Sources and Methodology
"Evidence-informed" is easy to put in a tagline and hard to make checkable. This page is the checkable version: what we collected, how, and — most importantly — what each dataset cannot tell you.
Source hierarchy
We rank sources, and we say which tier a claim rests on.
| Tier | What it is | What we may write from it |
|---|---|---|
| Own data | Our own snapshots and queries | Descriptive statements about that snapshot only |
| Primary literature | Studies, case reports, official guidance | Accurate restatement of population, setting and limits |
| Retailer specifications | What a seller publishes | An audit of what is disclosed, not fit or safety evidence |
| Community discussion | Public forum threads | Questions people actually ask, not frequency as evidence |
| Editorial judgment | Our own frameworks and checklists | Labelled as our framework, not as a research finding |
A weaker source does not become stronger by being bolded, put in a component, or carrying our byline.
Dataset: retailer specification audit
Snapshot 2026-08-10 · Unit of analysis the product, not the variant row
We collected public product listings from open retailer sources and extracted two fields: ring diameter and cage length.
Cleaning rule. We drop values outside plausible ranges — 25–60 mm for ring diameter, 30–130 mm for length — because the out-of-range values are unit-conversion artefacts, not real products. One listing resolved to 1905 mm, which is 75 inches. Exclusion rates are published on the pages that use the data.
Why the denominator matters. A single product often sells several variants, so counting variant rows instead of products weights multi-variant listings several times over. We analyse at product level for that reason.
🔴 This is also where we failed. Getting the unit right is not the same as getting the population right. On 2026-08-12 we withdrew every disclosure percentage on this site because we could not defend which captured records count as chastity devices rather than accessories — three defensible rules gave answers differing by more than two to one. The product-level discipline described above was correct and insufficient. Full account in corrections.
What this dataset can support: statements about what retailers disclose, and how inconsistently.
What it cannot support: which device fits anyone, whether a size is comfortable or safe, manufacturing tolerance, or whether two retailers measure the same way. It is an audit of published text, not of physical objects.
Dataset: community discussion
We read public forum threads to learn which questions people actually ask and which distinctions confuse them. That is where topics come from.
Hard rules. We do not publish usernames. We do not quote posts verbatim. Where we describe patterns, we report aggregates only, and any illustrative scenario is written by us and labelled as a composite rather than presented as someone's real account.
🔴 What community data cannot do: frequency in a forum is not evidence of safety, efficacy, or prevalence. Forum posters are a self-selected group, and a thread count is not a measurement of the world.
Medical and safety literature
The published literature on these devices is concentrated in emergency case reports of entrapment and strangulation injuries.
Those reports describe harm that has already occurred. They are informative about failure modes and useless as evidence that ordinary long-term wear is safe. We did not find studies covering routine long-term use, and we say so on every page where the question arises rather than filling the gap.
How a number gets published
Every figure is registered before it appears in prose, with its query, sample size, snapshot date, method and limitations. A figure whose stored query does not reproduce it is not published — this is enforced by an automated check that fails the build, not by anyone remembering.
Percentages appear with their absolute number and denominator. A percentage without its sample is not information.
Versions
Datasets carry snapshot dates. When we refresh one, the figures on affected pages change together with the snapshot date, and material changes are recorded in corrections.