Corrections and Updates

Updated 11 August 2026

A site that quietly edits its mistakes cannot be checked. This page is the record.

What counts as a correction

The line is whether a reader could have acted differently because of the error. If yes, it gets logged with what changed and why.

Correction log

2026-08-11 — Sample unit corrected on the short-cages guide

What was wrong. The page reported length figures against "247 listings." That number counted specification rows, not products. A product selling several variants was counted several times.

Corrected to. 115 products. Because the unit changed, the derived figures changed with it: median published length from 60 mm to 51 mm, and the share at 65 mm and over from 47% to 34%. (Source: the corrected per-product query in data/derived/queries/chastity_size_distribution.sql.)

Why it happened. The query aggregated rows rather than grouping by product. The stored query has been rewritten to aggregate per product, and now reproduces the published figures exactly.

Also corrected on the same page. The title previously promised a comparison between "small" and "tiny." Only 6 of the 115 products use either word, so no such comparison is possible. The page now reports that finding instead of implying a comparison it cannot make.

2026-08-11 — Variant-level precision claim removed

What was wrong. The size guide stated that no listing published both measurements "in a form we could tie to a single variant." Our audit is product-level; we did not verify variant-level binding at all.

Corrected to. A statement that no product published both as clearly labelled fields, with an explicit note that this is a product-level audit and that any tool claiming variant-level accuracy would be asserting precision the source data does not contain.

2026-08-12 — A waiting condition we invented on a safety page

What was wrong. Our safety page listed signs that warrant medical care under the heading:

"Seek medical assessment if any of these persist."

The patient guidance we cite sets no such condition — it lists severe genital pain, marked redness or swelling or bruising, and pain when urinating as reasons to see a doctor, full stop. The same guidance separately describes blood at the tip of the penis or in urine as a sign of a serious injury; we had folded it into the same wait-and-see list.

Why it mattered. The word "persist" was ours, not the source's. It told someone whose symptoms eased — or whose bleeding happened once and stopped — that they could wait. That is the opposite of what the guidance says, and on this particular page, being wrong in that direction delays care.

Corrected to. The signs are now listed with an explicit instruction not to wait to see whether they settle, and blood is stated separately as a serious-injury sign to be treated as such even if it occurs once.

Also corrected in the same review. Three further changes to the same page: the emergency-removal tools named in case reports were removed from patient-facing text (they are supervised hospital procedures and served no decision-making purpose for a reader); the "48–72 hours" figure was removed as a countdown anchor, because a number that prominent outweighs the sentence denying it is a threshold; and an "if you are not sure" path was added, defaulting to the stuck route rather than leaving it unassigned.

How it was found. Independent editorial review, followed by re-reading the cited source directly. The error had passed all ten of our automated gates — those check whether a claim is registered and sourced, not whether our wording matches what the source actually says.

2026-08-12 — Two overlapping groups presented as a clean split

What was wrong. A draft of the fitting guide reported that 66 listings publish ring size as a selectable option and 187 publish it only in description text, laid out as two rows of a table. That reads as a partition of the same 253 products. It is not: 34 products appear in both groups, because they have a size list and mention sizes in the description. The rows double-counted those 34.

Corrected to. A three-row breakdown that is genuinely exclusive — 32 option-only, 34 both, 187 description-only — with an explicit note that these describe where we parsed the number from, not what the retailer's page offers a buyer.

Why it mattered. The error inflated the apparent contrast between "sizes you can choose" and "sizes you can only read about." That contrast is the point of the table, so overstating it flattered our own finding.

How it was found. Independent editorial review questioned whether the two groups were mutually exclusive. They were not, and a single query settled it. The page was corrected before publication; no reader saw the split version.

2026-08-12 — We withdrew every disclosure percentage on this site

What was wrong. Every percentage we published about retailer disclosure divided by the same thing: how many chastity devices we captured. We could not defend that number.

