AI Notice

This is an estimate, not a test

Milesleft estimates functional degradation and remaining wears from what is visible in a photograph. That is all it has to work with. It is not a test, it is not a measurement of the fabric itself, and it produces no certificate, grade or verdict.

A vision model reads four visual indicators off the two photographs: translucency at the heel and ball, bobble density, how far the rib sits open at rest, and how far the surface has sealed. Those four numbers are the only thing the model produces. Everything after that — the four bars, the range of wears left, which function goes first, the retirement conditions — is arithmetic that runs in your browser against a fixed model. The model does not count wears.

Because the arithmetic is fixed, the same photographs always give the same read-out. That is deliberate: a figure that moves about between runs would be worse than useless for deciding what to do with a pair.

What it cannot know

Three things change the answer substantially and none of them are visible in a photograph:

• How many times the pair has actually been worn. Whatever you tell us is a recollection, and recollections run low.
• How it has been laundered. Wash temperature, detergent load and above all whether it has been through a tumble dryer change the curve more than almost anything else.
• Your own sensitivity to discomfort. Two people can wear the same pair to opposite conclusions and both be right.

Every figure is therefore given as a range, and the range widens when there is less to go on. There is no single number anywhere in a read-out. Every read-out also carries its own list of what it could not know about that particular pair, and that list is never empty.

No health judgement of any kind

This service makes no health judgement whatsoever. Reduced cushioning is described as a physical fact about the fabric — how much loft is left in the pile — and it is never used to infer any effect on your feet.

The service reaches no conclusion about pain, blistering, skin, circulation, or any other matter concerning your body, and it never will. It is not a medical device, it performs no diagnosis and it offers no clinical opinion.

It accepts no photographs of feet and no facility to upload one exists anywhere in it. That is an absence of a feature, not a filter applied afterwards.

If what you write in the note is about how your feet feel rather than about the state of the sock, the whole analysis is skipped and you are told:

Anything about how your feet feel needs a podiatrist, not a sock tool. We only look at the sock.

That branch runs before anything is computed, on your device, and no result is produced at all. Anything concerning how your feet feel should go to a podiatrist or a doctor.

Looking worn is not the same as being finished

A pair that looks thoroughly used often has most of what matters intact. Bobbling is usually a surface phenomenon: short fibre ends work loose and tangle on top of the knit while the loop structure underneath is untouched. Matting — the loops themselves collapsing into a sealed plateau — looks superficially similar in a photograph and means something quite different.

Every read-out therefore carries a section saying plainly which pairs can still be worn and which signs of age do not shorten what is left. That section is shown at the same visual weight as the degradation bars, is never watermarked on the free plan, and is never behind a payment. A tool that only ever said throw it out would create exactly the waste this one exists to prevent.

No verdict on any brand or product

The service evaluates no brand, reaches no conclusion about the quality of any product, ranks nothing for durability, and gives no restocking or purchasing advice. Any maker’s mark the model reads in a photograph is masked before it appears in a read-out, a shared image or a link preview.

When a read-out says a pair should be retired, it stops there. It does not say what to get instead, and it never will.

The terms of service expressly prohibit using the output of this service as evidence or support in a return, a complaint, a warranty claim, a chargeback or a quality claim. It is not built for that and is not reliable for it. See section 3 of the terms of service.

Prohibited high-risk uses

The service and its output must not be used for medical purposes, or for any legal, financial, safety-critical, employment, credit, housing, insurance or judicial decision about any person. These are not edge cases we have not thought about; they are prohibited.

Your photographs are not training data

Nothing you upload and nothing you write is used to train, fine-tune, evaluate or benchmark any model — ours or anyone else’s. Photographs are kept for 30 days so you can reopen a read-out and are then deleted automatically. You can delete any of them, or your whole account and everything in it, at any time. See the privacy notice.

Getting it wrong

There are pairs this will read badly: unusual constructions, dyes that behave oddly under backlight, photographs taken at a steep angle or in poor light. If a read-out looks wrong, tell us what the pair actually looks like and what it said. That is how the model gets better, and it is the only thing we use those reports for.

Use a read-out as one input to your own judgement about a pair of socks. That is the whole of what it is for.