How our claims are verified

This site makes checkable claims on purpose. This page is the method behind them — what we survey, how we source comparisons, and how the honesty is enforced.

Facts verified 2026-07-19 — corrections: hello@emailfast.dev

The claims ledger

Every load-bearing factual statement on this site is a numbered entry in an internal claims ledger, each carrying the exact evidence that backs it — a code path, a config value, a test, or a policy page. Pages reference claims by id, and the site build resolves them. The build fails if a page cites a claim the ledger doesn't recognize, or uses a phrase we've told it we may not honestly make (for example, a certification we don't hold or a metric we have no live history for). The marketing cannot ship ahead of the product because the tooling won't let it.

The platforms we surveyed

We reviewed these twelve against their own public pages in July 2026:

Mailchimp · Substack · beehiiv · Buttondown · Kit · Loops · SendGrid · Mailgun · Postmark · Resend · EmailJS · Paubox.

Two of our claims lean on this survey, and they range over different subsets — so we're explicit about the scope of each:

• "None offers customer-held encryption keys" (the keys claim). A negative statement true across all twelve — none of them, whether a newsletter tool or a delivery API, offers customer-held keys for message content.
• The rented-delivery claim. This does not say every platform we surveyed rents its delivery — several of the twelve (SendGrid, Mailgun, Postmark, Resend) are the delivery infrastructure other senders rent from. It names specific consumer newsletter platforms whose own pages disclose an upstream provider — beehiiv (SendGrid), Buttondown (Mailgun/Postmark), Substack (Mailgun) — and mirrors each source exactly.

Our per-platform comparisons — honest about where each of them wins — are on the comparison pages.

How comparative claims are sourced

• Mirror the document, don't infer. Each comparison is phrased to say what the source page says. "Buttondown's subprocessor list names Mailgun and Postmark" mirrors a document; we avoid inferences dressed as facts.
• Scope and date are load-bearing. Every comparative claim states the survey scope (the twelve above) and its verification date, so "every" can never quietly overreach.
• Evidence captured. Each comparative claim will be backed by an archived snapshot of the source page, captured and kept alongside its ledger entry, so a claim can be re-checked against exactly what the source said the day we verified it.
• Quarterly re-verification. Comparisons decay as platforms change, so they're re-checked each quarter and re-dated. When one is wrong, our corrections policy says how fast we fix it.

How "open tracking that refuses to lie" is measured

open tracking that refuses to lie: privacy-proxy prefetch “opens” are detected and not counted as human reads. Concretely: a privacy proxy (Apple Mail Privacy Protection, corporate scanners, link-prefetchers) fetches a tracking pixel automatically, without a human ever opening the message. We classify an open as machine-generated when it carries a known proxy/bot signature or lands within a short window of the send (before a human plausibly could), and we exclude those from human open counts.

Its error bounds, stated honestly. This is best-effort classification, not certainty. A new proxy we haven't catalogued can slip a machine open through until we add its signature; an unusually fast genuine reader near the send can be over-excluded. We tune the window and signatures from real traffic and would rather under-count a human open than report a robot as a reader — the opposite of the industry default.

Endorsements and material connections

We publish no customer quotes today. When we do, any quote or endorsement from a customer who received free or discounted service (for example, an early design partner) will disclose that connection plainly — because the honesty is the point, not an embarrassment. That commitment is adopted now, before the first one ships.