Bike-fit video repeatability: compare takes, not screenshots
Separate several cycles in one clip, independent takes of an unchanged setup and before-and-after conditions. Keep disagreement visible instead of selecting a preferred pair.
Published 8 October 2026 · OpenBikeFit
Camera field guide
Keep the inconsistent take.
Take 1
142°
Take 2
143°
Take 3
148°
Fictional takes: 142° / 143° / 148°
The matching first pair does not erase the third take.
An agreement check is not absolute accuracy or a fit verdict.
On this page
Understand before recording
Compare eligible local recordings
Use the existing free workflow and read its source and scope checks; opening a page does not accept consent or verify a rider.
Open toolA smooth trace can be internally consistent while the recording setup is wrong. Replaying that trace produces the same result because it contains the same evidence. To ask whether a camera observation can be repeated, you need another recording made under the same declared condition. To ask whether a later condition differs, you need a different comparison with both sources eligible. These are separate questions, not interchangeable counts of screenshots.
This guide explains the OpenBikeFit recording-condition workflow and its limits. It does not tell you to make a hardware change or guarantee that matching videos represent correct physical measurements. The examples use fictional angles and app-specific rules. A confirmed technical condition is useful context, not a certificate of installation, comfort, accuracy or clinical safety.
Three different levels of repetition
First, cycles within one clip let the engine summarize repeated pedalling and observe cycle-to-cycle variability. Second, independent takes within one unchanged condition test agreement across separately started recordings. Third, before-and-after conditions compare eligible sources from distinct stages or dates. A clip duplicated twice remains one observation; it does not pass the independence requirement simply because it has two filenames.
Keep the level written next to a claim. A representative knee maximum within a clip is not a repeated physical measurement of saddle height. A pair of agreeing takes is not automatically an eligible cross-date comparison. The local workflow stores lineage and protocol with the condition so that these distinctions are not reduced to the date shown on a card.
Preserve the unchanged condition
Between baseline takes, keep the bicycle setup, camera side and placement, lens, capture mode, rider context and declared protocol consistent. Do not move a component while collecting an unchanged condition. If the camera or bicycle support moved, start a new appropriate condition rather than relying on a repeated display number. A declaration records what you say stayed the same; the app cannot inspect everything physically.
The current engine also checks source context, the selected Fit Plan, safety acknowledgment state, capture protocol and ruleset. A later edit can make earlier evidence no longer current. Those checks do not establish that every physical detail matched, but they prevent software from treating incompatible saved sources as the same comparison merely because their names or timestamps look plausible.
Check every accepted take
The current recording-condition logic needs at least six clean cycles and moderate or high measurement quality for every decision metric in each take. It checks estimated knee extension, hip closure, and representative elbow and trunk angles, with matching sides. A missing or unusable metric is not replaced by whichever other number happens to agree. One accepted take remains a collecting state; a second is required.
For two technically accepted takes, the app compares their metric difference with combined observed noise. If they disagree, the workflow can request one final take. With three accepted takes, all relevant pairwise comparisons and adjustment-direction signatures must agree. It never picks the most convenient two and hides the remaining inconsistent take. If inconsistency persists, hardware advice remains locked for that condition.
A fictional agreement example
Suppose two eligible takes report knee extension values of 142 and 143 degrees, each with app noise set to 2 degrees. Their difference is 1 degree; the app combines those noise values as the square root of 2 squared plus 2 squared, about 2.8 degrees. This example illustrates a comparison rule, not a measured error bound or a statistical confidence interval for a rider.
Now add an eligible third take at 148 degrees with the same fictional noise. It differs from the first by 6 degrees, so the attractive 142/143 pair cannot make the whole condition consistent. Keep the third result and inspect what changed. More takes are not a license to keep trying until two produce the desired answer. The app-specific minimum noise floor is not a promise of two-degree physical accuracy.
Repeatable is not necessarily accurate
An unchanged oblique camera or a consistently misplaced landmark can yield matching numbers across recordings. Repetition reveals some variation, but it cannot prove that the shared error is absent. A different laboratory method can use different landmarks and produce different means even when each instrument is repeatable. Do not claim that a repeated phone result has been validated against another system unless that exact comparison was actually performed.
Keep capture quality and independent agreement visible beside the result. Do not convert an observed difference into an expected millimetre change or a guarantee of comfort. If the question concerns actual bicycle dimensions, repeat physical coordinates with named points and an instrument. A joint angle, a mechanical coordinate and a rider report should remain separately labeled evidence.
Compare dates only after source checks
The cross-session comparator is for eligible confirmed conditions with a compatible frozen bicycle setup and protocol. It does not turn two arbitrary saved dates into a valid before-and-after study. The workspace rechecks selected sources at the deliberate action; changed, missing or ambiguous sources require a fresh choice. Reading this article does not open a fit or alter its saved data.
Use the fictional analysis to learn the states without collecting a rider video. When you have real, eligible local recordings, use the existing camera comparison tool and read its reasons if it declines a comparison. No automatic retry or replacement result is implied by this guide. Local copies can be lost when device data is cleared; a display or screenshot is not a complete recovery archive. Resolve incomplete evidence instead of labeling it no change.
Practical questions
Frequently asked questions
Can I reuse the same clip as the second take?
No. A replay, duplicate clip or another screenshot repeats the same observation. The unchanged-condition workflow requires independently started recordings with the same declared setup.
Why does a third take block two matching takes?
The app keeps all technically accepted takes in the condition. Agreement must hold across all required metrics and pairs; selecting a convenient pair would hide contrary evidence.
Does a confirmed condition prove an accurate fit?
No. It checks technical eligibility and repeatability under the declared protocol, not absolute accuracy, physical installation, comfort or clinical outcomes.