This public example shows what OpenBikeFit checks and what it refuses to conclude. It uses a generated pedalling sequence, not a person, a customer result or your device. Reading this page does not ask for a camera, an account or a saved measurement.
Starting the interactive demo creates a local session and asks for the adult/on-device consent shown by the fit tool. This page itself is only an explanation.
Illustrative image. The example below describes synthetic data, not a diagnosis or a fitted position.AI image
The fixture supplies two different programmed conditions. The useful result is not a single angle or a fit score: it is a traceable sequence of accepted input, repeated measurement and comparison. The numerical reading below does not predict the effect of a physical change.
01Input
Can this side view be measured?
The fictional rider is filmed from the side. The analysis checks framing, visibility, stability and usable pedalling cycles before reporting joint angles. Poor input is a reason to record again, not a reason to make a bike change.
02Repeat
Does a second take agree?
Several cycles give an angle range rather than one flattering frame. The unchanged setup needs an independent take; disagreement asks for another recording instead of selecting the most convenient number.
03Re-check
What changed after one trial?
Two different programmed conditions illustrate the comparison. Their movement difference is checked against observed variation. This is not a model of a physical saddle adjustment or a claim of benefit for a rider.
These are the tested synthetic scenario's qualitative states, not a live result. The real camera path may reject a clip or withhold a recommendation when quality, repeatability or safety conditions are not met.
Contents · read the worked report
Start with the source, not the green label
This is a complete worked reading example, not a report about you. Four generated sequences pass through the same analysis, repeatability and comparison functions used by OpenBikeFit. The results are printed on this page during the site build. Reading them does not run a camera, create a fit, inspect saved records or accept the interactive tool. The visible numbers therefore illustrate software behaviour, not a measured rider or a successful bicycle adjustment.
Conditions A and B use different programmed leg geometry. Each condition has two separately seeded takes, so their small tracking variations are not identical replays. That is still a simulation: four programmed takes are not four independent real experiments. In particular, B is not the predicted consequence of raising a real saddle by a stated number of millimetres. Keep this distinction when reading the apparently large knee difference below.
Keep four takes attached to two declared conditions
A1 and A2 belong to the unchanged fictional condition A; B1 and B2 belong to the separate fictional condition B. All use a drive-side view, the demo_synthetic pose engine and the rgb_basic analysis mode. Each sequence contains 240 frames over eight seconds at 30 frames per second, with a declared 720 by 1280 image shape. These are generator settings, not a test that your phone can capture or track that protocol.
In a real workflow, several cycles inside one clip cannot replace a separately prepared take of the same declared setup. The condition must retain every accepted take and its exact analysis. A disagreement is information, not permission to remove the least convenient recording. If the tool requests another take, preserve the reason and repeat the protocol; do not manufacture agreement by comparing only two attractive frames or by silently changing the bicycle between takes.
Four sources. Two conditions. One comparison per metric.
Knee sources A1 127.1° and A2 127.0° → A centre 127.1°. B1 144.9° and B2 145.5° → B centre 145.2°. Each condition noise: 2.0°. Difference: +18.1°; combined noise: 2.8°. A software comparison, not proof of rider benefit.
Read quality, confidence and agreement as different signals
The generated takes have full tracked landmark coverage, nine detected cycles and no capture-quality issues. Brightness is supplied as a fictional luminance input of 128; no real image has been checked. A quality score of 1 in this deliberately clean fixture is not 100 percent accuracy. Framing and side-view checks are software heuristics, and an accepted clip can still contain perspective, landmark-placement or calibration error.
All displayed joint summaries have a confidence band labelled high in this fixture, while each two-take condition has a band labelled moderate. Those labels answer different questions. Summary confidence combines available tracking, cycle and variation evidence; condition confidence is conservatively capped when only two takes exist. Neither is the probability of a correct fit, an uncertainty interval certified against a physical instrument, or a promise that the joint centre was located correctly. Agreement can repeat a systematic error.
Name the angle and the summary before comparing values
The knee row uses the maximum image-plane interior angle from the accepted cycle summary, labelled as an extension estimate near the bottom of the stroke. It is not a guaranteed measurement at the exact crank six-o’clock position. The hip row instead uses the minimum projected angle near the top of the stroke. The elbow and trunk rows use the midpoint of their summary minimum and maximum; they are not an average of every video frame or a selected crank-event measurement.
