Reviewed concrete example
Define the comparison question
Before you compare BPM counting sessions, state what differs. Are you comparing two passages, two input methods, before and after a planned break, or repeated attempts under one condition? An A/B label does not create control by itself.
Keep everything else as comparable as practical: pulse level, source version, section boundaries, device, browser, recent window, timeout, planned interval count, and trimming rule.
Match the evidence unit
Do not compare quarter-note taps in A with half-note taps in B. Do not compare mean of per-interval BPM on one side with BPM converted from mean duration on the other. Preserve units and definitions.
If only summaries remain, record interval count, mean and median milliseconds, minimum, maximum, and the exact preprocessing. The comparator can calculate transparent deltas without reconstructing raw data.
Balance order and repetition
Order can mix with practice, fatigue, attention, or source drift. For input-method comparisons, alternate A–B–B–A or another planned order. Run multiple separate pairs rather than merging everything immediately.
Repetition does not prove causation, but it shows whether a single pair is unusual.
Worked summary pair
Session A: 24 intervals, mean 500 ms, median 499, min 484, max 518. Session B: 24 intervals, mean 506, median 504, min 488, max 526.
Median-duration rates are about 120.24 and 119.05 BPM, a B-minus-A difference near −1.19. Ranges are 34 and 38 milliseconds. Counts match.
A narrow conclusion is: “For these matched passages, B had a 5-ms longer median interval and 4-ms wider range.” Do not conclude that a person worsened or a device is universally slower.
Keep raw and trimmed versions straight
If both sessions use a predeclared first-gap removal, record raw summaries and apply the rule equally. If only B has an obvious extreme, preserving it may be the honest comparison; a sensitivity analysis can show results with and without the declared value.
Never call trimmed and raw summaries equivalent.
Signed deltas
Use B minus A consistently. Positive milliseconds mean B is longer at the chosen pulse; positive BPM means B is faster after conversion. These signs move oppositely because rate is reciprocal.
Range difference describes endpoint spread, not distribution or significance. Count ratio describes balance, not quality.
What summary-only comparison cannot do
It cannot reveal order, drift, quartiles, MAD, CV, or whether extremes occurred at boundaries. It cannot pool exact per-interval BPM. Two different raw sequences can share the same summary.
If causal or inferential claims matter, use an appropriate experimental design and statistical method beyond this product.
Documentation template
Write question, source and passage, pulse, A/B condition, order, device/browser, settings, interval count, raw or trimmed status, duration centers, rate conversions, ranges, signed deltas, anomalies, and limits.
Store notes in your approved system. BPMCounter.click does not save or synchronize comparisons.
Privacy and ethics
Use generic labels rather than names. Do not turn descriptive timing summaries into employment, audition, health, disability, or disciplinary evaluation. The tool supplies no performance score.
Inputs remain in browser memory with no account, database, analytics payload, or persistent storage.
Separate observation from conclusion
A strong comparison note has two layers. Observation lists the two source summaries and signed deltas. Interpretation names plausible context and uncertainty without turning it into fact. For the worked pair, observation is “B median interval is 5 ms longer.” Interpretation might be “the second condition, source passage, or input timing could contribute.”
Do not replace multiple pairs with one favorite example. A small table of every planned pair shows repeatability and order effects. If a session violates a predeclared rule, retain an exclusion note and reason.
When raw values are available
Use raw interval tools to compare distribution and order before reducing to summaries. Check factor-of-two events, boundary artifacts, and drift separately. The summary comparator is a fallback record, not a reason to discard richer evidence. Preserve raw and summarized versions under clear labels.
Stop conditions
Decide in advance how many intervals and repetitions each condition receives, plus what operational failure requires a rerun. A fixed stop rule prevents one side from accumulating extra attempts because its early results looked less convenient. Report every planned pair or a documented exclusion.
Give each preserved session a stable label and collection time before entering its summary.
Frequently asked questions
Must counts be identical?
Matching improves one aspect of comparability, but procedure and passage still matter. Report count differences.
Can I average A and B?
Keep them separate unless a defensible pooling method and raw evidence exist.
Does a signed delta show cause?
No. It describes direction under the entered summaries.
Should I compare recent windows?
Only when both use the same size and comparable session positions.
Write the question, fixed conditions, order, count, and preprocessing rule first. Run separate sessions, enter their summaries, and report signed differences with caveats.
