How the estimate works, and where it is blind

The service estimates the risk that a garment will transfer dye onto other items in the same wash. It does this from two things: the fibre content printed on the composition label, and the depth and family of the colour as seen in a photograph.

This is an estimate. It is not a colourfastness test. A colourfastness test wets, rubs and measures the actual fabric under controlled laboratory conditions to a published standard. Nothing on this site does that, and nothing on this site should be described to anyone else as having done that.

The judgement itself is made by a fixed table held in this application — fibre chemistry, colour depth, colour family, washes so far. The table is deterministic: the same inputs always produce the same reading, and the reasoning shown on your card is the actual reasoning, not a plausible-sounding explanation generated afterwards.

An AI model is used for exactly two narrow tasks:

• reading the fibre names and percentages off the label photograph;
• estimating the depth band and colour family from the garment photograph.

The model does not decide your risk level, does not write the reasons, does not write the first-wash plan and does not write the self-test. Those come from the table and from fixed text. This division is deliberate: a model asked to judge would produce a fluent answer whether or not it had grounds for one, and fluency is not the same thing as being right.

The model can misread a label — a worn print, a bad angle, an unusual fibre name. What it read is always shown to you on the card, so you can see whether it got the label right before you trust anything built on top of it.

The dyeing method — reactive, direct, sulphur or acid — is not printed on a composition label. The service cannot read it, and neither can you. It is one of the factors that decides whether an item bleeds.

Two garments can carry identical composition labels, identical colours and behave completely differently in water, because they were dyed by different methods. A reactive dye forms a covalent bond with the fibre; a direct dye does not. Nothing on the garment tells you which you are holding.

The same applies to:

• whether the item was given a fixing treatment after dyeing, which can change the outcome entirely and is never disclosed;
• your water — hardness and mineral content change how much dye comes off and where it settles;
• your detergent, its dose, your machine, and whether the drum is overloaded, since abrasion between garments is what lifts surface dye;
• how the item was stored, and what it has already been pressed against while damp;
• anything printed, coated, embroidered or bonded onto the garment, which behaves differently from the base cloth.

Because of section 3, this service never states that an item will survive a mixed wash, never states that mixing is fine, and gives no assurance of any kind about what will come out of your machine.

The lowest of the three levels is expressed as USUALLY SAFE, and usually is explicitly not always. That level means only that nothing in the composition or the colour pointed to a transfer risk. It is a statement about what was found, not a clearance.

The three levels are also drawn using shape and wording rather than a red-amber-green scale. A page about colour cannot use colour to carry its verdict, and a person who does not see hue the way a designer does must still be able to read it.

Where the evidence is ambiguous, the reading errs towards washing separately. An unnecessary separate wash costs one cycle. A wrong reassurance costs a drum of clothes that cannot be put back.

Because we cannot read the dyeing method, a physical test on your actual garment is more reliable than our estimate. It takes thirty seconds, costs nothing, needs nothing you do not own, and damages nothing.

1. Take a plain white cotton cloth — an old tee shirt, a handkerchief, a cotton pad. It must be white and it must be cotton.
2. Dampen it with cold water and wring it out so it is wet but not dripping.
3. Find a hidden part of the garment: an inside seam allowance, the inside of a hem, under a collar.
4. Press the damp cloth firmly against that spot and hold it there for about thirty seconds. Press hard — the pressure matters more than the time.
5. Lift the cloth away and look at it in daylight, near a window, not under a bulb.

Any colour at all on the white cloth means dye is moving. A faint tint is enough to change what you do — it will be far stronger across a whole garment in a drum of water than it is under your thumb. If the cloth stays white, you have learned something the estimate on this page cannot tell you.

The test is on the free plan, is never watermarked, is never blurred and is never placed behind a payment. A tool that hid the more reliable method behind a paywall would be a worse tool, and it would push people into gambling a wash rather than checking.

This service recommends no dye fixative, no colour catcher, no detergent, no wash bag and no other product, and it has no commercial relationship with any brand or retailer of such products. Nobody pays to appear here, because nothing appears here.

What you get instead are general handling steps that cost nothing: cold water, inside out, a short cycle, a low spin, out of the machine promptly, dried away from anything pale.

We also do not repeat the household remedies that circulate about this subject — soaking in salt or vinegar and similar. They are not supported by evidence and passing them on would be dressing up a guess as advice.

The service does not evaluate any clothing brand, does not maintain a list of brands whose colours run, and never presents dye transfer as any brand’s fault. Deep colour on a natural fibre behaves the way the chemistry makes it behave. That is not a defect and it is not anybody’s failing.

We accept no responsibility for what happens in your machine. The decision about what goes in the drum is yours, taken with information we have told you is incomplete.

That limitation concerns the accuracy of the estimate only. It is not an exclusion of your statutory rights in relation to this service itself. Under the Consumer Rights Act 2015 this service must be of satisfactory quality, fit for purpose and as described, and section 47 of that Act voids any term attempting to exclude those rights. If the service is faulty or does not work as described, you have rights against us and nothing here touches them. See the terms of service.

• As evidence in a claim, return, chargeback or complaint against a retailer, manufacturer or brand. The output is an estimate from a photograph and is not capable of establishing anything about a specific garment.
• As a substitute for laboratory colourfastness testing, or as a basis for telling anyone that testing has been done.
• For any medical, legal, financial, safety, employment, credit, housing or judicial decision.
• To tell a third party that dye transfer has been ruled out for an item.