Corrections

Our editorial policy says that substantive corrections are noted on the page itself rather than edited away quietly. This is the log of them. Every entry gives the date, what was wrong, how long the error was visible, and where the correction now sits.

A correction log with nothing in it is not evidence of accuracy. It usually means the publisher has not looked, or has looked and decided not to say. We would rather this page grew.

What counts as a correction

Three kinds of change get an entry here and a dated note on the page itself:

• A factual claim that was wrong. A described platform behaviour that the platform does not have, a number that was not what we said it was, a citation that did not support what we hung on it.
• Something displayed to readers that was untrue, including metadata — a publication date, a review date, an author credit.
• A claim we made about ourselves — about how the site is funded, how it is written, or what it promises — that the site was not actually keeping.

Three kinds of change do not get an entry, because logging them would bury the ones that matter: fixing a typo or a broken link, rewording a passage that was correct but unclear, and adding a source to a claim that was already accurate. Those are ordinary editing.

The log

2 September 2026 — the Telegram absolute, finished on the second attempt

What every affected page now says: Telegram has no per-post ranking, which is the narrower claim every argument on this site actually needs, and it does run a similar-channels recommender whose documented input is overlap between subscriber bases (Telegram API documentation). The nine pages are the Telegram growth guide, members explained, premium members, auto views, best times to post, Premium for channel owners, stories and boosts, the Instagram followers guide and Instagram reach. Eight of them now carry a dated note saying what they used to say, which is what the rule at the top of this page promises and what neither earlier attempt delivered.

The method belongs in the entry, because “we looked again” is what the previous two entries also said. The whole corpus was searched for each form the absolute takes — no ranking system, ranks nothing, delivers everything, no discovery algorithm, no recommendation of any kind, documents no ranking system, no ranking step at all — and the search was re-run afterwards with the correction boxes stripped out first, so that a note quoting the old wording could not be mistaken for the old wording surviving. Outside this page, what the second search returns is: the qualified form (“no per-post ranking”) on seven pages; two statements scoped to posts that are exactly true as written, one saying Telegram documents no ranking system for channel posts and one saying it documents no ranking of individual posts; one line that quotes the old phrase in order to argue against a misreading of it; and one that quotes this site’s own shorthand, “Telegram delivers everything to everyone”, in order to say what is wrong with it. No unqualified absolute survives.

2 September 2026 — four counts this site published about itself, all stale

Folding 45 documents into 31 falsified every hardcoded count on the pages that describe this site, and not one of them was updated in the same commit. The disclosure page said “twelve of our forty-five pages” carried a commercial link. The about page said “39 of our 45 pages” carried a review date, and repeated the twelve-of-forty-five figure a section later. This page said we cannot re-check “forty-five pages” every week. All four were false the moment the merge landed, and they were visible for the rest of that day.

What makes this the second occurrence rather than the first is the 26 August entry below. It explains that the about page’s review-date sentence was rewritten precisely because a hardcoded count “would be false again the moment a new page went up” — and then hardcodes a replacement, which went stale exactly as predicted, seven days later. So the about page no longer carries a number for it: it says what a review date means and leaves you to see which pages show one. The disclosure page does still carry a count, because a disclosure that will not say how much of the site is commercial is not much of a disclosure. It now reads seven of thirty-one.

Seven rather than ten because three commercial cards were removed the same day, and the reason is the same in each case: the card sat where the page had just finished arguing against the thing it links to. One was on bot starts, whose own conclusion is that a purchased bot start is the weakest product in the catalogue. One was on auto views, which sold the same service a reader had followed a link from post views to read about, so the same purchase was put in front of them twice. One was on how to grow a channel, which tells a new owner in as many words not to buy subscribers at that stage. Boxing a paid link under a verdict against the product is a pattern rather than an accident, and it is not one this site can argue is compatible with the rest of what it publishes.

26 August 2026 — three promises this site was not keeping

An internal review found three statements about Novamya’s own practice that were not true as written.

