Corrections log

31 August 2026 · Coffee nap

I described a study arm that does not exist, and a nap length the paper never states

Previously: “Twelve sleepy participants… under different conditions: 200 mg of caffeine alone, a 15-minute nap alone, or caffeine immediately followed by the nap, against placebo.”

This page had been carrying a note saying its figures came from secondary descriptions and still needed checking. I have now read the published abstract of Reyner and Horne (1997), Psychophysiology 34(6):721-5, PMID 9401427. Two things were wrong.

First, there was no nap-only condition. The abstract states plainly that the non-nap comparisons were caffeine only and placebo. I had invented a third arm, which also made the claim “it beat either component alone” unsupportable, because the study never tested a nap on its own.

Second, the nap was not stated to be 15 minutes. The abstract gives a 30-minute break containing the caffeine and the nap, and never gives the nap's length. The under-15-minute figure appears in the abstract's opening sentence, describing the same group's earlier work as background. I had lifted a number out of a paper's summary of prior research and attached it to the study itself.

What survived checking, and is now sourced to the abstract rather than to secondary descriptions: twelve participants, 200 mg of caffeine, a two-hour monotonous afternoon drive in a simulator, incidents reduced to 9% of placebo for the combination against 34% for caffeine alone, and the finding that naps consisting of “nonsleep dozing” still helped.

The page now describes the design correctly and carries a note explaining the error in place. The full text is paywalled and I have not read it, so the page asserts nothing the abstract does not state. This is the same mistake as the NASA percentages: not an invented source, but a claim about what a source establishes.

31 August 2026 · Privacy

The privacy page said there was no analytics on this site. There was, and it had been there all along

Previously: “There is no analytics package, tag manager or session recording on this site. Nobody here is measuring what you read.”

Cloudflare Web Analytics runs on every page. It is injected by Cloudflare at the edge rather than written into this site's code, which means it does not appear anywhere in the source, and does not appear if you fetch a page with a command-line tool. It appears only when a real browser requests a real page.

That is how it went unnoticed through two previous reviews of this page, including one earlier the same day. Both times the check was made by reading our own files, and the claim being made was about what visitors' browsers actually do. Those are different questions, and only the second one matters to a reader.

The page now names it, along with the five Google domains that AdSense contacts, in a table produced by loading the site in a browser and recording every outside request. Two things about the Cloudflare beacon were verified rather than assumed: it sets no cookies, and it writes nothing to local storage.

Logged by: the author. Found by measuring the live site rather than reading its source, which is the only method that would have found it.

31 August 2026 · Privacy

The privacy page said this site carried no advertising, for a day after advertising went live

Previously: “Cookies. MySleepMate sets no cookies. There is no consent banner because there is nothing to consent to.” And: “Third parties. There are none.” And: “There is currently no analytics, tag manager, advertising pixel or session recording on this site.”

Google AdSense was added to every page on 30 August 2026. The privacy page was not updated until the 31st, which means that for a day it told readers three things that were no longer true: that no cookies were involved, that no third party received their visit, and that there was no advertising pixel on the site.

The page had explicitly promised the opposite. It said that if an advertising pixel were added, the page would be updated to name it and say what it collects “before or at the same time as it goes live”. That is the promise that was broken, and it is worse than the factual error, because the commitment was the thing readers were being asked to trust.

The page now names Google AdSense, states plainly that loading any page causes the browser to contact Google and hands over IP address, user agent and page address, links to Google's own controls and to industry opt-outs, and carries a note about having been late.

Logged by: the author, prompted by a review of whether the site met AdSense's own requirements. Nobody reported it.

31 August 2026 · How Long Should A Nap Be

“155 minutes” corrected from a duration of benefit to the study’s last measurement, and an unsupported replication claim withdrawn

Previously: the 10-minute nap “improved alertness, fatigue and cognitive performance immediately on waking, with benefits lasting over two hours”, and the study “has not been replicated at scale”.

Two problems, both found by reading the paper itself rather than a summary of it. In Brooks & Lack (2006), 155 minutes is when the testing stopped, not how long the benefit lasted. Post-nap assessments ran at 5, 35, 95 and 155 minutes, with sleep-latency measures at 65, 125 and 185. Some of the 10-minute nap's effects were still detectable at that final assessment, which means the study cannot say when they faded, only that they had not yet. The page now says exactly that, and carries a callout warning against the reading it previously invited.

Separately, the claim that the study “has not been replicated at scale” was withdrawn outright. No literature search had been done to support it. Saying a finding has not been replicated is an evidential claim like any other, and this site should not make one it has not checked.

The same pass added the study's design, which turned out to be stronger than first described: 24 healthy adults, within-subject with counterbalanced order rather than five separate groups, after nights restricted to five hours, with nap conditions timed to precise minutes of actual sleep rather than time lying down.

Logged by: the author, from the full text. Caught the day the page was published, before it had been indexed.

31 August 2026 · Sleep Numbers

NASA nap study: the 34% and 54% figures re-attributed, and the “26-minute nap” framing corrected

Previously: “The widely quoted outcome figures are roughly 34% improvement in performance and 54% in physiological alertness against the no-rest group”, with the references section listing the NASA technical memorandum as the source of those figures.

On checking the memorandum itself, Crew Factors in Flight Operations IX, held by the US DOT National Transportation Library: those two percentages do not appear as stated results. What the report prints are counts: 34 microsleep events in the rest group against 120 in the no-rest group across the final 90 minutes of flight, and 81 psychomotor vigilance lapses against 124. The percentages are a later formulation of the same work, repeated onward through secondary coverage.

The entry now reports the counts as the study's findings, states plainly that the percentages are not printed in the report, and describes the protocol correctly: a 40-minute scheduled rest window in which pilots slept a mean of about 25.8 minutes, rather than “a 26-minute nap”, which described a measurement as though it were a designed dose.

Also corrected in the same pass: the CDC figure now quotes the brief directly and notes that the CDC measures sleep per 24-hour period rather than per night. That figure, 30.5% of US adults under 7 hours in 2024, NCHS Data Brief No. 559. Was checked against the brief and is unchanged.

Logged by: the author, after reading the primary source. The page had been live with the weaker attribution since 31 August 2026.

16 August 2026 · Privacy

“Nothing is saved to local storage” narrowed to cover the website only

Previously: “None of it is transmitted to a server, stored in a cookie, saved to local storage, or associated with you.” written when this site existed only as a website.

The Android app adds a bedtime reminder, and to raise a reminder it has to remember the time you chose, the warning you asked for, and whether it repeats. Those three values are written to local storage on the device.

The sentence remains accurate for the website in a browser. It was not accurate for the app, so the privacy page now carries a section describing what the app stores rather than leaving a claim in place that had quietly stopped being true. No change to what the calculator itself does, and nothing new is transmitted anywhere: the reminder is scheduled by the phone's own alarm service, with no server and no push notification involved.

Logged by: the author, on release of the app rather than after the fact.