Standards

How we report

Every story is built to be checked. This page shows how a story is made, the live record of every check we have run, the corrections we have issued and the rules you can hold us to.

The process

From the show floor to the page

AI drafts and checks. A named person approves. These are the steps a field story goes through, with the models that do each AI step today.

  1. Recorded with consent

    A reporter on the show floor records an interview only with the speaker’s consent.

  2. Transcribed

    Speech-to-text produces a timed transcript, so every quotation can be traced to the second it was said.

    Model: microsoft/mai-transcribe-2 via OpenRouter

  3. Researched

    Background and figures come from official sources, and every claim carries its citation.

    Model: openai/gpt-6-luna via OpenRouter

  4. Written, then checked

    One model drafts. A model from a different company checks every sentence against its evidence. Quotations must match the recording word for word.

    Model: anthropic/claude-opus-5.5 via OpenRouter; checked by openai/gpt-6-sol via OpenRouter

  5. A person approves

    A named person reads the checked draft and approves it or sends it back. AI cannot publish.

Verification ledger

Every check, as it happened

Each entry comes from the newsroom’s own records and links to the published story. Approvals carry the name of the person who made them; meet them here. Private interview material, notes and drafts never appear.

quotes verified against recordings
7
stories approved by a named person
6
stories published from show floors
57
corrections issued
0

As of , read from the newsroom database when this page loaded.

Corrections log

What we got wrong

No corrections issued yet.

Report a correction

Tell us the story and the specific claim. We treat accuracy complaints as reporting work.

Editorial standards

The rules we hold ourselves to

Provenance

Facts, Judgments and Decisions stay separate

Everything we publish is built from three things that are recorded differently and never collapsed into one field.

  • Facts are source-backed observations or records: an interview, a document, a published statement, something our reporter saw.
  • Judgments are assessments, ours or a model’s, stored with the model, version and confidence where those exist.
  • Decisions are explicit outcomes, made by a named person or by a stated policy, of what we do next.

When you read a Stratawire story you can tell which is which: original reporting is labelled Original reporting, and the sources behind a story are listed with it.

Interviews

How quotations are handled

Interviews are recorded with the participant’s consent, and what ends up in quotation marks is checked against that recording before publication. Filler words may be removed for readability; words are never changed.

Translated quotations are checked against the original language, and material uncertainty is flagged rather than guessed.

We never invent interviews, quotations, firsthand observation or access. AI never creates a person’s words. It can transcribe and translate what a person actually said, and nothing more.

Tools

AI use and model disclosure

Real reporters work with AI staff. AI assists research, transcription, translation, drafting and checking. A named person decides what is published; AI cannot publish.

For material AI work we record the provider and model used, and we list every AI job with its model and provider on the newsroom page. Our decision model, Jev, advises on judgment calls such as source sufficiency and review priority. It is not a writer and not an oracle, and its output is advice, never an edit.

Media

Synthetic media is never documentary evidence

We do not use AI images. Synthetic imagery, audio or video would be labelled for exactly what it is and never presented as a photograph, recording or footage of a real event.

Real media carries its provenance: who captured or supplied it, on what basis we may use it, and what was done to it. Where we have no photo with rights, we use a graphic labelled “Graphic: The Stratawire”, never a picture passed off as a photo.

Accountability

Corrections

When we get something wrong, we fix it prominently and record the fix. Corrections are auditable, not quietly overwritten: the record says what was wrong and what changed, and the count appears in the ledger on this page.

To request a correction, use the contact page and name the story and the specific claim. Accuracy complaints are treated as reporting work, not as public-relations work.

Independence

Conflicts of interest and disclosures

Commercial relationships, personal connections and any interest that could bear on a story are disclosed, and where a conflict is material the person steps back from the piece.

No commercial relationship buys coverage. The separation rules for partner content below apply to every arrangement, without exception.

Verification

Accuracy checks, not approval

We may check factual claims (names, titles, figures, technical detail) with the people we interviewed. Checking is verification, not approval: interviewees do not see copy before publication, do not sign off stories and cannot veto what we write.

Commercial

Partner content separation

Partner and sponsored material is published in its own clearly labelled lane: a “Partner content” label, its own visual treatment, and placement apart from newsroom reporting.

Partners have no influence over editorial coverage, and partner content is never presented as reporting, interviews or original observation. Newsletter and sector-briefing sponsorship is stated within the item it supports. The formats and the full rules are on the advertising page.

Partner-content formats and the full separation rules are on the advertising page.