The take inventory retains the minimum, maximum, decision value, cycle variation and cycle count so that these definitions remain inspectable. A range such as 54.5 to 127.1 degrees describes the knee motion summary of A1, not a recommended range to achieve. For the same two planar segment directions, an interior angle of 145 degrees corresponds to supplementary flexion of 35 degrees. That arithmetic changes the label convention, not the rider, physical saddle height or certainty of the observation.
Confirm each condition before calculating its difference
For each decision metric, the condition compares the separately generated take values against a combined agreement limit. Its centre is the median of the underlying take values, displayed to one decimal place. Its noise allowance uses the larger of the per-take cycle variation, a two-degree technical floor and the largest disagreement between takes. Small variation inside a smooth generated sequence therefore cannot produce a claim of sub-degree certainty.
Both fictional conditions are confirmed by the current rules: all four decision metrics are repeatable and the generated adjustment-direction signatures agree within each condition. Confirmed is a narrow software state, not confirmation of a bicycle assembly or an accurate anatomical measurement. The centre is calculated before display rounding, so averaging the already rounded labels by hand can differ at the final decimal. Keep the reported centre and its source values rather than replacing either with a tidier number.
Read the difference together with the noise boundary
The comparison subtracts the A centre from the B centre for the same decision metric and side. It combines the two condition noise allowances by the square root of their squared sum. Here each allowance is 2.0 degrees, giving approximately 2.8 degrees combined noise. The software uses a strict greater-than test against the unrounded combined value, not a greater-than-or-equal test against the displayed 2.8 label. A displayed difference alone is not enough to interpret the outcome.
The knee centre changes from 127.1 to 145.2 degrees: plus 18.1 degrees, beyond that noise boundary. Its software label is improved because its distance to the current illustrative knee reference band decreases. The hip changes by plus 2.5 degrees and remains within noise. The elbow and trunk do not change at the displayed precision and also remain within noise. The four rows stay separate; no overall fit score is created by counting green labels.
Knee extension estimate
Maximum interior angle across accepted cycles; estimated near bottom of the stroke. Side: right.
Software reference: 140.0°–150.0°. Context, not a personal target.
A1 decision value
127.1°
Summary min 54.5° / max 127.1° Cycle variation 0.7° 9 cycles; high summary confidence
A2 decision value
127.0°
Summary min 54.4° / max 127.0° Cycle variation 0.9° 9 cycles; high summary confidence
B1 decision value
144.9°
Summary min 58.2° / max 144.9° Cycle variation 0.9° 9 cycles; high summary confidence
B2 decision value
145.5°
Summary min 58.2° / max 145.5° Cycle variation 0.9° 9 cycles; high summary confidence
Condition A · confirmed · moderate confidence
Centre 127.1°; noise 2.0°; largest take disagreement 0.1°; agreement limit 2.8°.
Condition B · confirmed · moderate confidence
Centre 145.2°; noise 2.0°; largest take disagreement 0.6°; agreement limit 2.8°.
B − A: +18.1° · Combined noise 2.8° · Beyond noise Software label: improved. Not a rider comfort or physical-fit verdict.
Hip closure estimate
Minimum projected hip angle across accepted cycles; estimated near top of the stroke. Side: right.
Software reference: 40.0°–55.0°. Context, not a personal target.
A1 decision value
42.2°
Summary min 42.2° / max 95.1° Cycle variation 0.5° 9 cycles; high summary confidence
A2 decision value
42.1°
Summary min 42.1° / max 95.0° Cycle variation 0.5° 9 cycles; high summary confidence
B1 decision value
44.7°
Summary min 44.7° / max 102.9° Cycle variation 0.3° 9 cycles; high summary confidence
B2 decision value
44.7°
Summary min 44.7° / max 102.9° Cycle variation 0.5° 9 cycles; high summary confidence
Condition A · confirmed · moderate confidence
Centre 42.2°; noise 2.0°; largest take disagreement 0.1°; agreement limit 2.8°.
Condition B · confirmed · moderate confidence
Centre 44.7°; noise 2.0°; largest take disagreement 0.0°; agreement limit 2.8°.
B − A: +2.5° · Combined noise 2.8° · Not beyond noise Software label: within noise. Not a rider comfort or physical-fit verdict.
Elbow representative angle
Midpoint of the cycle-summary minimum and maximum; not a chosen pedal-event frame. Side: right.
