How we check
A pattern passes 8 checks on its way to this site. 6 of them ran on all 113 patterns here and 2 could not run at all. Together they prove that the arithmetic works and that the page describes what the arithmetic makes. None of them proves the object is worth making, because nobody has made one.
The eight checks
In the order they run. Each pattern page shows the same list for itself.
Format and stitch chain
Proves that the pattern is workable as arithmetic.
Run by machine on every patternCatalogue audit
Proves that the catalogue entry agrees with the pattern.
Run by machine on every patternGeometry
Proves that the shape the counts describe is a shape.
Run by machine on every patternRender matches the pattern
Would prove that the picture beside a pattern shows the object its rounds make.
Could not runNot a duplicate
Proves that no two records share a displayed name, checked by machine across the whole library.
Run by machine on every patternRecognisability
Would prove that the features which make an object read as itself are each present — as a piece, as a run of colour, or as a finishing note.
Could not runTarget on the record
Proves that the phrase each page targets is on the record with what it is claiming, and that a search volume appears only where a measured cluster stands behind it.
Run by machine on every patternPage readiness
Proves that there is a page.
Run by machine on every pattern
What an author credit means here
An author credit here is not a claim that someone hand-designed every stitch. These patterns are drafted with AI help and then checked by machine, round against round; the credit names the person who drafted this one, ran those checks and decided it was fit to publish — and who to come back to when it turns out to be wrong. Nobody has crocheted them yet, so what is promised is the arithmetic, not the look in the hand.
Meet the authorsFound a round that will not add up?
Note the pattern, the piece and the round number while you have them in front of you. This site has no address taking reports at the moment; the report page says what to keep.
Reporting a roundWhat each one proves, and what it does not
The result under each one is the line the records themselves carry, written when the check was run on 2026-09-04 and read off the 113 files rather than restated here.
Format and stitch chain
What it proves. That the pattern is workable as arithmetic. Every round is pinned to the one before it: the instruction says what it consumes and what it produces, and the declared count has to be exactly what comes out. A round that asks for more stitches than the round below has is refused.
What it does not. It says nothing about what the counts describe. A workable chain can still make the wrong object.
Catalogue audit
What it proves. That the catalogue entry agrees with the pattern. The stated finished size is compared against the size the counts derive, every colour named in the palette has to be worked somewhere, and a name that promises a texture is checked against whether any note describes it.
What it does not. Its warnings are judgement calls, not failures. 66 of them stand across the library and each is a reading somebody has to do.
Geometry
What it proves. That the shape the counts describe is a shape. A disc worked from a ring of six has to gain six a round to stay flat; a piece that closes has to decrease far enough to actually close; a left and a right of a pair have to have the same number of rounds or they will not collapse together.
What it does not. Each of its three rules is guarded, and a rule that did not apply to a pattern did not pass on it: the closing test runs only where the pattern is a single piece, the flatness test only where the record calls itself flat, the pair test only where a left and a right are both named. The model behind all three is a model — it works in stitch widths derived from the hook, not from anybody’s hands.
Render matches the pattern
What it would prove. That the picture beside a pattern shows the object its rounds make: every piece the pattern lists present, no piece it does not list, the palette’s own colours, and no texture this format cannot produce — no ribbing, no openwork, no clusters.
What it does not. Even running, it would compare a picture against a written description of the pattern. It would not compare it against a finished object, because there is no finished object.
Not a duplicate
What it proves. That no two records share a displayed name, checked by machine across the whole library. The comparison is on the name and not on the stitch counts, because the opening 6, 12, 18, 24, 30, 36 appears in dozens of pieces and structural similarity would flag almost everything.
What it does not. A name check cannot tell two records of the same object apart under two names, and nothing has compared the subjects: whether the library holds one thing twice is a judgement somebody has to make, and no result of making it is on the record. Two patterns for genuinely different objects can also still overlap in what a reader wanted.
Recognisability
What it would prove. That the features which make an object read as itself are each present — as a piece, as a run of colour, or as a finishing note. A dragon with no snout, no horns, no wings and no tapered tail passes every arithmetic check above and is a lumpy bear.
What it does not. It checks the pattern against a written list of features, not against whether anybody looking at the result would name the animal.
Target on the record
What it proves. That the phrase each page targets is on the record with what it is claiming, and that a search volume appears only where a measured cluster stands behind it. A page with nothing defensible to target records that instead of borrowing a fat term it does not answer.
