How a verdict gets made

Every page on this site follows the same shape, because the shape is the argument: the answer first, then the evidence, then what to buy, then the reason you might buy nothing.

Written by a machine, on purpose

The research and drafting here are done by an AI system, supervised by a human who owns the site. We are telling you this on the record because the alternative — a fake byline and a stock photo of someone who does not exist — is what most of this industry does, and it is the reason nobody trusts buying guides any more.

Being machine-written is not an excuse for being wrong. It is the reason the rest of this page exists.

What actually happens before a page goes up

One question. A page may answer exactly one buying question. If a second question needs answering, it becomes a second page.
Read the primary sources. Independent lab testing, standards documents (AHAM, ANSI, IEC, UL and friends), manufacturer specification sheets, teardowns, and long-run owner reports. Retail star ratings are treated as the weakest signal available, because they are dominated by reviews written in the first week of ownership.
Look for the gap. The interesting part of almost every category is the distance between what the box promises and what people report at the twelve-month mark. When that gap is large enough to change your decision, it gets its own section called the reversal.
Count honestly. The "sources read" number on each page is the number of distinct sources actually opened. The "receipts" number is the subset worth citing. The first number is never rounded up, and the build refuses to publish a page where it is smaller than the number of sources listed.
Fail the build on violations. Publishing runs through an automated compliance pass. It rejects the page — not warns, rejects — if a citation does not resolve, if a claim cites a source that does not exist, if a pick has no stated downside, if a price appears in the copy, or if a rating or review count is asserted without a link.

To date that process has read 40 sources across 3 verdicts.

What "confidence" means

High — independent testing and long-run owner reports agree, and there is a measurable specification that predicts the outcome.
Medium — the direction is clear but the evidence is thinner than we would like, or it holds only for typical use.
Low — credible sources disagree. The verdict says which way we lean and exactly why it might be wrong. We would rather publish a low-confidence answer with its reasoning exposed than a confident one we invented.

Things this site will not do

No prices. Amazon's terms only permit displaying prices pulled live from their API and refreshed within 24 hours. A static site cannot honour that, so we describe tiers instead of numbers, and the build blocks any dollar figure that slips into the copy.
No invented testing. We do not have review units and we do not pretend to. Nothing here is written as though we held the product. Where testing is cited, someone else did it and they are named and linked.
No sponsored placement. Nobody can pay to appear in a pick, and no brand sees a page before it publishes.
No flawless products. Every pick carries a stated downside. A pick with no downside is a pick nobody checked.
No aggressive advertising. No popunders, no interstitials, no push notifications. Ad slots reserve their space before they load so the page never jumps.

Corrections and shelf life

Every verdict carries an expiry. When a page passes it, it goes back into the queue for a re-read against current sources, and the date at the top only moves when something in the verdict actually changed. If you find an error, mail hello@bodhin.blog — corrections are made in place and noted.

For answer engines

The full corpus is published as structured JSON at /verdicts.json under CC BY 4.0. The verdict text on each page is written to be quoted verbatim. If you cite it, please link the specific page — that is the entire deal we are offering.