The about page said “every article carries the date it was published and the date it was last reviewed”. Only 23 of the 45 pages then on the site carried a review date. Thirty of them were re-read against their sources in the course of this review, so the figure stood at 39 of 45 on the day this entry was written — the counts in this paragraph describe the corpus as it was on 26 August 2026, before the consolidation to 31 documents. But the claim was false when it was written and would be false again the moment a new page went up, so the wording changed too. It now says what is actually the case: a review date appears once someone has read the page back against the platform it describes, and a page without one has not been re-checked since publication. We considered the alternative — stamping a review date on every page — and rejected it, because a review date applied in bulk to make a site look maintained is the exact practice the editorial policy disavows two paragraphs further down.

The same page said commercial links sit “inside the text where they are relevant”. They do not. They render in a bordered box set apart from the text. The description now matches what is on the page.

The disclosure page repeated the same inaccuracy, describing the link as sitting inside the body text. Corrected there too.

26 August 2026 — Telegram views are not counted once per account

Telegram Post Views — the most-read page on this site — stated without qualification that “views are counted once per account”. That is only true for about four days.

Telegram’s own channel FAQ says: “After a short while (around 4 days), Telegram will forget that you’ve seen a post and will count you again.” So a view total on an older post is not a count of unique readers, and the gap widens the longer a post stays useful. The consequence matters for anyone comparing posts: a two-month-old post and yesterday’s post have not had the same opportunity to accumulate re-counts.

The section has been rewritten and now quotes Telegram directly rather than paraphrasing it. The page also now carries Telegram’s own admission that these figures are approximate, which is unusual for a platform and worth knowing.

26 August 2026 — an invented finding on the social proof page

Our page on the psychology of social proof — since folded into the Instagram followers guide, which now carries the two experiments it was built on — claimed that “the difference between 50 followers and 2,000 is enormous” while “the difference between 20,000 and 60,000 changes almost nothing”, and presented it as a research finding. It was not. No study was cited because none was consulted; the numbers were invented, and they sat on a page whose own excerpt promised readers “what the research supports”. It was visible for 21 days.

The passage is deleted. The page now rests on two real randomised experiments — Muchnik, Aral and Taylor’s 2013 study in Science, in which a single positive vote raised a comment’s final rating by about 25 per cent, and Salganik, Dodds and Watts’s 2006 study, in which 14,341 participants showed that social influence makes outcomes both more unequal and less predictable. Where the page still reasons beyond what those studies establish, it now says so in the sentence rather than in a footnote.

This is the worst kind of error a site like this can make, because inventing a number and dressing it as research is the specific behaviour the rest of the site spends its time warning readers about.

26 August 2026 — eight claims about platform behaviour that the platforms contradict

Every page asserting how a platform works was read back against that platform’s own documentation. Eight assertions did not survive. All eight were stated as fact and none of them had a source, which is how they lasted three weeks.

• “Telegram posts are not indexed for global search.” They are. Telegram’s API reference documents channels.searchPosts searching “globally in all public channels”, including channels the searcher has never joined. The honest qualifier — which we now give — is that the search is metered: Premium accounts get an allowance, after which each search costs Stars. So it is a real discovery surface, not a plannable one. The same page also claimed hashtags only work inside a chat you are already in; the same method searches them globally.
• “Telegram stories do not forward.” This was the central argument of our stories page, and it is wrong. Telegram documents both sharing a story into a chat, where it renders as a forward from the original poster, and reposting it with a header naming the source. There is even a flag whose purpose is to prevent forwarding, which would be unnecessary if forwarding did not exist. The section has been rebuilt around what is actually true of stories — they expire — rather than around a mechanism we had wrong.
• “Telegram has no recommendation of any kind.” Half wrong, and the half that was wrong mattered. Telegram has no per-post ranking, which is what the argument needed. But it does run a channel recommender, and its documented input is overlap between subscriber bases. That is now stated wherever the absolute used to be — and it makes cross-promotion, not keywords, the real discovery lever on Telegram.
• “Buying members breaches Telegram’s terms of service.” It does not. We re-read the full English Terms: the words “artificial”, “inflate”, “fake”, “engagement” and “followers” do not appear in the document. Telegram bans spam and scams generally, and its advertising guidelines forbid advertising engagement services on Telegram Ads, which is a different thing. Meta, YouTube and TikTok all do prohibit buying engagement explicitly, and we now quote each of them. The asymmetry is real and no other guide states it.
• “YouTube counts a view after roughly thirty seconds.” No such threshold exists in YouTube’s documentation. YouTube says views “are counted the moment a video starts to play across all formats”, and that it separately discards low-quality playbacks during validation. The thirty-second figure is folklore.
• “Instagram’s ranking system does not read your follower count.” Instagram’s own ranking explainer names follower count among the signals it uses for Reels. Corrected on both pages that said otherwise.
• “Instagram counts a view on a short display threshold.” Meta publishes no duration threshold for a view — only that a view is a time the content “was played or displayed”. Removed rather than guessed at.
• “Instagram impressions” as a live metric. Meta deprecated the impressions metric for all API versions on 21 April 2025, replacing it with views. Our reach-versus-impressions explanations were describing a metric that no longer exists. They now describe reach and views, and note the retirement — which is also worth knowing when a panel offers to sell you “impressions”.

