Skip to main content
BikeSize

How we source and update product and geometry data

A stack figure with no source is just a number someone typed. Every geometry record on BikeSize carries where it came from, how it was captured, when it was last checked, and how much we trust it. This page explains that record, and what happens when part of it is missing.

Where the numbers come from

  • Geometry comes from official manufacturer geometry charts, spec sheets, and PDFs. Marketplace listings and third-party recreations of a geometry table are not accepted, no matter how confident they look.
  • Size charts come from the brand's own published fit chart. How those are stored is covered in how we normalize manufacturer sizing.
  • Product details come from manufacturer product pages, size charts, support articles, and manuals, plus retailer listings for price and availability. The ranking between those is set out in how we select and verify gear.

What is recorded with every geometry record

Each model family stores, as structured fields:

  • The source: a label, the URL, the publisher, and whether that URL is an official manufacturer domain
  • The capture method: official geometry chart, official PDF, official spec sheet, or migrated from earlier in-repo data
  • The date the rows were last reviewed against that source, as a plain calendar date
  • A confidence level, plus an optional note about what is still pending

Measurements are stored in millimetres and angles in degrees, so imperial sources get converted once on the way in rather than being re-converted by every page that reads them. Stack and reach are required on every size row. Everything else, from head angle to standover, is only present when the manufacturer published it.

Confidence, and what each level is allowed to do

  • High
    What it means
    Transcribed straight from the official geometry chart and spot-checked afterwards.
    What it may power
    Anything. High confidence is only permitted when the source is an official manufacturer domain, and validation rejects the record otherwise.
  • Medium
    What it means
    Taken from an official page but not re-verified value by value, or migrated from older in-repo data that is waiting on a direct re-check.
    What it may power
    Public pages, alongside its source link and review date. Most of the current geometry dataset sits here, and we would rather say so than round it up.
  • Low
    What it means
    Partial or inferred.
    What it may power
    Nothing on its own. Low-confidence rows are not allowed to carry a public page.

How often it gets rechecked

Data is re-checked against its source when a model year changes, when a brand reworks a chart, and at minimum every twelve months. That matches the review cycle in the editorial policy. Scraped product details also record the timestamp of the capture, so a price or a spec on a page can be traced to the moment it was read rather than presented as timeless.

Where a figure is old enough to matter, the honest move is to check the manufacturer before you buy. Prices and stock in particular go stale faster than any static page can track.

What validation blocks before publication

These rules run automatically on every build, so a bad record fails the site rather than reaching a reader:

  • Measurements outside physically plausible ranges
  • Unit mix-ups, such as 56 entered in a field that expects 560 mm
  • A missing source URL, or one that is not a real web address
  • High confidence claimed without an official source
  • Empty or duplicated size labels
  • Stack or reach that decreases as sizes go from small to large, which is almost always a transcription slip
  • A review date in the future

What we do not do

  • We do not measure bikes or test products. No number here comes from hands-on measurement, ride testing, or lab work.
  • We do not fill gaps by estimating. A missing measurement stays missing.
  • We do not quietly refresh an old figure to the current model year and leave the review date alone.

Corrections

If a number here disagrees with the manufacturer, report it on the public issue tracker with a link to the official chart. Reports are handled in the open so other readers can see what has been flagged. We aim to respond within five working days, and a confirmed error is corrected at the source record, which updates every page that reads it.

Back to all methodology pages.