Editorial standards

No sponsored placements. No gifted-review obligations. No manufacturer input on scores, verdicts, or rankings. No display advertising of any kind. The site earns Amazon Associates commissions when readers buy through our links, and those links appear only on products we actually recommend.

Scores and verdicts are set before any affiliate link is placed and never change after. Products we evaluate and reject carry no buy link at all, which is the structural proof the policy is real: we leave commission on the table on every rejected product.

Every affiliate link is direct and uncloaked, carries nofollow sponsored semantics, and hands the reader to the live Amazon listing, which always shows the current price, availability, and promotions. We display no LIVE prices today and no stock claims, because Amazon owns that data and keeps it current: every price on this site is the exact amount from our fact-checked research record, labeled as such with a link to the live listing, never a real-time feed. The day the hourly Creators API sync activates, the surfaces it feeds upgrade to a "Live price" label (refreshed at least hourly, always falling back to the fact-checked price), so every labeled price states exactly which it is. The star ratings beside recommendations are the customer-rating evidence from the same research record, shown as proof of demand, and the live listing owns the current count.

The one tracking tag is a single configuration value, not baked into pages, so it can be swapped or retired without touching content. If the Associates relationship ever ends, links stop earning but keep working, and the site keeps publishing.

Report an error by email with the page and the claim. We verify every report against the underlying source record, and material corrections follow this procedure:

1. Verify the report against the sourced claim and its origin.
2. Fix the content, update the score if the underlying fact moved it, and correct the record so the fix propagates to every referencing article.
3. Record the correction in the machine-readable record, flagged as a correction (THE FRESHNESS LAW: the byline's Last-verified date moves with the re-checked record, so the stamp always tells the reader the truth about the fix).
4. Reply to the reporter by email confirming the fix.

Typographical fixes land silently; factual corrections are always recorded. We do not delete criticism or quietly reverse verdicts.

AI assists drafting under a strict contract: drafts are assembled from the verified research records into the site's fixed templates, anything malformed is rejected, and every batch passes a human review gate. The drafting step never touches the templates, the scores, or the facts, and nothing publishes unreviewed.

The architecture is deliberately AI-agnostic: any AI, or a human with a text editor, produces the same schema and therefore the same rendered pages. If the AI dies tomorrow, the site keeps publishing.

Every page labels which of these it is:

• Research-verified scoring: verified spec records, independent review corroboration, and the fixed rubric. The standard tier for new category launches.
• Hands-on bench testing: named tester, the protocol printed on the page. Labeled explicitly whenever it happens.
• Community-reported: owner reports and forum evidence, labeled as such and weighted accordingly.

We never label a desk review as hands-on testing. The tier is on the page, and the how-we-evaluate page carries the full protocol.