Solutions · Media + journalism

Published journalism
is permanent.
Corrections aren’t.

A statistic that doesn’t survive a statistician’s Twitter thread. A source whose conflict of interest wasn’t surfaced. An investigative claim whose primary document has an alternative interpretation the reporter didn’t consider. These are the corrections that define a publication’s credibility. Augle runs the adversarial review before publication — so corrections become optional rather than inevitable.

Investigative reporters stress-testing evidence before publication
Editors reviewing statistical claims and source independence
Science journalists translating research findings for public audiences
Data journalism teams validating analysis before publication
Session configuration
Guardian modeEditorial integrity
DepthStandard · Deep
DocumentsDraft articles · research papers · data analyses · source documents
SVS checksSource independence · statistical misrepresentation · primary vs. secondary source · press release vs. peer review
OutputFinding · confidence grade · unresolved objections · audit trail
The problem

The story is right. The statistic in paragraph three isn’t.

Most journalism corrections are not about the central claim — they’re about a supporting statistic that was misrepresented, a source whose independence wasn’t adequately verified, or a research finding that was drawn from a press release rather than the actual study. These are the errors that don’t change the story but define how the story is remembered. Augle catches them before publication.

Questions journalism teams run on Augle

Does the statistic in this paragraph accurately represent what the cited study actually found?
Is the research finding we’re reporting from the peer-reviewed paper or the press release, and do they say the same thing?
Does this source have a conflict of interest that isn’t disclosed, and what is the strongest argument against their position?
What is the most credible counter-argument to the central claim in this investigation, and have we addressed it?
Are any of the studies cited in this science story subject to replication failure or retraction?
Statistics that misrepresent the underlying study

Editorial integrity mode monitors statistical claim accuracy throughout the session. Relative risk presented without an absolute baseline, a correlation stated as causation, a subgroup finding presented as a main effect, or a percentage change calculated from an unrepresentative base — each is identified before it enters the published article. These are the statistics that statisticians correct on social media the morning after publication.

Press release findings that diverge from the actual study

The Guardian’s SVS distinguishes press releases from peer-reviewed publications and checks the original source against the claim being made. A press release that overstates the study’s findings, omits the sample size limitation, or drops the confidence interval — and that the reporter has relied on without reading the original — is flagged. The paper says something different from the press office.

Sources with undisclosed conflicts of interest

Editorial integrity mode monitors source independence throughout the deliberation. A quoted expert who has received funding from an industry the story concerns, a think tank whose positions are funded by a party with a direct interest in the story’s framing, or a ‘independent’ report that was commissioned by the subject of the investigation — these are flagged for editorial consideration before publication.

The strongest counter-argument you haven’t addressed

The Contrarian takes the role of the most credible critic of your story’s central claim — surfacing the alternative explanation for the evidence, the methodological objection to the central thesis, the data that cuts against the narrative. A story that has been tested against its strongest counter-argument, and has an answer for it, is a story that holds up. One that hasn’t been tested doesn’t know whether it holds.

How it works in practice

Three sessions.
Three journalism moments.

Each scenario illustrates realistic deliberation behaviour across investigative journalism, science reporting, and data journalism.

Use case 01 of 03 · Media + journalism
The Investigative Pre-Publication Review
Senior Investigative Reporter · National newspaper · Pre-publication legal and editorial review
Standard depth · document ingestion

Does the evidence in this investigation into a major housebuilder’s planning application practices support the central claim, and what is the strongest counter-argument the subject’s legal team will advance?

