Calculation Reference

Inputs: event type and positive finite marks. Rule: minimum for timed events; maximum for distances, heights, and points. Example: 12.10 s beats 12.25 s; 5.80 m beats 5.72 m. Limitation: incompatible event or class records must not be compared. Updated: 2026-08-01.

x = ln(mark)
m = median(x)
MAD = median(abs(x - m))
robust sigma = 1.4826 * MAD
robust CV percent = 100 * sqrt(exp(robust sigma^2) - 1)

Every stored exact-date mark across all available years enters the consistency cohort. Fewer than five results produce no estimate. Classical sample variation remains in technical details. Consistency is repeatability, not improvement or readiness. Updated: 2026-08-01.

timed = (old - new) / old * 100
higher-is-better = (new - old) / old * 100

Moving from 12.00 s to 11.40 s is 5.00% improvement. Moving from 5.00 m to 5.50 m is 10.00% improvement. Old and new values must both be positive and finite. Updated: 2026-08-01.

timed progress = standard / performance * 100
field progress = performance / standard * 100
timed improvement needed = max(performance - standard, 0)
field improvement needed = max(standard - performance, 0)

Values above 100% mean the published benchmark has been met or exceeded. Missing or incompatible values return no calculation. Updated: 2026-08-01.

performance score = -ln(mark) for timed events
performance score = ln(mark) for higher-is-better events
robust slope = median of all pairwise exact-date slopes
robust intercept = median(score - slope * elapsed days)

Only exact dates in the latest season enter the descriptive line. The all-career least-squares development trend and next-season continuation were withdrawn. The line is descriptive and not causal. Updated: 2026-08-01.

Every unique positive non-excluded stored mark contributes. Exact-date marks enter rolling-origin selection among the all-history median, a 365-day half-life recency-weighted level, a Theil–Sen log trend capped at 35% annual change, and a robust ensemble. Coarse-date marks enter a capped robust median anchor without receiving an invented day. Sparse estimates use conservative minimum spread floors and remain preliminary. Each range widens with the horizon and is model spread, not a probability interval. Updated: 2026-08-01.

For supported events only, StrideSignal applies the published Moinat, Fabius, and Emanuel (2018) models:

100 m: P0 = P - 0.0449w + 0.009459Pw - 0.0042w^2
200 m: P0 = P + 0.090w - 0.010w^2
Long jump: P0 = P - 0.029w
Triple jump: P0 = P - 0.069w + 0.009w^2

P is the recorded mark and w is wind in m/s. These values are labeled estimates, never official corrected results. Other events and missing wind return the recorded result without an invented correction. Updated: 2026-07-30.

interval time = current cumulative time - prior cumulative time
interval distance = end distance - start distance
velocity = interval distance / interval time
percent of total = interval time / total time * 100

Every segment must cover a positive continuous distance and cumulative time must increase. Invalid split sets return errors rather than a misleading velocity. Updated: 2026-08-01.

Pearson correlation measures linear association. Spearman correlation applies Pearson correlation to average ranks and handles ties. Pairing requires exact dates inside 21 days and a matching test protocol. Fewer than 10 pairs produce no coefficient; 10–19 are exploratory; interpretive wording begins at 20. Same-season fallback is removed. Correlation does not establish confidence or causation. Updated: 2026-08-01.