Get the extensionGet the app Support us Sign in

Corrections

Where the engine makes a judgement we cannot defend, it goes here. This ledger is public and append-only: entries are amended with a dated change log, never quietly edited or removed. It has two parts: the mistakes we found in our own engines’ work, and what readers have filed.

Anyone may file against a specific lexicon entry or a specific score — no account needed. Useful reports name the article, the word or the number, and say what the right answer would have been.

What kind of problem is it?

Mistakes we found ourselves

Errors our own engines made in stories we published, with what was wrong, what we did about it, and the changelog entry that records the fix. 13 so far, newest first. They are kept in our code, not in the database, so they stay as written.

FixedAreaWhat went wrongWhat we didChangelog
6 October 2026 Pictures A photograph we found for a story, rather than the outlet’s own, was shown when its caption did not name the story: a British coach on a story about a boat collision in Congo, a picture captioned “Placeholder” on a story about robots, an American campus security car on the court case of Andrew Mountbatten-Windsor. A found picture is shown only where its own caption names what the story is about. It was removed from every page already published, as well as from new ones. Pictures that name the story, and summaries that follow the report
6 October 2026 Summaries A summary was printed that stopped in the middle of a clause (“…weigh up the cost of keeping.”, “…predicting 140.”), ran two sentences of the report together, or had a later sentence about people it never named (“One of the women…”). A summary was not written again when the outlet changed a headline or rewrote its report. A summary that fails those checks is not printed on a card, a story page, a list, the Brief, Good news, an alert or an email; the next best is used, or none. A summary is written again when the outlet changes a headline or its report. Pictures that name the story, and summaries that follow the report
6 October 2026 Story grouping The Newtown stabbing went out under the one headline its other newsrooms least agreed with, which read as though the woman arrested had been the one attacked. The Daily Brief said a player had reached the semi-finals the morning after he reached the final, because the rounds of a competition had been grouped as one story. A story is named by the headline its newsrooms agree with most. The stages of a competition (quarter-final, semi-final, final) are separate stories, and a story grouped earlier is shown by its newest stage. One event, one story: the stages of a tournament kept apart, places and sections put right
6 October 2026 Places A street named after a city was placed in the city (Sydney Road is in Melbourne), as was a place used only as a reference point (“halfway between Sydney and Melbourne”), and a story about somebody’s Australian links was placed in Australia. Those are no longer placed. Stories stored before the change keep their old place until they are placed again. One event, one story: the stages of a tournament kept apart, places and sections put right
6 October 2026 Mood Mood read a seismologist’s “shock” as tabloid drama, read the words in “no reports of injuries or damage” as though somebody had been hurt, and counted what may happen as heavily as what did. An earthquake report in which nobody was hurt read 0, grief. A seismologist’s “shock” is not drama, “no reports of injuries or damage” reads as the relief it is, and what may happen counts half as much as what did. Terms kept whole, plain words only where they fit, and every reading in one unit
6 October 2026 Religious framing The reading counted a faith word even where it was a name. A lawyer called Christian, “Mr Christian” and a Buddhist association were read as somebody’s faith named beside a crime, and a story about a man accused of spying read 70. A faith word counts only where it names a person by their faith. Terms kept whole, plain words only where they fit, and every reading in one unit
6 October 2026 Rewriting “Tedesco in tears after colossal huge game” was published as “Tedesco in tears after colossal large game”: a plain word was swapped in beside a word that already said the same thing. A word the phrase already says is taken out rather than doubled, and “deadly” becomes “fatal” only for an event somebody died in. Terms kept whole, plain words only where they fit, and every reading in one unit
6 October 2026 Rewriting Our version took the first word off terms that are names of things, and “Man shot dead by police, critical incident declared” was published without “critical”. The same rule applied to “critical condition”, “serious injuries”, “major crime”, “dangerous driving”, “deadly weapon”, “rare earths”, “fully franked” and “grand final”. Those terms are kept whole. Terms kept whole, plain words only where they fit, and every reading in one unit
5 October 2026 Rewriting A loud verb was swapped for “criticised” even where it was the event. The West’s “Cars torched in Taylors Hill driveway”, a firebombing, was published as “cars criticised”, and in a summary “erupted in flames” was published as “began in flames” and “is under fire after” as “is after”. The same fault made “A man was bashed” read “criticised”, “savaged by a dog” read “criticised by a dog”, “battered to death” read “affected to death”, “Fire gutted the home” read “reduced the home” and “Mount Etna erupted” read “began”. A verb with a physical sense is replaced only where what it acts on shows the figurative one (a plan, a policy, a government). Fixed on 5 October; it is recorded here, and in the changelog, from 9 October. Corrections lists our own mistakes, and the trust pages say only what the code does
5 October 2026 Rewriting “Hit out at the report” was published as “criticized at the report”, which is neither good English nor Australian spelling. It reads “criticised the report”, and every word we put in is spelled the Australian way on the site, in the extension and through the API. Fewer loaded words in our headlines and summaries
5 October 2026 Rewriting Our version took out “apparently”, “seemingly”, “presumably” and “arguably” as if they were decoration. They say how sure the reporter is, so “The Australian man apparently killed himself” was published as “killed himself”: a claim the report had not made. Those words are kept, and so are “hardly”, “barely”, “virtually” and “practically”. Words that say how sure a report is are kept
5 October 2026 Religious framing The reading’s highest scores were on reports where faith was the subject, such as a racist tirade against a Muslim schoolgirl on a Sydney bus, which is the opposite of what it is for. “A mass shooting” and “a towering mass of water” were read as the Catholic Mass, and a “fitness guru” as Hinduism. A victim’s faith, a place, a festival and a word about prejudice no longer count, and the ordinary words that happened to be on its list are gone. Language Load uses the corrected reading. Religious framing that reads what it is meant to
5 October 2026 Sections and places A story was filed under Culture because “artificial intelligence” contains “art” (an OpenAI story about a Medicare hack), and “transport” put one in Sport. A column by a British MP for Newcastle upon Tyne was placed in Newcastle, New South Wales. The Victorian earthquake was shown with a picture of Christchurch. Sections are matched on whole words, a story from abroad is given no Australian place unless it says Australia, and a picture whose caption names another place is not shown. Stories in the right place, and Good news that is good

Filed by readers

No reader has filed a correction yet. That is a statement about the record so far, not a claim that there is nothing to find — the limits we already know about are listed on the method page, and selection bias is one we cannot detect at all.

What belongs here

A company that removes other people's editorial judgement and hides its own errors has no argument left. This page is the argument.