CartographerSettled: the planning application approval rate at the three local authorities is statistically anomalous. Contested: whether the anomaly is explained by the housebuilder’s site selection strategy — they may systematically select sites with pre-existing planning support. Unknown: whether comparable housebuilders operating on equivalent sites have materially different approval rates.
ContrarianStrong objection (Unresolved): "The subject’s legal team will argue that the anomalous approval rate is a function of site selection, not undue influence — that they choose sites where approval is already likely. Without a controlled comparison to comparable housebuilders on comparable sites, the statistical anomaly is consistent with both the story’s thesis and the alternative explanation. The story currently doesn’t have an answer for this."
MethodologistConstruct validity: the approval rate comparison uses all planning applications in those local authorities as the baseline. If the housebuilder systematically submits applications for sites with strong pre-existing planning merit, the correct baseline is comparable sites, not all applications. The current methodology doesn’t control for this. Confidence: Contested.
PragmatistTwo options: (1) Obtain comparable approval rates from other major housebuilders in the same LPAs on comparable site types — if the anomaly persists under controlled comparison, the story is stronger. (2) If the comparison can’t be done before deadline, reframe the statistical finding as unexplained anomaly rather than implied influence, and present the alternative explanation in the article. Either is more defensible than the current framing.
Session output
FindingContested — statistical anomaly established but alternative explanation not controlled for
Key gapNo controlled comparison to comparable housebuilders on comparable sites
Legal riskSubject’s legal team has a credible site-selection alternative explanation
Next stepObtain comparable data OR reframe as unexplained anomaly with alternative presented
The value

The alternative explanation was the legal team’s entire pre-publication letter. The reporter had considered it but the article didn’t address it directly. A reframed statistical claim with the alternative explanation presented and addressed turned a legally vulnerable story into a defensible one — without changing the central finding.

Use case 02 of 03 · Media + journalism
The Science Story Review
Science Editor · Weekly magazine · Pre-publication fact check on a clinical research story
Standard depth · document ingestion

Does our draft story on a new study linking ultra-processed food consumption to cognitive decline accurately represent what the study actually found, or are we overstating the findings?

Guardian SVSPrimary source check: the draft cites the press release headline (‘Ultra-processed foods linked to 28% higher dementia risk’) and the peer-reviewed paper. The press release figure is a relative risk reduction. The paper’s Abstract states absolute risk: 0.68% vs 0.53% per year. The draft uses the relative figure without the absolute baseline. Editorial integrity mode flags: statistical misrepresentation — press release vs. paper discrepancy.
MethodologistTwo additional issues: (1) The study is observational — the draft uses causal language (‘causes’, ‘leads to’) not supported by the study design. (2) The study population is Brazilian adults aged 35–74. The draft’s headline implies general population applicability. The paper’s Discussion explicitly cautions against generalisation to other populations.
ContrarianModerate objection: "The lead researcher’s institution receives significant funding from a food industry body that has an interest in the study’s findings being treated as non-generalizable. The story doesn’t mention this. A critical reader will find it. The disclosure should be in the article, even if you assess it doesn’t affect the findings."
Session output
SVS flagRelative risk cited without absolute baseline — press release vs. paper discrepancy
FindingDraft overstates study findings on three dimensions
Edits neededAdd absolute risk · remove causal language · qualify population scope · disclose funding
The value

The press release was accurate about the relative risk — it just omitted the absolute baseline, the population scope caveat, and the causal language limitation. Four targeted edits turned a story that would have generated a correction from the study’s authors into one they could endorse.

Use case 03 of 03 · Media + journalism
The Data Journalism Review
Data Journalist · Digital outlet · Pre-publication analysis review on housing affordability
Standard depth · document ingestion

Does our analysis showing that housing affordability has deteriorated faster in regional cities than in London hold up to scrutiny, and what is the most likely methodological challenge?

MethodologistConstruct validity issue: the analysis uses median house price to median earnings as the affordability ratio. In regional cities, the denominator includes a higher proportion of part-time and lower-wage earners than in London, where high earners are more concentrated. The ratio may reflect labour market composition differences rather than pure affordability deterioration. This is the methodological challenge a housing economist will raise.
ContrarianModerate objection: "The ONS publishes an affordability ratio that uses lower quartile earnings and lower quartile house prices specifically to avoid the labour market composition issue. The analysis uses median-to-median. If the lower quartile ratio shows the same pattern, the finding is robust. If it doesn’t, the choice of denominator is the story’s central vulnerability. This should be checked before publication."
PragmatistRun the same analysis using ONS lower quartile affordability data. If the regional vs London pattern holds, the story’s conclusion is supported by two independent measures and the methodological objection is addressed. If it doesn’t hold, the finding needs to be reframed. Either outcome is better than publishing a median-to-median analysis without knowing whether the ONS measure agrees.
Session output
FindingContested — methodology may conflate affordability with labour market composition
Key gapLower quartile ONS measure not run — this is the standard comparator for the central claim
Next stepRun ONS lower quartile analysis before publication. If consistent, story is robust. If not, reframe.
The value

