How to report an error and see what changed.
Updated September 2, 2026If a number or explanation looks wrong, email corrections@datingpoolreport.com. Include the page link, the exact statement or result, why you think it is wrong, and the original source when you have one.
Send privacy or security reports to privacy@datingpoolreport.com. Do not include personal data that we do not need.
The Research Desk checks the claim against the cited source. We check who the number counts, the dates, the model version, and whether we can repeat the result. A newer source is not always proof of an error. But we update pages that answer a current question when newer comparable evidence exists.
We add a visible note for a meaningful error in a number, source, method, credit, or explanation. The note says what changed and when. We can fix small spelling, punctuation, format, or access issues without a formal note when the meaning stays the same.
In short. These changes keep the calculator from showing more detail than the data supports. The technical entries below say exactly what changed.
July 24, 2026 — uncertainty and provenance closeout. The data receipt now shows source-input uncertainty separately from uncertainty about how data from different sources may overlap. Neither range is a final confidence interval. Census survey error from replicate weights is still not calculated, and exact state ranks remain hidden. We also downloaded the official ACS and NHANES source files again. They reproduced the checked-in model exactly. Their SHA-256 values are recorded and checked by the data manifest.
July 24, 2026 — sparse effective-tail guard. Model 2026.07-evidence-v3 keeps the repeatable normal model by sex and age. But it no longer gives an exact height result when the modeled tail has fewer than five effective NHANES observations. It shows a one-sided limit: “1 in X or rarer.” This uses an approximate 95% Wilson upper bound based on effective sample size. The calculator does not show exact comparisons after that limit. This fixes the false precision allowed by v2 at extreme heights, even though the underlying decade groups had only about 215–639 effective observations. The bound is a stability guard. It is not a complete NHANES design-based confidence interval.
July 22, 2026 — measured height model. Model 2026.07-evidence-v2 replaced the percentile-table-only lookup. It used sex- and age-specific normal distributions fitted to measured NHANES examination records with the CDC examination weight. It removed the retired arbitrary tail floor. It did not yet include the sparse-effective-tail guard now used in v3.
July 22, 2026 — height-tail guard. The calculator previously extended the CDC table beyond its published 5th–95th percentile range. It used a 0.2% lower floor in the upper tail. That rule made false plateaus, including an exact-looking women’s six-foot result. Model 2026.07-evidence-v1 removed that extension and carried documented bounds through the result.
July 22, 2026 — Pool Index label. We changed “most balanced” to “highest combined estimated share.” The geometric-mean score measures the size of both selected groups together. It does not measure equality.