Home / Journal / Spec-to-bulk tolerance
Why crochet reorders drift, and how to write tolerance before it happens
Spec-to-bulk tolerance is the written allowance between the approved sample and the production run: how far a finished piece may sit from that sample on size, weight, fill, stitch density and shade, and what happens when it falls outside. It matters because the thing that most often breaks a crochet reorder is not price and not design — it is a second run that arrived slightly different from the first, at a moment when nobody had agreed what "slightly" was allowed to mean.
What drift is, and what it is not
Drift is the distance between a reference and a production run. It is not the same thing as a fault. A crochet toy that is two millimetres taller than its sample, made correctly, from the approved yarn, on the approved pattern, is not defective; it is what handmade production looks like. The problem starts when the reference itself was never fixed, so the first carton becomes the de facto standard and the second carton is measured against a memory.
There are three places drift enters a crochet line, and they are worth separating because they are fixed in different ways.
| Source | What it looks like | Where it is controlled |
|---|---|---|
| Yarn lot | A shade that reads slightly warmer or cooler than the previous run | Approved physical standard plus a lab dip before dyeing |
| Hand tension | A limb a few millimetres longer, a body slightly firmer | Gauge swatch, single artisan per SKU where volume allows, and a measured band |
| Fill and finishing | A toy that stands on one run and slumps on the next | Fill weight per piece, checked by scale rather than by eye |
Only the first of those is a material problem. The other two are process problems, and process problems are cheap to solve at the specification stage and expensive to solve after the container has sailed.
The five tolerances that decide whether a reorder matches
Most sourcing documents specify far too much and tolerate nothing. A crochet spec that runs to nine pages of photographs but never states a number is not a specification; it is a mood board. Five numbers do most of the work.
| # | Tolerance | Why it earns its place |
|---|---|---|
| 1 | Finished dimensions | The only thing the retail buyer can measure on the shelf; protects the size claim on the listing |
| 2 | Finished weight | Fastest proxy for fill density, and it catches under- and over-filling in one reading |
| 3 | Fill quantity per piece | Decides whether the piece holds its shape through packing and shipping |
| 4 | Stitch density / gauge | Decides whether the surface reads firm or loose, which is what buyers call "quality" before they can explain why |
| 5 | Shade against a physical standard | The one attribute a photograph cannot settle, and the one that most often triggers a claim |
Two supporting checks belong beside them but are not tolerances in the same sense: seam and attachment integrity, which for soft toys is a pass-or-fail gate rather than a band, and surface finish, meaning the absence of loose ends, visible joins and stray fibre. Those are binary. Either the piece survives the pull test or it does not.
Why hand-crochet drifts more than machine knit
A machine holds tension to a setting. A person holds tension to a habit, and habits shift at the margins — with the hour of the day, with the humidity, with a change of yarn batch running through the fingers, and with who is holding the hook. With a network of 30,000+ artisan weavers across four provinces in China, that human variable is not a flaw to be engineered out; it is the reason the craft exists at all.
What a serious maker does about it is not to demand machine consistency from human hands. It is to reduce the number of variables deliberately: one gauge swatch per SKU signed off before production, the same yarn lot reserved for a single order wherever the volume allows it, and a named artisan or team kept on the same SKU across reorders so that the second run is made by hands that already know the pattern. Where a reorder is placed with a different team, the tolerance band is what absorbs the difference — and the band only absorbs it if it was written down first.
Writing tolerance into the purchase order
The purchase order is the only document both sides will still be able to find in six months. Six sentences are enough, and each one closes a specific argument.
- Name the reference. "Approved sample dated 12 March 2026, signed by both parties, one retained by each." Not "as per sample".
- State the measurements of that reference. Height, width, and any dimension the customer will notice, in millimetres or centimetres, on the actual signed piece.
- State the band. A number, not an adjective. The band should be wide enough that a well-made piece passes and tight enough that a badly-made one does not.
- State the shade condition. The light, the distance, and the fact that judgement is against the physical standard rather than a screen.
- State the consequence. What happens to a piece outside the band: rework, replacement, or acceptance at an adjusted price. Silence here is what turns a tolerance into a dispute.
- State the reorder expectation. Whether the next run is expected to match the same sample, and what happens if the yarn lot cannot be repeated.
A tolerance clause is not a way of accepting worse goods. It is a way of defining what "the same" means before anyone has a reason to disagree about it. Buyers who write one tend to receive fewer surprises, not lower standards.
Inspecting bulk against the approved standard
Inspection is where a written tolerance pays for itself, because it turns a judgement call into a measurement anyone can repeat. The order of checks matters: cheapest and most decisive first.
- Weight, on a sample drawn across cartons. Not the top layer. A carton that is correct on top and light underneath is the failure mode weighing is designed to catch.
- Dimensions, on the same drawn sample. Measure the piece the customer will measure, in the state it will be sold — filled, dressed, unblocked if it will be sold unblocked.
- Shade, beside the approved standard. Daylight and warm interior light, side by side, at normal reading distance. A shade that only differs under a phone flash is not a shade problem.