Three reasonable ways of deciding what counts as a chastity device gave us 629, 448 and 273 records — a spread of more than two to one. Our automated classification put a coffee table, a lace teddy, a puppy cage and several key safes in the device set, and threw out real cock cages because their titles happened to contain "case" or "cuff." Both error directions, in the same rule.

What was withdrawn. Every share and rate: ring-diameter disclosure, cage-length disclosure, both-fields disclosure, the length distribution bands, the material-naming counts, and the median figures that depended on them. They are marked as withdrawn in our figure ledger, and an automated check now refuses to publish anything so marked.

A correction to this correction (12 August 2026). The sentence above was true of our content files and not of our page components. The measuring guide kept publishing two of the withdrawn disclosure rates, because the check only ever read article text and that table was written directly into the page. The table has been removed, and the check now reads the built pages themselves rather than the sources they came from. We are recording this here because a site whose main claim is traceability does not get to fix that one quietly.

What survives, and why. One product-level finding: no product we captured publishes both a ring diameter and a cage length as clearly labelled fields. That count is zero under all three denominators. When a count is zero, the choice of denominator cannot change it — which is exactly why this is the finding we build on and the percentages were not.

A second error inside the first. We had also reported "zero structured material fields" as though it described retailers. It described us: our extractor does not implement material extraction, so finding no rows in a field we never populate says nothing about what retailers disclose. Withdrawn on the same day.

What we are not doing. We are not re-deriving a better percentage and publishing it quickly. Settling this needs a human-graded sample, and until that exists a number from us would be false precision dressed as rigour.

How it was found. Independent editorial review rejected our first correction as insufficient and named specific misclassified products. They were right, and checking took one query. The full classification rule and its tests are in engine/classify_products.py.

2026-08-12 — A median that did not match our own ledger

What was wrong. The short-cage guide reported a median published ring diameter of 45 mm. Our own figure is 40 mm, and recomputing it produces 40 mm under both the original and the corrected denominator. The 45 was simply wrong and could not be reproduced from the data it cited.

Corrected to. 40 mm, from 216 devices with a readable ring diameter.

Why it went unnoticed. Our number-traceability check verified that each published figure pointed at a registered entry — but it never compared the number printed on the page with the number in the entry. It reported "every number is traceable" while the two disagreed. The check now compares values, and a deliberately mismatched figure was used to confirm it fails.

2026-08-12 — Our headline sample was not what we said it was

What was wrong. Every page on this site described our audit as covering "629 chastity device listings." It did not. The 629 figure was everything we captured from eight retail sources, and it included items that are not chastity devices at all: bust harnesses, a waist cincher, garter and hobble belts, a butt plug, spare rings, and a pair of gloves. Counting devices only, the sample is 448.

Corrected to. Every figure has been recomputed against the 448 devices. The disclosure rates all moved:

Why it mattered, and which way it cut. The wrong denominator made retailers look worse than they are — every disclosure rate was understated by several points. That is the direction we are least likely to catch on our own, because the error flattered our own finding. We are stating it plainly for that reason.

What did not change. The central finding is unaffected: not one device publishes both a ring diameter and a cage length as clearly labelled fields we could verify. That count is zero under either denominator.

How it was found. Independent editorial review challenged whether the 629 were all devices. They were not. The classification is now a separate script with its own must-fail tests (engine/classify_products.py), because a denominator that flatters us needs a check that is not us. 63 captured items remain unclassified and are excluded from every denominator rather than assumed to be devices.

Reporting an error

We want error reports, including ones that make us look bad.

Contact, honestly stated. This site is pre-launch: the domain is not yet registered, so no address on it can receive mail. Publishing one would be a contact route that silently fails. A working reporting channel will appear here before launch, and this note will be replaced rather than deleted.

What we will not do

• We will not delete criticism that turns out to be fair.
• We will not backdate. A page corrected today says so today.
• We will not treat a rewrite as a correction record. If a figure changed, the old figure is stated here, not just replaced upstream.