Method, sources and limits
What we do, how we check it, and - the part most directories omit - what we do not know.
Where the data comes from
Facts come only from primary sources: the teacher's own site, the venue's own site, the host studio's site, or the organiser's own booking page. We read aggregator listings only to learn that an event exists, and we store nothing from them. If a primary source does not confirm a fact, the fact does not appear here. That rule costs us coverage. It is also why a figure on this site survives being checked against the organiser's own page.
Prices
In this tier the teacher announces the event and the venue frequently owns the room rate, so a large share of genuine, well-run programs publish no price at all. We record that as a state rather than a blank:
| Price disclosure | Instances | Share |
|---|---|---|
| Published | 584 | 53% |
| Partial price only | 256 | 23% |
| Price not published | 185 | 17% |
| Price on request | 42 | 4% |
| Price set by the venue | 34 | 3% |
Only published renders a number. Every other state renders its own label. We do not estimate, average, or convert currencies, and a price is never shown without its basis - per person, per unit, or per night.
Practice hours
The daily practice figure is computed from the organiser's published timetable, not copied from their marketing. Overlapping blocks are counted once. Meals, philosophy, lectures and free time are excluded. Where any practice block has no published times, we publish no figure at all rather than an undercount - which is why many programs here carry no hours figure.
Coordinates are derived, not published
No venue in this index publishes its own coordinates. Every pin we hold was worked out from an address, a gazetteer or a map link, so each one states its precision. 346 of 385 venues have a coordinate: 185 placed at the property, 133 at the settlement it sits in, and 28 only in the surrounding area. Confidence is a separate axis from precision - a pin placed at a property can still be a low-confidence guess - so, cutting across all three, 117 of the 346 are low-confidence and marked as indicative. A low-confidence pin does not count toward a listing's completeness score, whatever its precision, because we are not confident in it and the score should not claim we are.
What we hold back
100 of the 588 programmes we have ever collected are currently quarantined: a quality check found a problem serious enough that publishing the listing would cost more trust than it is worth. A quarantined programme is not deleted and not counted as something we never found - it is held until the problem is fixed or the finding is confirmed wrong, and it counts here so the gap between what we have collected and what we publish is never silent.
How often we are wrong
We draw a random sample of published dates and re-check every field against the page we cite. Not "is the field filled in" - is it right. The first audit was on 2 August 2026 and returned 64.6%. We repaired what it found and audited again the next day, on a fresh sample.
| Field | 2 Aug | 3 Aug |
|---|---|---|
| Venue | 87% | 90% |
| Dates | 70% | 82% |
| Price | 62% | 82% |
| Price disclosure state | 70% | 82% |
| Teacher | 72% | 80% |
| Status (open / sold out) | 25% | 45% |
| Overall confirmed correct | 64.6% | 77.1% |
Clean on all six fields went from 3 of 40 to 15 of 40. But 16 of 40 records still carry at least one demonstrable error, which is exactly what it was: the average record improved and the share of defective records did not. We publish both halves of that.
Do not book off this index without opening the source link. For a single-venue retreat with one named teacher, our auditor now says the data is good enough to act on. For a modular teacher training, it is not - that is where the surviving errors cluster.
The error class that improved most: prices. 160 figures were recovered from pages we had recorded as publishing none, and "partial price only" fell from 82% misuse to 38%. Records with no citation at all went from 10% to 0.4%.
The class that survives is modular teacher trainings: a course running one weekend a month for nine months, recorded as the weekend we happened to read. If a training here shows a short span, treat the dates as a fragment and open the organiser's page.
Status is our least reliable field - only a quarter could be confirmed at all, because most pages simply do not say whether a place is still open. Read it as "what the page said when we fetched it", never as live availability.
An unverifiable field is not a pass. It means a reader who follows our citation cannot confirm what we told them, and we count it against ourselves.
One claim we had to withdraw while writing this: we said every fact links to a page we fetched. That is true of every programme, and true of only about a fifth of individual dates - the rest cite the page that announced the programme rather than the page that announced that date. The distinction is small in wording and large in practice, so the wording changed.
What we currently do not know
- 39 of 385 venues still have no coordinate at all.
- Several venues render prices in JavaScript our reader cannot execute, or behind a login. Those programs show "price not published" even though a price is visible to a human. That is a limit of our fetcher, recorded as such.
- Some sites block automated reading entirely. Those organisers are absent from this index, not judged by it.
- Mean completeness across the index is 0.781 of a possible 1.000. We publish the number rather than round it up.
Corrections
Every listing is editorially collected and unclaimed. Corrections are free, apply immediately, and never require payment. Correcting your data does not change your ranking, in either direction. See claim a listing.