The Winding Index — Methodology & Data
This page explains exactly how the Winding Index was built — where the data came from, how it was audited, what was thrown out and why. We publish the method alongside the findings because a statistic you cannot inspect is just an opinion, and because we would rather show our working than ask to be trusted.
What the dataset is
The Index is compiled from the Aevitas TPD database: 4,918 records covering 106 watch brands, each recording a model's turns per day (TPD) and winding direction. The database has two strata. The larger is our long-standing store database of manufacturer-published settings, built up model by model over years of selling and setting watch winders. The smaller is a research stratum added in October 2026: records with full reference numbers and movement names, each checked against published sources, with the source recorded on the record.
The audit: from 4,918 records to 4,302 published
Before anything was published, every record was passed through an audit (October 2026). The outcome:
| Outcome | Records | What it means |
|---|---|---|
| Published in the Index | 4,302 | Records that survived every check below |
| Quarantined | 473 | Held out of all published statistics — see below |
| Excluded as unverified | 117 | Research records whose source could not be confirmed to publication standard |
| Removed as superseded duplicates | 23 | Duplicate records of the same watch, resolved to a single canonical record |
| Deleted as non-automatic | 3 | Manual-wind and quartz models that do not belong in a winding dataset |
Conflicts. Eight groups of records described the same watch at different settings. Each was resolved on the evidence — the record with the traceable source wins over a later unsourced entry — and in every case the resolution, and the reason for it, is recorded on the record itself. One pattern recurred: a batch of newer entries had recorded 725 TPD, the midpoint of the commonly published 650–800 range, for watches whose sourced records said otherwise. Those batch entries lost, every time.
The quarantine, in plain language. Three brands — Graham, Breguet and Girard-Perregaux — show suspiciously uniform settings across large parts of their ranges in our bulk data. Cross-checking showed the bulk records faithfully reproduce a long-published third-party settings database, which means those brands' figures ultimately rest on a single upstream source whose defaults we cannot independently confirm. One source, however long it has been copied, is not verification. So those 473 records are quarantined: they appear in no statistic, table or brand page in the Index, and they will stay out until each brand is checked against a second, independent source. We would rather publish less data than publish data we cannot stand behind.
Piaget was checked — and cleared. Piaget's figures also looked like a cluster (a median of 1,080 TPD is far above every other brand), so the audit spot-checked the brand record by record against an independent publication. All 71 records reproduced exactly, and the values are structured by calibre — the high settings belong to specific ultra-thin movements. Piaget's median of 1,080 TPD is therefore published as a verified finding.
Conventions you should know
Ranges and midpoints. Many sources publish a range rather than a single figure — 650–800 TPD is the commonest. A database needs one working number per watch, so a range is recorded as a single figure: the source's own default where one is given, otherwise the midpoint of the range. During the audit, 321 records were found to have been converted to midpoints in an earlier correction pass; a sample of 30 was re-checked against the published ranges and every one reproduced the convention exactly, so the pass stands — and the convention is disclosed here rather than hidden.
Direction labels. Sources spell the counter-clockwise direction two ways; the dataset normalises to one. Direction figures are stated as bidirectional, clockwise only or counter-clockwise only, as a share of the models counted.
Every statistic carries its n. Each figure in the Index states the number of models behind it. Brand statistics are computed only from that brand's published records; brands with too few records do not get statistics at all.
Known limitations
Coverage is uneven. The dataset grew brand by brand, so coverage is deep for some houses (IWC 274 published models, Omega 236) and thin for others. Rolex is the clearest gap: 25 audited records at the time of the audit. Researched additions are being added from published per-model sources, and the Rolex Index page states exactly how many models its figures rest on, including which additions are high-confidence.
Movement-level claims are not published. Only 307 of the 4,918 records name the movement. That is too thin a base for movement statistics, so the Index makes none — where two sources give the same movement different settings (the 650-vs-725 splits seen in some calibre families), the model record follows the better source and the disagreement is a known feature of this data, not a hidden one.
A compiled database is not a manufacturer certification. The Index reflects the best published settings we can source and check. Makers occasionally revise guidance; where a manufacturer corrects a setting, we update the record and note the change.
Citing the Index
The findings are free to cite — in articles, guides, videos and research — with a link back to the Winding Index. If you need the underlying figures for a specific brand or model, the TPD database is openly searchable, and questions about the data are welcome via our contact page.
October 2026 edition. The dataset is refreshed quarterly; annual editions of the report accrete on the Index page.
