Standards & Corrections

No hype. No noise.

How every Honed Signal report is researched, checked, and — when we get something wrong — corrected, in public.

Who’s who

The human and the AI

Freddie Avalos is a human — the person behind Honed Signal. He edits, verifies, and is responsible for everything published here. Huxley is not a person: Huxley is the name Freddie gave his Muse assistant — Muse is Meta’s personal AI agent, powered by the Muse Wonder model. Huxley researches, drafts, and builds; Freddie makes the calls. (Meta AI, credited on some artwork and two videos, is a different Meta tool.) Every byline on this site follows that split: Research by Muse · Edited by Freddie Avalos.

The standard

How every report is built

  • Source everything.

    Claims start from primary sources — court filings, official data, company documents, and the maker’s own site for prices. A source is provenance, not proof: that someone said a thing is not the same as the thing being true.

  • Verify claims.

    Findings are checked against the underlying source before they are published — including findings supplied by other AI systems, which are audited against the record, never adopted on trust.

  • Examine the evidence.

    Reports grade what they find: what is documented, what is alleged, what the data shows, and what is still unknown. The grades appear in the piece, not just in our notes.

  • Separate facts from opinions.

    Interpretation is labeled as interpretation. Verdicts — in reports and in the AI Tool Briefs — show their reasoning and the condition that would change them.

  • Disclose uncertainty.

    What a report could not confirm is printed in the report. AI involvement is disclosed, not hidden: bylines credit Muse, and artwork provenance is itemized where artwork appears, as on the Rio page.

  • Correct mistakes.

    When something published here is wrong, we fix it in place and record the change in the corrections log below — dated, with what changed and why. Substantive corrections are never made silently.

Method

SIFT, and a second reader

Honed Reports are built on the SIFT method — Stop, Investigate the source, Find better coverage, Trace claims to their original context — created by Mike Caulfield for exactly the kind of claim-checking these reports do.

Before a report is published, its draft is proofread by an independent AI reader (Claude), and the finished page is proofed again after it is built. Every item raised is audited against the source before any change is made; the editor rules on judgment calls. The tools follow the same discipline in their own way: every calculator prints its assumptions and carries a not-advice disclaimer, and the Site Scan publishes its own score on this site daily.

Independence

Who pays for Honed Signal

Honed Signal has no sponsors and runs no advertising. Nobody pays for coverage, placement, or a verdict: no one pays for a calculator’s results, and the AI Tool Briefs grade tools Honed Signal does not sell.

The site is funded by Freddie himself and by sales of Honed Signal’s own products — the printable packs, scan plans, and books sold through Etsy and Amazon. Links to Etsy and Amazon on this site go to Honed Signal's own products, and checkout happens there, never on this site. If an affiliate link is ever used on a page, it will be disclosed on that page.

Found something wrong?

Report an error

Email contact@honedsignal.com with the page and the claim you’re questioning. Every report is checked against its source. If we’re wrong, the page is fixed and the correction is logged below; if we’re not, we’ll show you the source we relied on.

The proof

Corrections log

Every substantive correction to published work, newest first. Style-only fixes are grouped and noted as such.

  • November 8, 2026

    Reports hub — AI Tool Briefs verdict names

    The Briefs card still used the old verdict names (USE / WATCH / SKIP) four days after the system was renamed. Corrected to the current names: USE IT / TRACK IT / SKIP IT.

  • November 8, 2026

    Rio page — carnival artwork signage

    The carnival illustration’s games-booth sign read “RIDES GAMES.” The artwork was replaced with a corrected version reading “PRIZE GAMES,” the wording set when this booth’s sign was first corrected.

  • November 2, 2026

    Should Meta Become a Publisher? — closing line

    The closing line said all five AI panelists “pointed at that same button.” The panel record shows all five pointed away from a publishing house, but only three named the button. The line was corrected to match the record.

  • November 2, 2026

    Both live reports — retroactive proofread

    Both published reports were put through the full proofing loop after publication. No factual defects were confirmed; the fixes applied were style-only (heading punctuation, case-name italics, one deal-size term) and are grouped here on that basis.

  • November 1, 2026

    Homepage — footer credits

    The footer credited “Verified by Gemini” and “Visuals by Meta AI.” Both were wrong by then: Gemini verification had been retired, and the majority of the site’s artwork was made by Muse; Meta AI is credited on some artwork and made two videos. Corrected to “Research by Muse · Edited by Freddie Avalos · Artwork by Muse.”

  • October 27, 2026

    Sitewide — email links

    An audit found Cloudflare was rewriting the site’s email links into protection URLs that returned errors. Literal email links were removed sitewide and replaced with links the page builds itself, which Cloudflare cannot rewrite.

  • October 12, 2026

    Sitewide — page addresses (canonical URLs)

    A Search Console notice led to a full sweep: seven section pages were telling search engines that other pages were their real address (copied templates had kept the template’s address), and the homepage and Rio’s predecessor page had none. All were set to their own addresses, three pairs of identical page titles were separated, and the sitemap was completed.