Software reference: 150.0°–170.0°. Context, not a personal target.
A1 decision value
172.4°
Summary min 172.4° / max 172.4° Cycle variation 0.0° 9 cycles; high summary confidence
A2 decision value
172.4°
Summary min 172.4° / max 172.4° Cycle variation 0.0° 9 cycles; high summary confidence
B1 decision value
172.4°
Summary min 172.4° / max 172.4° Cycle variation 0.0° 9 cycles; high summary confidence
B2 decision value
172.4°
Summary min 172.4° / max 172.4° Cycle variation 0.0° 9 cycles; high summary confidence
Condition A · confirmed · moderate confidence
Centre 172.4°; noise 2.0°; largest take disagreement 0.0°; agreement limit 2.8°.
Condition B · confirmed · moderate confidence
Centre 172.4°; noise 2.0°; largest take disagreement 0.0°; agreement limit 2.8°.
B − A: 0.0° · Combined noise 2.8° · Not beyond noise Software label: within noise. Not a rider comfort or physical-fit verdict.
Trunk representative angle
Midpoint of the cycle-summary minimum and maximum; comfort-first reference context. Side: sagittal.
Software reference: 40.0°–55.0°. Context, not a personal target.
A1 decision value
49.1°
Summary min 49.1° / max 49.1° Cycle variation 0.0° 9 cycles; high summary confidence
A2 decision value
49.1°
Summary min 49.1° / max 49.1° Cycle variation 0.0° 9 cycles; high summary confidence
B1 decision value
49.1°
Summary min 49.1° / max 49.1° Cycle variation 0.0° 9 cycles; high summary confidence
B2 decision value
49.1°
Summary min 49.1° / max 49.1° Cycle variation 0.0° 9 cycles; high summary confidence
Condition A · confirmed · moderate confidence
Centre 49.1°; noise 2.0°; largest take disagreement 0.0°; agreement limit 2.8°.
Condition B · confirmed · moderate confidence
Centre 49.1°; noise 2.0°; largest take disagreement 0.0°; agreement limit 2.8°.
B − A: 0.0° · Combined noise 2.8° · Not beyond noise Software label: within noise. Not a rider comfort or physical-fit verdict.
Separate a software reference from a rider benefit
Improved is the existing comparison label for movement towards a software reference band after the noise check. In this example it does not mean more comfortable, faster, safer, correctly installed or medically appropriate. The current reference interval is shown beside every row so that the label has an explicit basis. Trunk context is comfort-first here; changing the declared mode may change its reference interpretation, not the recorded movement itself.
Within noise means this comparison does not establish a difference beyond its observed allowance. It does not prove that the real movement is identical. Inconclusive is different: a difference can exceed noise without reducing or increasing distance to the reference, for example when both observations remain equally inside the band. Do not translate either state into “keep adjusting until green.” A physical setting, a video estimate and the rider’s experience remain distinct records with different evidence requirements.
Distinguish a current comparison from retained history
A current re-check needs the exact eligible before-and-after conditions, their clips and analyses, the current plan and context, and the current measurement rules. A newer baseline, missing source or mismatched stored comparison must not inherit the previous round’s chart or decision. In the live tool, an unavailable current comparison gives a recording or review next step instead of silently selecting an earlier pair. This public page has no saved session, so it is not asserting a current state for your device.
Reading example: an eligible A/B pair can support the numerical comparison above. If B’s referenced analysis is missing, keep the earlier journal text as history but do not present it as a freshly validated current result. If the rider has independently reported a ride response before the camera re-check, that response remains its own evidence; it does not fill in a missing movement comparison. These are explanations of states, not buttons that mutate a record or simulations of a crash recovery.
Carry physical references and open questions to the next step
An angle difference cannot be converted into a saddle-height instruction in millimetres. Segment geometry, projection, contact points and the movement strategy can all matter; this generated A/B pair is not a calibrated bicycle model. Before discussing an actual reversible change, preserve the measured bicycle coordinate, its exact endpoints and datum, the component configuration and the relevant manufacturer limits. A camera estimate cannot approve torque, insertion, compatibility or installation.
Use the separate blank technical worksheet to record physical references in your own controlled workflow. Its 33 blank fields are a general bicycle hand-off, not a dedicated video-report form or a medical record. Keep the original clips and the exact metric convention separately where appropriate; a session JSON does not include separate landmark or raw-video files. A useful question is “what can these sources distinguish?” rather than “which number should I copy onto my bike?” Missing information should remain unknown, not become a measured zero.