The ONS lower quartile analysis took two hours. It confirmed the regional vs London pattern — the finding was robust under the more rigorous measure. The story was published with both measures cited, which pre-empted the methodological objection and made the analysis stronger, not weaker. The alternative was publishing a median-to-median analysis and receiving a correction request from a housing economist.

How Augle works for journalism

The adversarial reader’s
review. Before publication.

1
Submit your draft and source materials

Upload the draft article, supporting research papers, data analyses, and source documents. Editorial integrity mode activates — the Guardian checks source independence, distinguishes press releases from peer-reviewed papers, monitors statistical claim accuracy, and flags original source vs. secondary report discrepancies before any agent receives the evidence.

2
The ensemble maps what’s established

The Cartographer classifies every claim as Settled, Contested, or Unknown based on the available evidence. The Methodologist evaluates whether the evidence supports the conclusions drawn — whether causal language is justified by the study design, whether the population scope applies to the story’s framing, whether statistical representations are accurate relative to the original source.

3
The adversarial reader’s objections are run

The Contrarian takes the role of the story’s most credible critic — a subject’s legal team, a statistician on social media, a rival journalist, or a hostile expert. Every objection specifies a resolution condition. Unresolved Strong objections appear verbatim. These become the pre-publication edit list, not the post-publication correction notice.

4
You receive a confidence-graded evidence record

The Synthesizer produces a finding anchored to the evidence — not the story’s thesis. Confidence grades per claim, unresolved objections verbatim, and reopen conditions. The audit trail is exportable — for legal pre-publication review, editorial governance, or the public interest defence records that regulated journalism requires.

Journalism session · configuration
Guardian mode
Editorial integrity — source independence monitoring, statistical misrepresentation detection, original vs. secondary source distinction, press release vs. peer review classification
Document types
Draft articles · Research papers · Press releases · Data analyses · Source documents · Leaked materials · Expert reports
Contrarian focus
Legal team counter-arguments · statistical objections · alternative explanations · source independence challenges · methodological weaknesses
Output package
Confidence grade per claim · unresolved objections verbatim · SVS record · exportable audit trail for legal and editorial governance
Session depth
Standard for features and science stories · Deep for major investigations with expert review at Phase 1/2 boundary
Why Augle for journalism

Corrections become optional.
Not inevitable.

Catches statistical errors before publication

Editorial integrity mode monitors every statistical claim against the original source — relative risk without absolute baseline, correlation framed as causation, subgroup findings presented as primary results, percentage changes from unrepresentative bases. These are the errors that generate corrections from academic statisticians within 24 hours of publication. Catching them before publication is the entire value.

Runs the strongest counter-argument before publication

The Contrarian surfaces the most credible objection to your story’s central claim — the alternative explanation a legal team will advance, the methodological objection a rival journalist will raise, the conflict of interest a reader will find. Unresolved objections appear verbatim with resolution conditions. A story that has been tested against its strongest counter-argument is a story that holds up under scrutiny.

Produces an auditable evidence record

Every session produces an exportable record — SVS verification outcomes, confidence grades per claim, every objection raised and its resolution status. For regulated journalism, high-stakes investigations, and pre-publication legal review, this record demonstrates that the evidence base was reviewed to the standard that the story’s claims require. It’s also the record that matters if a correction is later disputed.

Publish stories that hold.
Not ones that correct.

Join the waitlist and run a session on your next investigation before it goes to publication.