unghosted

How unghosted.io measures Rolex waitlist times

This page describes how we collect, normalize, and present Rolex waitlist data on unghosted.io. The same principles apply across brands where noted.

Sources

Wait time data on unghosted.io comes from two places:

  1. Direct submissions through our submit form. No email required, anonymous by default.
  2. Public buyer accounts posted on watch forums and community sites, compiled into structured reports.

Most of the dataset was compiled from public buyer accounts; a smaller share comes from direct form submissions. Every entry has a source field indicating where it came from. Source links are recorded for reports captured through the review pipeline from September 2026, and for the 2024 megathread import. They are not available for earlier imports.

How we count

Every report represents one person's experience with one watch from one authorized dealer. We don't average across multiple watches per buyer. We don't synthesize numbers from market commentary or buying guides. Editorial content from third-party sources is excluded from waitlist statistics.

Purchase history

Purchase history is the buyer's prior purchase relationship with the dealer who made this sale, not the size of their collection. A first Rolex from a dealer where the buyer already bought other watches or jewelry at that store is not “no prior purchases.” A large collection bought elsewhere, with nothing at this dealer, is coded as no prior purchases here.

Rows coded Unknown stay in all-time coverage. They are excluded from published purchase-history splits (first-time vs prior-purchase) and from the countable history tiers those splits use.

Usable reports (the public denominator)

Homepage totals, BrandStatBlock counts, wait-pole percentages, and quick-share charts use the same denominator. A published report counts when its wait_time is a known completed or open-list bucket. We discard:

Still-waiting reports stay in that denominator so an open list does not silently inflate completed-wait rates. Legacy labels outside the catalog are dropped from both poles and quick-share, not remapped. Stored “Under 1 month” is a known coarse label: it counts in that denominator and in the within-three-months share, on the same rung as 2-4 weeks.

That usable N is not the same as every published row, and it is not the 12-month median sample. Title tags and meta descriptions no longer embed a live count. When a number is quoted, it is one of three:

Under-1-month poles count walk-in through about four weeks. Over-a-year poles count 1-2 years and longer. Quick-share asks a related question (share receiving within about three months) against the same N. When a percentage comes from fewer than 50 usable reports, we render it with the raw count (for example, “6% (2 of 36)”) so small samples cannot look precise. We do not put a percentage in a title tag below 40 reports, the same floor as reference-level charts.

Brand-wide medians are a poor summary for this dataset: waits are often bimodal and model-dependent. We prefer pole shares and model-level breakdowns for public claims. On-page citation blocks may still show a 12-month quotable median when the sample clears our citation floor, always labeled with its own n.

Two clocks: purchase time vs collection time

Wait-index work uses two different time axes, and they are not interchangeable. Purchase-quarter cuts group reports by when the buyer says they received the watch. Collection-date series group reports by when the report entered our dataset. A gap on the purchase-quarter axis often means “below our publication threshold” or “few people recalled that year,” not “the market went quiet.” Clustering in 2023 and 2026 on purchase dates reflects acquisition waves in who submits to Unghosted as much as it reflects AD wait conditions.

The rolling collection-window index (fixed recent sample by submission time) is the continuous headline series. Purchase-quarter breakdowns stay as a separate report cut, with thin quarters suppressed rather than plotted as empty market years. Snapshot history records which metric was computed on which calendar day so the two clocks stay auditable.

Why we show median, not mean

Wait times are right-skewed. A small number of extreme waits (years for VIP-only watches like the Daytona Le Mans or Patek Nautilus) pull averages upward in ways that mislead the typical buyer. The median is the middle value when reports are sorted by wait bucket. We take that value at floor(n/2) on the sorted list. When the sample size is even, that is the upper of the two middle reports (the longer of those two waits), not an average of the two buckets. If those two middle reports fall in different buckets, the calculated median is still the upper bucket, but the typical wait we show is the range spanning both. Cost of Waiting and grey-market midpoint math keep the single upper bucket, because a span is not a price.

Wait durations on a bucket boundary

Buyers often state a wait as an exact number of months or years: 3 months, 6 months, 1 year. Those numbers sit on the edge of two adjacent duration labels. We code the stated duration to the bucket that names that number as its upper bound. 1 month and 3 months are 1-3 months. 6 months is 3-6 months. 12 months or 1 year is 6-12 months. 2 years is 1-2 years. 3 years is 2-3 years. 5 years is 3-5 years. Durations shorter than one month stay on the week labels (under 1 week, 1-2 weeks, 2-4 weeks). Vague bounds such as “under 3 months” or “a few months” are not assigned a duration bucket.

Sample size honesty

A typical-wait figure (calculator, wait answers, and family medians on brand pages) needs at least 10 reports in the 12-month window. Below that we show the count and invite a report. We do not invent a median. If a region slice misses that floor, we show the national median and label it as national. Family slices that miss the floor stay empty; they do not fall back to the brand.

Recency visuals use an 18-month window. Current versus historical is the purchase month, not a stored flag. If that window has fewer than 10 dated reports, the figure is all dated reports, labelled all-time, still with a floor of 10. Slices under 10 all-time show the count and no median.

Public wait-share percentages use a higher floor of 40 reports. Below that we name the count instead of a bare percentage. Purchase-history two-way splits need 15 reports on each side. Those floors are separate from the 10-report typical-wait floor.

Canonical references

When a buyer knows the reference, we prefer the canonical Rolex reference number (e.g., 126710BLNR for the Batman / Batgirl GMT-Master II). Model/reference is optional on the submit form - a brand + family + wait report is enough to publish. When a reference is provided, Rolex GMT-Master II, Cosmograph Daytona, Land-Dweller, and Submariner use a canonical picker so the same watch is not counted under multiple names.

Updates

Data is updated continuously as new submissions arrive. Page summaries are generated on every page load, so the numbers reflect the current state of the database. We do not cache statistics for periods longer than a single request.

No email gate

Anyone can read all of our data without signing up, providing an email, or creating an account. We believe waitlist data should be public. No paywall, no signup wall, no email capture before viewing reports.

Unghosted does not use affiliate links. There are no sponsored placements, no dealer relationships, and no paid content of any kind.

What we don't publish

Brand journeys on /journey are private planning tools. They never appear in public statistics unless the user explicitly submits a wait report through the existing submit form. Journey tokens are excluded from the sitemap and stripped from analytics URLs.

Selection and survivorship bias

Two filters sit between real AD experiences and the numbers on this site. First, who reports: we depend on people who choose to post or submit. Buyers who got a watch quickly may be more willing to share, and forum threads also over-index on long waits and complaints. Both directions are selection bias; we do not claim the sample is a random draw from every Rolex list.

Second, what we count as usable: we track declined, list-closed, still-waiting, and gave-up outcomes. Declined, list-closed, and gave-up leave the public denominator. Still-waiting stays in so open lists do not inflate completed-wait shares. A figure like “23% waited over a year” is measured against that shared N. See Usable reports above.

Geographic coverage also skews toward the US. Thin Rolex families and most non-Rolex brands remain sparse; sample-size caveats on those pages are the product, not a footnote. Density is the bottleneck, which is why we ask for reports from the page you are reading.

Limitations

The selection and survivorship filters above are the main limitations. Beyond those, some Rolex families and most other brands still have samples that are not yet statistically meaningful for model-level claims. We try to be transparent about this through sample-size caveats throughout the site.

Contact

Questions or corrections: hello@unghosted.io.