Repeatable bike measurements: compare readings, not noise
A practical repeatability protocol for manual bike coordinates: relocate the tool, keep each reading, check the reference and separate observed spread from a claimed setup change.
Published 8 October 2026 · OpenBikeFit
Measurement field guide
Keep all readings, not just a mean.
Observed span: 743 − 740 = 3 mm
Mean ≈ 741.3 mm; not guaranteed accuracy
Fictional readings, not a fit result.
On this page
- 01Write the protocol before the first reading
- 02Make each reading an independent placement
- 03Keep the original series visible
- 04Do not average away a protocol problem
- 05Compare matched before and after series
- 06Keep measured, modeled and reported evidence separate
- 07Use a record that preserves the method
- 08Decide what the observations can support
Record before interpreting
Use the free blank worksheet
Keep the physical reference, units and repeated readings together without creating a saved bike.
Open toolA before-and-after value is easy to write and harder to interpret. The bicycle might have changed, the endpoint might have changed, or the measuring method might have varied. Repeated readings help expose that distinction without pretending that a tape measure has become a laboratory instrument.
This protocol concerns manual bicycle dimensions, not camera confidence, clinical outcomes or instrument certification. It records what your own procedure produces. The goal is to keep the observations usable and to recognise when the evidence is too weak for a dimensional claim.
Write the protocol before the first reading
Define the task and quantity: for example, direct BB-centre-to-named-saddle-point distance in millimetres, or horizontal Bar X to the clamp centre. Specify the bicycle alignment, endpoint rule, tape path or coordinate method, tool, side and loading condition. Take a moment to check whether an earlier record uses the same definition.
A published protocol is valuable because it names points, not because its labels can be borrowed without its method. Velogic's measurement definitions distinguish saddle, bar, hood and frame endpoints. Your manual record should likewise expose its own definitions. It does not inherit that system's calibration, accuracy or repeatability merely by using similar field names.
Make each reading an independent placement
Read the value, write it down, remove the tool and set it up again. Refind the declared endpoint rather than leaving the tape or pointer in place. Check that the bicycle support and datum have not shifted. Use the same method for each placement; switching between a diagonal tape and a vertical component creates different quantities, not repeated observations.
Do not tighten, loosen or move components between baseline readings. This is a recording procedure, not a mechanical-adjustment instruction. If a secure measuring arrangement cannot be maintained, stop and record the limitation. A second person may help identify alignment or reading mistakes, but agreement between two people is not a calibration certificate.
Keep the original series visible
Fictional readings of 740, 743 and 741 mm have a minimum of 740, a maximum of 743 and a span of 3 mm. Their arithmetic mean is about 741.3 mm. Keep all three values beside any summary so another reader can see what happened. A mean alone conceals the difference between consistent readings and a widely scattered series.
The span is the range of these observed values. It is not a guarantee that the true value lies within them, and it is not a statistical confidence interval. A consistent offset in the datum or instrument can affect every reading in the same direction while producing a small span. Precision in display and accuracy of measurement are separate ideas.
Do not average away a protocol problem
A fictional series of 740, 747 and 741 mm includes a much larger disagreement than the previous example. First check the endpoint, tape alignment, instrument placement, support and transcription. Do not simply discard 747 because it is inconvenient or average the series into a value that looks acceptable.
If a specific mistake is identified, retain the original observation with that explanation and take a new complete series. If the cause remains unknown, record that the method is unresolved. There is no universal millimetre threshold on this page that certifies every bicycle, tool or purpose. Apply a tool's own declared validation rules when using that tool, without turning them into physical accuracy guarantees.
Compare matched before and after series
Suppose a fictional baseline series is 740, 743, 741 mm and a later series is 742, 745, 743 mm. The means differ by 2 mm, while each observed span is 3 mm. The arithmetic is correct, but those few readings alone do not establish a certain 2 mm physical change. Check matching conditions and the source of variation before drawing that conclusion.
Keep component identities, point definitions, method and alignment alongside both dates. A new saddle reference or a changed bar endpoint invalidates a simple like-for-like comparison even if the field names are identical. Preserve separate named records when they answer different questions. Missing baseline data cannot be reconstructed by confidence or extra decimals.
Keep measured, modeled and reported evidence separate
An installed dimension is a physical observation under a method. A component calculator's result is a modeled geometric change under declared inputs. A rider's report is a different kind of evidence. Recording any one of them does not automatically confirm either of the others. Label the source instead of placing all three into one column called result.
The same distinction matters when a camera check is involved. A camera date has its own setup and quality conditions; it is not an independent verification of a tape's endpoint. For a manual-coordinate task, keep the coordinate protocol as the baseline. Do not convert joint-angle confidence into a guarantee about the bicycle's installed dimensions.
Use a record that preserves the method
The blank worksheet supports an unentered paper-first record. My bikes offers an optional local passport and guided measurement flows; entering or saving there is a separate action. The setup card can present a saved bike with references, but a document's neatness does not repair an unresolved measurement. Keep a copy of the original readings in your own controlled workflow.
A repeated-reading sheet should include the dates, units, individual values, endpoint, origin, method, conditions and any excluded reading with its reason. Mark unavailable values Not measured rather than zero. External paper, downloads or shared copies are not recalled when a page closes; choose how those records are retained separately from reading this guide.
Decide what the observations can support
If the point and method can be reproduced, the record can support a clearer dimensional comparison or a better workshop question. If they cannot, the useful next step is resolving the measurement, not prescribing a component change. Preserve uncertainty visibly so the next person does not mistake an unresolved reading for a verified baseline.
Repeated manual readings do not establish comfort, safety, compatibility or an ideal position. Any later mechanical action still needs exact component instructions and an appropriate physical check. This protocol helps prevent overclaiming from a number; it is not an approval to adjust or a guarantee that a small difference matters to a rider.
Practical questions
Frequently asked questions
Should I take the same measurement more than once?
Yes, repeat the placement and refind the same named endpoint under the same conditions. Keep each reading and investigate disagreement instead of selecting a favourite.
Does a 3 mm reading span mean my tool is accurate to 3 mm?
No. It describes that series only. A common offset can affect every reading, and the observed span is neither an instrument specification nor a confidence interval.
Does a smaller average prove that a setup improved?
No. First establish a comparable dimensional observation. A modeled value, physical reading and rider report are separate; none alone proves a better fit or safe installation.