No black boxes.
Every price, index, and percentage on this site is computed from data we collect ourselves, with rules simple enough to print on one page. This is that page.
Where the data comes from.
Our collectors visit the world's major watch marketplaces, specialist pre-owned dealers, and certified pre-owned programs every night, and record each listing's price, specifications, condition, box and papers status, and images. We identify sources by category rather than by name, and every listing on this site links back to the original.
We record asking prices. Realized auction prices are shown separately where we have them and are never blended into dealer averages.
112,683
Listings
78,280
Models
1,783
Brands
117,949
Monthly data points
Cleaning, every night.
Scraped data is messy: currency glitches, typos, a watch listed at ten thousand times its value. Before any number reaches a chart, a nightly hygiene pass runs three rules:
Sanity band
Prices below $50 or above $5,000,000 are quarantined. They stay in the archive but never enter an average.
Outlier quarantine
Within each model, any price more than 10x above or 10x below the model's average is quarantined. The pass repeats until no outliers remain.
Deduplication
Each listing URL can exist only once in our database, and exact duplicate records are removed nightly.
Fair value: percentiles, not averages.
The price guide on every model page is built from the full distribution of that model's asking prices. Averages get dragged around by one weird listing; percentiles don't.
The confidence badge is earned, not decorative: high needs 30 or more listings with the middle half of prices spread less than 50% around the median; medium needs 10 or more listings spread less than 90%; anything thinner is labeled low and you should treat it accordingly. A listing is flagged "below market" or "above market" when it sits more than 8% from the median.
The market index: same watch, month over month.
Comparing this month's average price to last month's is a classic trap: if cheaper watches enter the market, the average falls even when every individual watch got more expensive. Our index only ever compares the same reference to itself across consecutive months, then chains the median of those ratios together.
Median, not average. Between March and April 2026 the average ratio across our constituents was +13.4% while the median was −0.3%. Seven of 100 references had moved between +132% and +357%, almost certainly because the mix of listings behind those references changed rather than because any watch reprised. An average lets a handful of those drag the whole market reading; a median reports what the typical tracked watch did.
The published parameters. Every one of these is a named constant in the builder, so the number on the chart can be reproduced from the same inputs.
| Base value | 1,000 points at the first month of the series |
| Constituents | The 100 most-listed eligible references |
| Weighting | Equal. We publish asking prices, not traded volumes, so weighting by value traded would claim more than the data supports |
| Minimum listings | A reference counts for a month only if that month rests on 3 or more distinct listings |
| Stability filter | A reference is excluded if its monthly listing count swings more than 4x, or its price range spans more than 6x. Both are signatures of a catch-all bucket rather than one watch |
| Step coverage | A month only moves the index when at least 60% of constituents appear in both months. Otherwise the level is carried forward and the point is marked provisional |
| Rebalance | Every 6 months. Constituents are never re-picked mid-period, or the index would drift with our coverage instead of with price |
| Series start | March 2026 |
Why the series starts in March 2026. Our price history holds earlier rows, but they are seed data rather than observation: before March the listing count behind each reference is frozen, unchanged month after month. Only 28 references vary across those months against 3,921 afterwards. Charting them would produce a longer line that means nothing, so the index starts short and real instead of long and invented.
What we won't pretend.
Asking is not selling. Most of our data is dealer asking prices. Where we show sold prices, they are labeled as such. Coverage is uneven. A Submariner has hundreds of data points; an obscure independent may have four, and the confidence badge will say so. Archives grow in bursts. When we add a new source, thousands of listings arrive at once; our time series are built to ignore these ingestion spikes rather than report them as market moves.