Choose a reading path before starting an interactive tool
If the terms are unfamiliar, read the camera setup, angle conventions, confidence and repeatability guides linked below. They explain how to prepare and label evidence without starting a capture. The methods guide distinguishes a body-input calculator, recorded movement, remote human review and in-person work. This numerical example is software output, not a remote consultation or a substitute for a qualified assessment when your question needs one.
Other publishers also separate their report from the service around it. MyVeloFit describes a key-angle report and a separate expert-review option; BikeFastFit describes report outputs alongside a public explanatory academy; Retül describes a final physical bicycle map after its fitting process. Those narrow process descriptions are not accuracy rankings, endorsements or guarantees adopted here. Our example numbers come only from the unchanged OpenBikeFit generator and analysis functions, not from their customers, report images or measurement tables.
Report-reading questions
Can I read this example without a camera or a saved fit?
Yes. This page is a server-rendered reading example with fixed fictional output. It does not request camera permission, read your saved fits, create a local session or accept the interactive service. Opening the separate demo or camera tool is a different, explicit action with its existing gates.
Does improved mean the rider has a better bike fit?
No. It is the software comparison label for reduced distance to an illustrative reference band after the noise check. It is not a comfort report, performance result, diagnosis, installation approval or proof that the bicycle change caused the movement difference.
Why is the hip change within noise even though the number changed?
The displayed hip difference is 2.5 degrees, below the approximately 2.8-degree combined allowance. A changed displayed number is not automatically a distinguishable movement change. Within noise does not prove identical real motion; it limits what this particular comparison supports.
Why not compare one best frame before and after?
A selected frame can use a different point in the pedal cycle or hide tracking variation. This example retains multiple cycles and two separately generated takes per condition. In real recording, independently prepared takes of the same declared setup check agreement; several frames from one clip do not replace them.
Is 145 degrees the same as 35 degrees of knee flexion?
Only when both labels describe the same two planar segment directions and use supplementary interior-versus-flexion conventions: 180 minus 145 equals 35. Different landmarks, projections or cycle events are not made equivalent by that subtraction. Neither number is a saddle-height measurement.
Can I calculate a saddle change in millimetres from this example?
No. The generated conditions use different programmed geometry, not a measured physical adjustment model. Keep physical bicycle dimensions, camera angles and rider reports separate. Exact component instructions, current physical measurements and appropriate professional judgment remain necessary for real work.
Does a confidence band labelled high or full landmark coverage prove accuracy?
No. Tracking coverage, software confidence and agreement between takes are different from error against a calibrated physical reference. A systematic projection or landmark-placement error can repeat across clear recordings. This deliberately clean generated fixture does not validate a phone, a rider measurement or the real-world accuracy of the system.
What should I retain if a current comparison becomes unavailable?
Keep the earlier journal and original evidence as history, with their source context. Do not substitute an older pair as the new current result. Follow the live tool’s explicit recording or review next step. This reading page does not repair, delete or revalidate any saved record.
Read the output
What you can take from it
A usable analysis separates the physical setup you recorded from estimates made by a 2D camera. It shows the movement range across multiple pedal cycles, the quality of the view and how much independent takes agree. Reference bands give context for a small, reversible question to test; they are not a universal prescription.
If you change one bike coordinate, measure it against the same physical reference, keep the camera protocol comparable and record another condition. The re-check can report whether a movement difference exceeds observed noise. It cannot prove what caused that difference or whether the new setup is better for you.
Do not infer
No verdict from one frame
A camera cannot diagnose pain, judge medical safety or promise performance. Clothing, lens position, occlusion, cadence and pose-estimation error can change a displayed angle. A reference interval drawn from published research may not describe an individual rider. If a symptom or injury is involved, this educational tool is not the right decision-maker.
OpenBikeFit deliberately refuses low-quality side views and asks for repeatable evidence before suggesting a hardware trial. The evidence and methodology page lists each active claim, source and limitation.
Choose your next step
Explore first, then use your own measurements
Run the interactive synthetic demo to inspect the actual report and re-check flow without a camera or trainer. Or use the free calculator to record a measured starting point before trying the camera workflow. Both paths stay on your device; neither sends a fit record to OpenBikeFit.
