How we record facts and compare tools

One registry, one source per fact, one date

The site is generated from a single data file. Each tool has an entry holding its plans, its list prices, the limits the vendor publishes, the URL we read them on and the date we read it. Pages never contain a price as text: they contain a token, and the generator replaces it at build time with the value from the registry.

That is a deliberate constraint rather than a technical detail. It means a page cannot quietly drift away from its source, a stale figure cannot survive in a paragraph somebody forgot to update, and the build refuses to publish if a page tries to hard-code an amount. The source ledger prints the whole registry: every tool, the exact page we read, when we read it, and the caveat we recorded on that source.

What we compare, and on what criteria

Comparisons use the same four criteria every time, in the same order, so two pages can be read against each other.

Where two tools bill in units that cannot be compared, we say so instead of forcing a conversion. A per-mailbox monthly fee and a pack of test credits are not the same kind of number, and adding them into one figure would invent a comparison that does not exist.

What we refuse to do

• No causal claim from a before and after. If placement improved while a tool was running, that is a sequence, not a cause. Everything else about your sending changed at the same time. Our pilot protocol exists precisely because a plain before and after proves nothing.
• No deliverability guarantee. Nobody can offer one, and a vendor that does is describing its own hope.
• No invented test. A headline promising a test only appears when a test exists. Sample reports we wrote ourselves are labelled as fictional on the page that shows them.
• No unattributed vendor claim. If a figure comes from the vendor's marketing rather than from an independent observation, the page says so.

The place of AI

Logos and product screenshots on this site are always the vendors' own published files: we never generate a screenshot, a dashboard, a placement report, a test result or a brand mark. The page headers, the section covers and the illustrations of the guides are generated by AI and redrawn to one style, as a visual aid only; guide illustrations say so in their caption. They never show a real product, a real price, a real inbox or a deliverability outcome, and they prove nothing about any tool.

Where our knowledge stops

Being useful on this topic means being explicit about the edges. These are ours.

• We have run no warmup campaign and hold no mailbox, so nothing here is first-hand operational experience with these products.
• Prices are list prices from public pages. Negotiated, annual, promotional and enterprise pricing is not visible to us, and some vendors quote on request.
• Vendors change plans and limits without notice. The ledger date tells you how fresh a figure is; it does not promise the figure is current today.
• Sending limits at Google and Microsoft are taken from their own published documentation, which does not describe every account type or every tenant policy.
• One list verification vendor could not be read because its site refused our request, so it is absent rather than estimated.

Corrections

If a figure here is wrong, the fastest fix is to tell us where the right one is. Write to contact@whichwarmup.com with the page and, if you have it, the vendor URL that contradicts us. A sourced correction changes the registry, and the change propagates to every page that uses the figure at the next build. We do not negotiate corrections with vendors: the vendor's own public page wins.