Editorial policy

The evidence ledger

Every factual claim in a Cernary article is a row in a ledger before it's a sentence in the article. Each row records the claim, the source it came from, the type of source (manufacturer page, manual, independent test, news report, owner forum), the date we retrieved it, and how confident we are in it. The article marks each claim inline with a ledger number (E·01, E·02...), and the full ledger sits at the bottom of the page. If a claim can't be traced to a source, it doesn't go in the article.

Sourcing

We prioritize sources in roughly this order: the manufacturer's own spec pages and manuals, independent instrumented tests where they exist, established review outlets, and owner reports on forums and retailer reviews when a pattern shows up across many independent accounts. A single forum post never becomes a claim on its own; a recurring pattern across dozens of unrelated posts can, and we say so plainly when that's the level of evidence behind a claim.

When sources disagree, we don't average them and move on. We name the disagreement, say which source we weighted more heavily, and explain why. Marketing copy on its own is never treated as evidence for a performance claim; a manufacturer spec sheet is treated as evidence for what the manufacturer states, not as proof of real-world performance.

Verification pipeline

A claim starts as a draft entry: source, quote, retrieval date. It is marked unverified until an editor reopens the original source and confirms the quote still matches what's published, at which point it flips to verified. We also actively look for evidence that contradicts a claim before publishing it, not just evidence that supports it; where we found real counter-evidence, the article says what it is instead of hiding it.

Claims carry a review date. Specs get revised, firmware changes, prices move and companies get acquired or shut down products, so a claim doesn't sit unchecked forever; it comes back up for re-verification on schedule.

How AI is used to produce this site

Cernary's production pipeline is human-directed and AI-assisted, and we'd rather explain that plainly than have you guess at it. An editor decides which product or comparison to cover and defines the scope. Sources are gathered and the evidence ledger is built from documents we can point to: manufacturer pages, manuals, published tests, forum threads. An AI drafting step turns that sourced ledger into prose, matched to the ledger entries it's written from. A human editor then reviews the draft against the ledger, edits it, and is the one who publishes it.

What this means in practice: the facts in an article are sourced and checked the way described above, regardless of which step wrote which sentence. What it does not mean: nobody at Cernary has physically unboxed, run or timed the products in an article marked research-based, and the prose never claims otherwise. Every article carries a testing status badge for exactly this reason. We do not test products hands-on, and no article claims hands-on results.

Independence

Articles disclose any affiliate links at the top of the piece and next to each link; see the affiliate disclosure page for the current arrangement. No manufacturer, retailer or PR contact gets to see or approve a verdict before it publishes, and none pays for coverage.

Corrections

We get things wrong sometimes; a supplier changes a spec quietly, a reader spots a transcription error, a source turns out to be less reliable than we thought. Read the full corrections policy or send what you found to contact@cernary.com.