Where our market data comes from

What we collect

Live asking prices for tracked watch references, collected automatically every hour from public eBay listings via eBay's official API. For each listing we capture the ask, condition, listing format, seller type, item location, how long it has been live, and detected variant attributes (dial, bezel, bracelet).

Collection is prioritized by a popularity model built from what dealers actually list and look up in Vericog — the most-traded references refresh every few hours, the long tail at least daily. Today eBay is our sole automated source; additional marketplaces are on the integration roadmap and will be listed here when they're live, not before.

How we clean it

Raw listings are noisy: parts listings, franken-watches, typo prices, duplicates. Each collected listing is deduplicated, run through interquartile-range outlier detection against its reference cohort, and assigned a per-listing confidence score. Detected variant attributes — dial, bezel, bracelet — are recorded with each listing so per-watch pricing comparisons don't confuse a green-dial with a black-dial; reference-level snapshots aggregate across variants. Outliers are kept but excluded from computed statistics, and raw observations are retained for 90 days.

What we compute

Cleaned listings aggregate nightly into per-reference daily snapshots: lowest ask, median ask, upper-quartile ask, live listing count, active dealer count, and the age distribution of live listings. From at least seven daily snapshots we derive 30- and 90-day trend signals, ask-spread compression, undercut frequency, and a market activity score.

The dealer pricing engine turns those snapshots into a maximum-buy figure, a recommended ask, and a risk flag — with condition and completeness adjustments. No signal is emitted for a reference until it clears the seven-snapshot minimum: if we haven't watched a market for a week, we don't pretend to understand it.

Ask-side by design (for now)

Our data is built from asking prices and listing behavior, not confirmed sold prices. eBay retired public access to sold-item history in 2025, and we won't synthesize numbers we can't observe. What asks give us is still substantial: where sellers cluster, how deep live supply is, how long inventory sits, and how aggressively it gets undercut — the median age of live listings is a direct read on how fast a reference actually turns.

This is also why we say "market activity" rather than "liquidity" in our public signals: true liquidity needs sell-through, and sold-price integration is on our roadmap. When it ships, this page will describe its source and method first.

Inferred sales

We track collected listings through their lifecycle: each collection pass refreshes when a listing was last seen live and records ask changes. When a listing disappears — and only when a later scan of that reference captured its complete live market, so absence is meaningful — we record an inferred sale at its last observed ask, with the listing's lifetime as inferred days-to-sell and a confidence score based on how the ask sat against the reference's median. Auction expiries, sub-day removals, and outlier asks are excluded, and references with very deep supply can't be fully scanned at our current API quota — so inferred sales skew toward references with moderate live supply, and we say so. Inferred sales are always labeled as such and are never blended with confirmed sold data — which we still don't claim to have.

Why public figures are banded

Public reference pages show qualitative bands — activity level, supply band, listing-age band — rather than exact dollar figures. Two reasons, honestly stated: exact reference-level pricing is the product our dealers pay for, and bands can't be mistaken for appraisals. Inside Vericog, dealers see the full-resolution numbers with their confidence scores attached.

Freshness

Collection runs hourly; snapshot aggregation and derived scores recompute nightly. Signals carry their computation timestamp internally, and any public signal whose underlying data goes stale is removed automatically rather than left to rot — a reference that stops meeting our data minimums loses its signals until coverage recovers.

What we don't do

We don't aggregate auction results, we don't have confirmed sold prices yet, we don't forecast prices to a dollar, and none of our output is an appraisal or investment advice. Where a Vericog screen shows illustrative or preview data — like the terminal mock-up on our market intelligence page — it is labeled as such.