Corrections

The policy

When we get something wrong, we say so on this page, under a name, with a date. We do not delete the error and we do not quietly edit it away. Corrections stay up.

A site that adjudicates other people’s claims has no standing to hide its own mistakes.

How we handle them

Factual error — anything that changes what a page asserts. Corrected in place, marked at the point of the error, and logged here.

Verdict change — where new evidence changes a conclusion. The original verdict stays visible, with the reasoning that replaced it.

Typos and broken links — fixed without a log entry.

Where an error appears in a video, the video’s description carries the correction. We do not re-cut published audio, and we say plainly when the audio still contains the error.

One rule that came out of the first correction

When any number in a published piece turns out to be wrong, every number in that piece gets re-checked against its source — not only the one that was flagged.

We did not do that the first time. The second error survived the first round of corrections precisely because nobody had questioned it.

Found one?

Bring the paper, or bring the discrepancy. The comments on the relevant episode are the fastest route. Corrections that come from readers are credited unless the reader asks otherwise.

Log

3 September 2026 — Episode 1, Your Watch Sees It Before You Do

Three figures were wrong or unsupported. All three are ours.

1. The claim appeared in three podcasts, not two. The narration and the original description said the claim showed up in two of the channels we scanned, independently. Our own scan records four occurrences across three distinct channels: Feel Better, Live More (twice), The Rich Roll Podcast, and Dhru Purohit.

2. We scanned 30 channels, not 34. Thirty-four channels were on the list. Four never resolved and were never scanned. The number we published was the size of the list, not the size of the scan.

3. A total we cannot support. The video cites 2,179 collected claims for that scan. We cannot reproduce that figure from our own records, and our working notes from the same week record a different one. The figure has been removed rather than replaced with a guess.

The 168 claims specific enough to test, the 45 sources, and the verdict are unaffected. None of the three errors touches the evidence.

The description now carries all three corrections. The audio has not been re-cut and still contains all three.

What changed because of it. The third error was the serious one — not because the number mattered, but because we could not audit it after publishing it. Our scans used to write into a single running file, so each run overwrote the last run’s totals. They now write an immutable, dated snapshot per run, kept separately from the running stock, and no figure about our own scanning gets published unless it can be traced back to the snapshot that produced it. The two earlier runs were reconstructed from what survived, and are labelled as reconstructed rather than measured — with the parts that could not be recovered left empty.

The four channels that never resolved were also diagnosed: three had changed their handles, and one had already been fixed in our configuration after those runs. All four are corrected now.