26 August 2026 — six citations that did not say what we attributed to them

A separate pass checked whether the sources we did cite supported the claims hanging on them. Six did not.

• Meta’s Transparency Center page on ranking was cited on three pages as Meta’s account of how Instagram ranks content. That page describes Facebook Feed and does not mention Instagram anywhere. Replaced with Instagram’s own ranking explainer.
• “None of the platforms publish weightings.” TikTok does, qualitatively: it states that a strong indicator of interest, such as finishing a longer video, “would receive greater weight than a weak indicator”. It is the only comparative statement any platform makes, and we were asserting it did not exist.
• “Hashtags are a weak signal, and Instagram has said as much.” Instagram has not said as much. Hashtags are simply absent from its ranking descriptions, which is an argument from silence — a reasonable one, but not a statement by Instagram, and the page now says which of the two it is.
• Engagement rate presented as a platform metric on four pages. Meta publishes no engagement-rate metric and no formula; the figure is an industry convention built from primitives Meta does publish. Every page that implied otherwise now says so, which is a genuine difference from the competing articles that print a formula as if it were official.
• A claimed waiting period before an attacker’s two-step password can be reset. We could not find this anywhere in Telegram’s documentation — the word “reset” does not appear in the relevant FAQ at all. Removed and replaced with the routes Telegram does document.
• A TikTok support URL cited on two pages now redirects to a page with no readable content. Replaced with TikTok’s newsroom description, which is where the substance moved.

Two of these are worth separating from the rest. Miscounting a metric is an ordinary mistake. Citing a page about Facebook as evidence about Instagram, and asserting that no platform publishes something one of them does publish, are failures of the specific discipline this site claims as its reason to exist. They are logged in the same detail as the others because a correction log that goes easy on the embarrassing entries is not a correction log.

22 August 2026 — three blog posts showed a publication date from before this site existed

The three posts in the blog displayed publication dates of 14, 22 and 29 July 2026. All three were written and published on 5 August 2026. The July dates were placeholders that were never true.

They were visible for seventeen days. We noticed on 13 August and changed the dates in the source files the same day, but the correction did not reach the site: the script that publishes these files read the date field when a page was created and ignored it on every later update. The fix on 22 August was to the script, and the dates changed with it.

A dated note sits at the top of the one post still at its own address, What Changed in Social Platform Policy This Year. The other two were folded into the guides they belonged to on 2 September 2026 — what an SMM panel is and the Telegram growth guide — with their old addresses redirected, and their notes went with the pages. A dated note about a former page’s publication date, sitting at the top of a guide that never carried that date, would confuse more than it corrects; the record of it is this entry.

This one is worth stating plainly rather than filing quietly. A site that spends most of its words telling readers to check whether a claim is traceable had three pages carrying a date that was not, and the policy promising that corrections get noted on the page had never once been used. Both of those are now fixed.

If you find something wrong

Send it to editor@novamya.com, or use the contact page. A correction does not need to be a formal complaint — a line saying “this paragraph does not match what the app does now” is enough, and is genuinely useful, because platform behaviour changes without announcement and we cannot re-check thirty-one pages every week.

If you tell us something is wrong and we conclude it is not, we will say so and explain why, rather than quietly not replying.