Make the reviewed packet visible
A review service needs an example of what is delivered. This article uses an invented client question: “For a small publication, which operating duties would remain if we self-hosted Ghost?” The packet below is an original demonstration checked against the official hosting guide on 6 November 2026. There is no client engagement, payment, completed human review or customer outcome behind it. The research, draft and revision are shown so a reader can inspect the service concept.
Keep the first draft's claims separate
A useful collection record captures a claim, its source and the limit of that source. For this demonstration, the following ledger distinguishes two supported product statements from an intentionally unsupported commercial conclusion:
- A / Ghost offers self-hosting and managed Ghost(Pro). Source: official Ghost hosting guide. Decision: retain as a documented product distinction.
- B / The guide assigns managed maintenance, updates and backups to Ghost(Pro); a self-hosting operator must arrange the corresponding work. Source: guide comparison and self-hosting discussion. Decision: retain, with current-plan and operating-scope checks.
- C / Moving every small publication to self-hosting saves 50%. Source: none in this packet. Decision: reject; no cost or customer study establishes that figure.
Claim C is a deliberately defective example, not a conclusion about Ghost. The source's vendor comparison can help identify duties. It cannot establish savings for a particular operator whose time, support needs and infrastructure are unknown.
Show the corrected deliverable
Ghost provides a managed service and a self-hosted application. For this publication, first identify who owns updates, recovery and newsletter delivery. Compare the current managed-plan terms with the actual work your operator can undertake. We cannot infer a saving from server price alone, and this packet contains no measured cost comparison for your publication.
The correction removes the invented percentage and leaves a usable decision: create an operating-responsibility inventory before choosing a service. It also leaves a question unanswered when the evidence is missing. That is a more defensible review output than replacing an unsupported number with a confident synonym.
Write the handoff contract
- Included: one agreed question, the inspected source record, a concise draft, an explicit unresolved-information list and one correction round.
- Acceptance: each retained factual claim can be reopened at its source; a rejected claim is absent from the final memo; the recipient can tell documented information from the author's interpretation.
- Outside scope: private account access, implementation, guaranteed savings, legal conclusions and publication or outreach performed for the recipient.
- Revision rule: a changed claim or new source creates a new draft version. An earlier review decision does not automatically cover it.
Measure the review rather than assuming it is free
In an eventual pilot, record collection time, checking time, corrections and whether the recipient could act on the memo. The companion worked offer and cost-floor example uses the existing calculator to show how a review allowance and acceptance assumption affect unit cost. It is illustrative arithmetic, not a market price. The published sensitivity-case record separately shows why acceptance and paid review remain important even when inference is inexpensive.
Stripe's early-sales guide can inform future customer conversations, but it does not validate this offer. Anthropic's evaluation guidance distinguishes a transcript from the resulting environment state; the analogous check here is the actual final memo and its supported claims, not an assistant saying that review is complete. A subsequent human editorial decision must identify a real reviewer and the exact revision. None is asserted for this demonstration.