Methodology for Certified Mail delivery benchmarks

PostalForm's delivery-benchmark protocol starts carrier transit at a verified USPS acceptance event and ends at a delivered event. Payment, provider creation, printing, and delivery attempts are recorded separately. Orders without reliable acceptance or delivery evidence remain visible as missing or incomplete observations rather than being silently counted as fast deliveries.

Use one physical Certified mailpiece as the unit. Record the calendar quarter of USPS acceptance in UTC, the extraction cutoff, destination state, selected receipt option, and fulfillment route. Keep replacements separate from their original pieces and report the rule used to handle duplicate provider records.

Include paid, non-test Certified mailpieces with reliable carrier records. Report exclusions for canceled-before-mailing pieces, duplicate records, missing service classification, and missing or contradictory timestamps. An order can contain more than one physical piece; order counts and mailpiece counts are not interchangeable.

Retain raw event names in the protected analysis source and document their normalized classification. Do not reinterpret “in local area,” “processed for delivery,” “label created,” or “in production” as delivered or USPS accepted. Preserve event occurrence time separately from webhook receipt time.

Report the total cohort, accepted pieces, delivered pieces, attempted-only pieces, returned pieces, and pieces still unresolved at extraction. A delivered-only median must be labeled conditional on delivery and accompanied by the unresolved share. It must not be described as the expected delivery time of every mailing.

For time-to-event estimates across the full cohort, right-censor unresolved observations at the extraction cutoff and publish the estimator and assumptions. Keep returned mail visible as a competing outcome; do not treat it as successful delivery. Report calendar hours or calendar days consistently, with a separate documented calendar if publishing business days.

State summaries require at least 30 accepted pieces in a state and service stratum under this protocol. This is an editorial minimum, not a guarantee of statistical precision. Suppress smaller cells, include sample counts and uncertainty, and avoid league tables that imply causal state or provider performance from different document or customer mixes.

Before release, reconcile mutually exclusive outcome totals, duplicate counts, negative intervals, timezone conversions, and raw-to-normalized carrier stages. Verify a sample against the underlying carrier chronology. Do not fill missing acceptance times from an order's creation date or a provider's estimated mailing date.

Publish only aggregate state/service cells meeting the threshold. Exclude names, street addresses, ZIP codes, order identifiers, tracking numbers, emails, payment references, document contents, filenames, and individual event histories. Keep the raw export access-controlled; publish the methodology, aggregate data, extraction window, and code revision instead.

Each completed release must identify its cohort dates, extraction cutoff, source systems, event mapping version, code revision, exclusions, counts, limitations, and correction history. A review should confirm whether the result supports a descriptive observation, an estimate, or no conclusion. This protocol does not promise a quarterly release when the data quality or sample size is insufficient.