Trim Tap Intervals with an Explicit Rule

Deletion must be part of the evidence

A tap outlier trimmer should never turn “that looks wrong” into an invisible action. This tool offers a small set of named rules, applies exactly one, and keeps the untouched sequence beside the result. Every removed value retains its original position and a reason.

Use it only after deciding why the rule fits your procedure. Removing warm-up movement before a trial is different from deleting the largest value after seeing the desired BPM.

Available rules

“Remove first” addresses a predeclared first-gap warm-up. “Remove last” addresses a predeclared stopping artifact. “Remove first and last” handles both boundaries. “Remove one shortest and one longest” is a symmetric extreme rule. The combined option applies boundary removal first, then removes one remaining shortest and one remaining longest.

When equal extremes exist, the earliest original position is removed. That tie convention makes the output reproducible. The page removes at most four intervals and requires at least three to remain.

These rules are mechanical, not endorsements. A first interval can be representative. An extreme can reflect real source change. Symmetric trimming can still bias a small list.

Side-by-side calculation

Both panes show count, mean and median milliseconds, per-interval mean and median BPM, minimum, maximum, and range. The comparison states numerical deltas rather than “improved” or “worse.”

Suppose the raw intervals are 710, 502, 499, 501, 498, and 760 milliseconds. Choosing “Remove first and last” retains 502, 499, 501, and 498. Raw mean duration is 578.33 milliseconds. Trimmed mean is 500. Raw range is 262 milliseconds; trimmed range is 4.

The contrast is large, but the tool does not claim the trimmed value is more accurate. The correct interpretation depends on whether 710 and 760 were genuinely planned boundary artifacts.

Extreme removal example

For 500, 498, 502, 497, 503, the symmetric rule removes 497 and 503, leaving 500, 498, 502. The retained mean remains 500. If two entries were tied at 497, the earlier original position would be removed and the other would remain.

Document the rule before quoting the trimmed result: “One shortest and one longest interval removed; tie goes to earliest position.” Without that line, a reader cannot reproduce the transformation.

Why no automatic IQR filter

Automatic thresholds can sound objective while hiding strong assumptions, especially for short tap sequences. An IQR rule may identify no value in one small list and several in another. A z-score assumes a distribution that the sequence may not support.

This launch tool keeps removal finite and easy to narrate. The separate interval analyzer still reports quartiles and IQR without deleting anything.

Read the delta carefully

If trimmed mean BPM rises, the rule probably removed relatively long durations or otherwise changed the rate mix. If mean shifts but median barely moves, a removed value had more influence on the arithmetic average than on the center. If range collapses, remember that range always depends on extremes.

No delta evaluates a musician, dancer, device, or performance. It describes the consequences of the chosen transformation.

Validation and reset

Every token must be numeric, finite, and within the stated millisecond bounds. A rule that would leave fewer than three values is rejected before calculation. The page never falls back to another rule.

Clear removes the source, selected rule, and both panes. Reload can also erase active state because no persistent storage exists.

Appropriate and inappropriate uses

Appropriate use includes a documented protocol that always discards a first practice gap, a classroom demonstration of outlier influence, or an audit comparing two preprocessing choices.

Inappropriate use includes trying rules until a preferred BPM appears, concealing removed values, treating spread reduction as proof of truth, or trimming a real tempo change merely because it is inconvenient.

The tool cannot verify the source, detect missed taps, or decide whether a boundary was equivalent.

Sensitivity note for small lists

Removing two values from five changes 40 percent of the evidence. The side-by-side count makes that proportion visible. As the initial list grows, the same fixed removal affects a smaller share, but it can still target important boundaries. Always report both counts and avoid interpreting a reduced range as an independent confirmation.

For a repeatable audit, record the chosen rule before pasting values and keep that declaration beside both outputs.

Frequently asked questions

Paste a synthetic or non-sensitive list, select the rule you planned, inspect every removed position, and copy both raw and trimmed summaries with the rule statement.