What it does not. It checks the claim is consistent, not that the page will rank or that the phrase is the right one.
Page readiness
What it proves. That there is a page. A title, a description, a lead paragraph, image alt text and prose for every piece, all present and all inside their bounds — and, in the other direction, that nothing claims more than it earned: a numeric gauge nobody swatched, a derived measurement with its marker stripped off, a durability claim about an object nobody has used.
What it does not. A field can be present, correct in length and still be dull or wrong. 114 editorial flags stand open across the library and each is a judgement rather than a defect.
The numbering has gaps, and they are worth naming. The gate has two further stages — whether the phrase a page targets is one the pattern can honestly answer, and whether the pattern has an image, alt text and a shelf to live on — that no script writes a result for. Nothing on any of the 113 records carries an outcome for either, so neither is in the list above and neither is counted in the 8. A check with no recorded result is not a check that ran.
The two that could not run
Their own words are in the list above. What follows from them is not, and it is the part worth carrying away: both are blocked on something missing rather than on something that failed, and between them they leave two specific things unchecked.
- Nothing has looked at the pictures. A picture on this site showing a piece the rounds do not make would not be caught by anything here, and nothing has counted how many do. If one looks wrong to you, that is a fault the checks cannot see, and it is worth reporting for the same reason a round that will not add up is.
- Nothing has checked the likeness. The arithmetic cannot tell a fox from a bear. A pattern can chain perfectly, close properly, sit inside its stated size and still be missing the one feature that makes the animal that animal — and until the feature lists exist, the only thing standing between a pattern and that failure is somebody reading it.
What no check here can establish
The boundary is the point of this page, so it is stated as a list rather than softened into a paragraph. Nothing above establishes:
- That the finished object looks good. The checks read counts. A shape can be arithmetically sound and still be an ugly thing.
- That the fabric behaves. Whether it drapes, whether stuffing shows between the stitches, whether a wall stands up: all of that is fibre, hook and tension, and none of it is in the data.
- That assembly is pleasant, or possible. Sewing eight pieces to a body is where a pattern usually stops being enjoyable, and no check has an opinion on it.
- That the instructions read well. Prose is checked for presence and for claims it may not make, not for whether it is any good to follow at ten at night.
- That anybody enjoyed making it. Nobody has made one. Every word on this site is drafted from the counts.
The one thing crossing that boundary is somebody with the pattern on a hook. A single finished object is worth more than its own correctness, because it tests everything that drafted it.
Why there is no badge
The gate can award a tier, and it has awarded none. verification.tier is empty on all 113 patterns, and it is empty because a tier is earned by every stage passing — and 2 of them could not be run at all. A badge printed over that would be a badge for work nobody did.
There is no state called tested, because nobody tested anything. The first tier that could exist above this one is earned by a maker who finishes one and shows it, which is why the pattern pages ask. Until then a page shows what ran and what did not, which is less impressive than a seal and is the thing that happens to be true.
What is not recorded yet
The records have room for four more facts about how a pattern was made, and they are empty. The reason to print an empty field rather than leave it out is that the gap is the useful information: the point of stamping a pattern is that when one defect turns out to be systematic, every pattern sharing the stamp can be found in one command. With these blank, that sweep is not available.
| Field | What it would hold | Today |
|---|---|---|
| provenance.drafterModel | Which model drafted this pattern. | empty |
| provenance.promptHash | Which prompt it was drafted from, so patterns sharing one can be swept together. | empty |
| provenance.checkRegistryVersion | Which version of the checks it was passed against. | empty |
| provenance.recognisability | The features that make the object read as itself — what stage 6a would check the pieces against. | empty |
No error rate until there is one to count
A figure for how often a pattern here turns out to be wrong would be the most useful number on the site, and there is not one. Such a figure would have to be built out of reports from people who worked a round and found it did not hold, and nothing has been counted, so nothing is printed. A number invented to fill that space would undo everything above it.
One thing worth knowing about a figure like that even once it exists: silence is not success. Most makers who hit a bad round assume the fault is theirs and stop, so a low report count and a good library look identical from here. That is why the controls that matter are the ones above and not a tally.
Reporting a round is where a report goes and what it needs. Where a pattern comes from before any of this runs is on how this works.