Dataset grain
Each full-fidelity row identifies provider, route, amount, standard or promotional mode, pay-in, pay-out, collection method, rate, fee, recipient amount, confidence, quote time and archive date.
The unique storage key also includes collection method, so publication analysis deduplicates multiple sources before comparison.
Known coverage imbalance
Provider totals differ by orders of magnitude, and some providers stop earlier or cover a focused corridor set. Raw observation counts must never be used as a popularity or quality ranking.
Pay-in and payout coverage are also uneven. Missingness is displayed rather than imputed.
Validity controls
Published comparisons require positive recipient amounts, known methods, supported confidence and exact quote matching. Promotional and standard observations remain separate.
Independent mid-rate fields are used only when present. Quoted delivery estimates are never described as measured arrival performance.
Timeliness and reproducibility
The catalogue and live quote system continue to update. Every report freezes its period and check date so a later total does not silently alter the article.
Readers can use the public statistics export, and the repository analysis script reproduces the selected August study.
How this result was produced
Eligible rows used high or medium confidence and an API or HTML source. We retained one provider row per exact key using confidence, source quality and recipient amount, then required at least two providers.
These are quote observations, not completed transfers. Results apply only to the named routes, dates, amounts and methods; they do not predict a future quote or measure delivery reliability.
No average of incompatible currencies, routes or methods is used. No missing value is converted to zero.
