Methodology

Every headline figure is the median asking price of live, verified listings for that exact reference in that market — never an average across markets, never a model-level blur, never a projection. One reference, one market, one median.

Each figure carries its listing count and its capture date, visibly. A market with three or more verified listings is marked solid; one or two is thin and disclosed as thin; zero claims no number at all. A single listing is presented as one listing’s price, dated — never as a market.

Observations outside 25–500% of the market’s own median are rejected as accessories or misparses. Native prices whose implied exchange rate deviates more than 40% from the currency’s file-wide median are discarded in favour of the cross-checked figure. Gulf currency pegs are verified within ±1%; any feed that moves a pair more than 12% in one refresh is rejected whole.

Prices are stored in their native currency, anchored in INR, and converted at display time — never baked at a stale rate. The conversion rate’s own source timestamp is shown beside every converted figure; past a 30-minute freshness threshold the board says so, visibly.

The board reflects live asking prices. Where a brand’s asking prices are known to run above actual clearing prices, the figure is labelled an asking signal and never presented as a transaction price.

A market is the seller’s jurisdiction — where the seller stands, established from the listing itself — never the currency a price happens to be quoted in and never the viewer’s location. A Hong Kong figure is what sellers in Hong Kong ask, whoever is looking at it, from wherever.

Price-on-request listings are counted but never priced — a withheld number is not a number, and no other field is ever substituted for it. Parts-market listings — dials, bezels, movements, bracelets sold alone — are rejected before aggregation; a bezel’s price inside a watch median would poison the figure. Rejections are internal telemetry and never surface as prices anywhere on the site.

Each (reference, market) sample states what it is. A census frame saw the whole live market. A truncated frame — capped collection that caught the low end only — publishes its floor price and listing count but never a median: a median of the cheapest slice would be fiction, so the pipeline nulls it and the site never recomputes one. An empty frame renders a page that claims nothing and asks not to be indexed.

Listings on the primary source platform quote in EUR as the platform base currency regardless of the seller’s jurisdiction — so the native capture is EUR, jurisdiction is read separately from the listing, and no currency is ever inferred from a market. Prices are then anchored in INR for storage and converted exactly once at display.

1107 references are currently tracked, 919 of them carrying verified prices in two or more markets. Attribute coverage (dial, case material, condition) is measured per run; a field below half coverage flags its references out of the index rather than publishing thin claims. Data refreshes as scrape runs complete — each figure’s own capture date is shown beside it, and pages republish within the hour of new data landing.

Asking prices, not transactions. Coverage follows where verified listings exist, so thin markets stay thin honestly rather than being modelled. Condition and accessory differences within a reference are disclosed as variant detail where captured, and cross-market comparisons deliberately stop short of like-for-like claims until tax treatment is captured per listing — a page will show two markets side by side and refuse to compute the difference.

No fabricated numbers, no rounded-up counts, no borrowed figures for unpriced markets, no seller identities, and no offers of our own — CALIBRÉ holds no inventory and marks up nothing. The structured data on every page mirrors exactly this: an aggregation of other parties’ listings, dated and counted.