- Seam and attachment integrity. For soft toys this is the pull test; a maker running a 90N pull-force check in-house is testing above the EN71 requirement, and the report is worth asking for. Our own TÜV Rheinland audit and 90N pull-force routine exists for exactly this reason.
- Surface finish. Loose ends, visible joins, stray fibre, embroidery alignment — the attributes that decide whether a piece looks like a sample or like a second.
Sampling depth is a separate decision from tolerance and should be agreed at the same time. The statistical side of it — how many pieces from how many cartons, and what acceptance number applies — is covered in our note on where soft toys actually fail mechanical testing, which is the attribute most often missed when inspection is limited to appearance.
When drift is a defect, and when it is normal variation
The distinction that resolves most claims is not how far one piece sits from the sample. It is whether the variation is controlled.
| Pattern | Reading | Reasonable response |
|---|---|---|
| Every piece slightly generous, consistently | Tolerance sizing issue, not a quality failure | Accept the run, correct the band or the gauge for next time |
| Most pieces correct, a few visibly different | Control failure inside the run | Sort and replace the outliers; investigate which team or lot produced them |
| Whole run correct on size, wrong on fill | Specification gap — fill was never written down | Rework the fill, then add fill weight to the PO |
| Shade shifts against the physical standard | Material or approval failure | Re-dye or re-knit depending on stage; dip approval discipline is the fix |
Read that way, a reorder becomes predictable. The maker's job is to hold the process steady and to say honestly which variables are human; the buyer's job is to decide, in numbers, how much human variation the retail promise can absorb. Both halves are needed, and only one of them can be written by the buyer alone.
Working on a reorder that no longer matches its sample? Send the approved sample's measurements and a photo of what arrived. We will tell you which of the five tolerances was never defined, and what to write into the next order. Read the sourcing FAQ or write to bianhaitao@kivastitch.com.
Frequently asked questions
What does spec-to-bulk tolerance mean?
It is the written allowance between the approved sample and the production run. Almost no handmade product is identical to its sample, so a buyer and a maker have to agree in advance on how far a bulk piece may sit from the sample on each measurable attribute — size, weight, fill, stitch density, shade — and on how a piece that falls outside that band is handled. Without the allowance written down, every difference becomes a negotiation after the goods are already made, which is the most expensive possible moment to have it.
Why do hand-crochet reorders drift more than machine-knit ones?
Because the variable in hand-crochet is a person, and a person is consistent but not identical. Tension in the hand changes slightly with fatigue, with humidity, with the batch of yarn running through the fingers, and with who is holding the hook. A machine holds tension to a setting. So the same pattern and the same yarn can produce pieces that differ by a few millimetres in a limb, or by a few grams in a finished toy, and both are correctly made. This is not a defect; it is the nature of the craft, and the tolerance has to be sized for it rather than argued with.
Which tolerances actually matter on a crochet order?
Five, in practice. Overall dimensions of the finished piece. Finished weight, which is the cheapest proxy for fill density. Fill quantity in toys, because it decides whether the piece stands or slumps. Stitch density or gauge, because it decides whether the surface reads as firm or loose. And shade, judged against a physical approved standard rather than a photograph. Buyers who write these five down, with a number attached to each, spend their inspection time on real problems instead of relitigating the definition of 'the same'.
How much size variation is normal between pieces?
It depends on the piece and on the stitch, so any single number quoted as a universal answer should be treated with suspicion. The workable approach is to measure the approved sample across the dimensions that matter to the customer — height, width, limb length where relevant — and to set a band that is tight enough to protect the retail promise and wide enough that a well-made piece is not rejected for being hand-made. The band should be agreed in writing and applied evenly to the whole run, not adjusted piece by piece until the numbers look good.
Is finished weight a better check than dimensions?
For filled toys it often is, because weight catches two failures at once: under-filling, which makes a piece feel cheap and collapse in transit, and over-filling, which distorts the shape and stresses the seams. Dimensions can be right while the fill is wrong, and the piece will still fail in the customer's hands. Weighing is also fast and needs no skill, which makes it a practical first-line check on a production line rather than an inspection ritual performed once at the end.
How should tolerance be written into a purchase order?
Name the approved sample by reference and date. State the measured dimensions of that sample. State the permitted band for each. State the lighting and distance under which shade will be judged. State what happens to a piece outside the band — rework, replacement, or acceptance at an adjusted price. And state whether the next order is expected to match the same sample. That is six short sentences, and it converts a future argument into a procedure that both sides can point at.
What should a buyer inspect first on arrival?
Weight and fill before anything else, because they are the cheapest signals and they fail earliest. Then the dimensions that the customer will actually notice, measured on a sample pulled from across the cartons rather than from the top layer. Then shade, side by side with the approved standard, in the light the goods will be sold under. Seam and attachment integrity comes next — for soft toys that means the pull test, and a maker who is serious about it will already have run one before shipping.
When is drift a defect rather than normal variation?
When it crosses a limit that matters to the end user, or when it is inconsistent within the same shipment. A run where every piece is a little generous is a tolerance issue and can be accepted or reworked as a batch. A run where most pieces are correct and a handful are visibly different is a quality control failure, because the buyer cannot sell a mixed carton. The distinction is not how far a single piece sits from the sample; it is whether the variation is controlled.