enough to make a well-fitting shoe look like a bad one. Between the last surface and the foot there is a layer of things the last file knows nothing about.
| Factor | What it does to the interior |
|---|---|
| Material | A stiff full-grain leather, a soft nappa and a knitted textile built on the same last give three different interiors, and two of them change shape during the first weeks of wear. Thickness and give both matter, and they matter differently in different zones. |
| Closure | A lace-up can accommodate a range of instep heights. A slip-on cannot. The same last, closed differently, tolerates a different population of feet, and the difference is largest exactly where feet vary most. |
| Sole and heel | Heel height, platform height and sole stiffness change the angle the foot sits at and therefore where the ball of the foot lands inside the shoe. A last dimension measured at one position is meaningless if the foot ends up somewhere else. |
| Grading | A factory usually measures one master size. The others are derived by grading rules that are rarely uniform: length, width and girth do not scale together, and the rules differ between factories. Assuming they do scale together produces errors that grow with distance from the master size. |
| Design tip | A pointed or elongated toe is air. The last is longer than any foot it is meant for, and the part that matters for fit ends well before the tip. Measure the whole length and you will recommend a size too large for everyone. |
| Coordinate convention | A prosaic one, but real: last files from different factories put length, width and height on different axes. Read one with the wrong convention and every measurement is a measurement of something else. |
None of these is exotic. Every last-maker and every pattern-cutter knows them. What is unusual is having to make all of them explicit, at the same time, in a form a computer can apply to a foot it has never seen and a shoe that has not been made yet.
Every parameter has its own rules
The second thing that surprised us was that fit is not one comparison but many, and they do not behave alike.
Toe clearance is a question of length, but the length that matters is measured from the heel to a point that depends on the shape of the toe box, not to the tip. Ball width is a question of where the widest part of the foot meets the widest part of the shoe, and both of those are positions, not just dimensions. Instep height is almost meaningless without knowing the closure. Heel width is measured at a point where the sole flares most, so the raw figure from a last file usually includes the sole and not just the cavity.
Each zone also fails differently. A shoe that is slightly too long is an inconvenience. A shoe that is slightly too narrow at the ball is a return. A system that treats a millimetre of error as a millimetre of error, regardless of where it falls or in which direction, is not describing how feet experience shoes.
And the rules are not symmetric. Too tight and too loose are not mirror images. They have different consequences, different tolerances, and different remedies, and they interact with the material in different ways.
The foot has to be right first
All of this is wasted if the foot measurement is wrong, and foot measurement from a phone camera has a poor reputation, largely deserved. We share the scepticism. So before we let anyone scan, we measured one reference foot by two independent physical methods and then scanned it, repeatedly, under the conditions a real customer would have: a sheet of paper on the floor, a phone, no tripod, no app store.
Bias +0.4 mm · σ 2.9 mm · 100 % within ±8 mm · 100 % accepted automatically
Repeatability of foot length on one reference foot across 14 scans with a standard iPhone, using foot-reconstruction technology licensed from the Instituto de Biomecánica de Valencia. The reference length was established by two physical methods that agreed to the millimetre. Width validated to ±1 mm against a pencil tracing. This is a repeatability study, not a population study; it tells you how consistently the method measures, not how it performs across every foot shape. That larger study is in progress.
The same measurement pipeline now runs in Safari without any app installed. The first browser scan landed within 1.2 mm of the reference.
The route to those figures was not clever mathematics. It was working out, one at a time, which small things ruin a phone photograph for measurement purposes: the resolution the browser actually delivers versus the resolution the sensor has, the moment the device angle is read relative to the moment the shutter fires, what the camera app silently does to orientation. Each of those alone produces an error of a few percent. Stacked, they produced a foot 16 percent too long and made a sound method look impossible. Removed one at a time, the method was correct all along.
We publish the problems. The conversion of these principles into a fit prediction is our method, and we do not publish that.
What we ask of a manufacturer
One file per last. Not the shoes, not the patterns, not the grading sheet: the last file, in whatever format your last-maker already produces. Ten lasts is usually enough to cover a substantial share of a range, because one last sits behind many articles.
The file does not leave us. It is normalised once, stored as measurements, and used to answer one question for one foot at a time: does this shoe fit this person, and if not, in which zone and in which direction. That answer works in your own store, in a retailer's store, and in the shopping agents that are beginning to buy on people's behalf. It also lets you check your own production against your own last, which is a use nobody expected and everybody wanted.
We are at Simac Tanning Tech and MICAM in Milan, 13–16 September 2026. If you make lasts, make shoes, or make the software that makes either, we would like to meet.