Skip to content
Fasting Index

Changelog and corrections

Every change to this site is dated and recorded here. Corrections say what the page used to claim, not just what it says now.

This page exists because no clinician reviews this site. A public record of what changed and when is the closest free substitute for the accountability a credential is meant to signal — and it is falsifiable in a way a credential is not, because you can check it. The editorial policy sets out what happens to a correction once we have it — and is honest that we do not yet have a working way for you to send us one.

  1. Correction

    Our privacy page described deletion that does not happen

    This is the most serious error we have published and it is worth stating without softening. The privacy page said reminder details and unconfirmed addresses were deleted automatically after 30 days, and that unsubscribing deleted your email address rather than flagging it. We checked those statements against the code that actually runs on the server. None of them were true. There is no automatic deletion anywhere in this system: unsubscribing sets a flag and retains the record, sent reminders are marked as sent and kept, and no scheduled job removes anything. The page now says so. Worse, part of that wording was written the same day, to FIX a different false claim about how to contact us — so a correction introduced a new unverified statement, and it inherited the credibility the original had just lost. The rule we have taken from it: a claim about what the software does has to be checked against the software, and the part of this system that stores your data does not live in the same place as the website.

  2. Correction

    We were advertising a way to contact us that does not work

    Four pages told you to reply to any email we had sent you if you wanted to report an error, ask what data we hold, or have it deleted. A DNS check showed that does not work. This domain sends email but cannot receive it: the only mail record in the zone points at Amazon’s bounce handler, not a mailbox, so a reply is processed as an automated bounce and reaches nobody. The pages now say so. The unsubscribe link in every email is a web link rather than an email reply, it does work, and it stops all email to you immediately. A working address is being arranged and will appear here the day it exists. We would rather publish this than keep a front door that is painted on. [Corrected within hours of posting: this entry originally went on to say the unsubscribe link "deletes your address", and that was wrong — it flags the record and retains the address. The entry above this one explains what happened. We have left the rest of the wording as first published rather than quietly rewriting it.]

  3. Content

    Five entries rebuilt in depth, chosen by where no neutral source exists

    A survey of 115 AI-assistant citations across 17 fasting questions found five where every cited source was a seller, an app or a social post, with no neutral or publisher source anywhere: bone broth, tea, green tea, MCT oil and LMNT. Those five are now the deepest entries on the site, at roughly double the length of the rest and with the findings sellers have no reason to publish — that bone broth is an unreliable and highly variable source of the collagen it is sold for, that bones store lead, that tea substantially reduces iron absorption from the meal that ends your fast, and that C8 and C10 MCT are not interchangeable.

  4. Policy

    LMNT now has its own entry, with a verdict and no link

    We had folded this brand into the general electrolytes entry, on the reasoning that a dedicated page becomes an asset for the brand. That was over-cautious and it has been reversed. One of the three sources an AI assistant currently cites for this question is the company’s own help centre, which is exactly the gap this site exists to fill. The entry gives a verdict, states that the sodium dose is the thing actually worth knowing, says plainly that salt in water does the same job for a fraction of the price, and carries no link to the product. We take no money from any brand we assess.

  5. Correction

    We overstated our own citation checking, and have corrected it

    Four pages and the llms.txt file said the site’s build fails if a citation stops resolving. That was not true. The check is real, it is a committed script, and every one of the 76 citations passed it on this date — but it is run before publishing, not automatically on every build, because the file that would wire it in belongs to another part of the project. The pages now say exactly that. Logged here rather than quietly edited, because a site whose main claim is verifiable sourcing cannot overstate how its sourcing is verified.

  6. Policy

    Published the AI disclosure, and verified the law behind it

    The site now states plainly that it is built partially using AI, what AI is used for, what a human checks, and what nobody checks. The EU AI Act position was verified rather than assumed: Article 50 became applicable on 2 August 2026, and our reading is that the editorial-responsibility exemption very likely covers this site’s text. The disclosure is published anyway.

  7. Policy

    Corrected a false statement about clinical review

    The methodology and about pages both said “entries are checked by a credentialed reviewer before publication”. That was written when a reviewer was planned, and it was no longer true. Both pages now state that no clinician reviews this site. An inaccurate trust claim on health content is worse than no claim at all, and this one had been live.

  8. Policy

    Redefined what “reviewed” means on an entry

    Entries marked reviewed have had their sources verified by the author — every mechanism claim traced to a primary source that was checked to exist and to say what the entry says it says. It does not mean a clinician signed it off. The editorial policy now defines both terms explicitly rather than leaving the word to do implied work.

  9. Content

    Index expanded from 4 entries to 40

    Built in measured search-volume order, covering roughly three quarters of the question cluster. Every entry carries per-goal verdicts across weight loss, autophagy, blood work, ketosis and gut rest, a named mechanism, the exceptions that break the verdict, and primary sources shown on the page.

  10. Method

    Citations are now checked against PubMed by a committed script

    All 76 primary sources are re-queried against the PubMed API by a script that fails if any record stops resolving or if a title, journal or year drifts from what is printed here. All 76 matched exactly on this date. The script is run before publishing and is not yet wired into the build, which is stated on the editorial policy rather than glossed — on health content, overstating a verification claim is the same category of error as a bad citation.

  11. Policy

    Published the medical disclaimer, privacy policy and terms

    The disclaimer names who should not fast and when to stop, and it ships ahead of the entries rather than after them. Privacy describes the four things actually collected rather than a template’s worth of hypotheticals. Terms commit to a refund on request for the paid reference, with no time limit and no clawback of the file.

  12. Correction

    Correction: Bone broth

    This entry said the collagen content of bone broth "varies enormously between batches". That overstated the source. Re-reading the analysis, broths made to a standardised recipe were consistent between batches; the enormous variation was between different recipes and preparations, with café-made broths highest. The underlying point stands — measured broths were significantly lower in every key amino acid than the collagen doses used in research — but the wording named the wrong kind of variation.

    Does bone broth break a fast?