Tuesday, 27 October 2026 Updated hourly from 226 outlets
Pricing Sign in

How Oldspeak works

Every outlet's headline on the same event, side by side, and a reproducible score for single articles.

How scoring works

A model answers the seven questions below from 0 to 3 for one article, citing evidence for each. Code then turns the answers into a score from 1 to 10.

  1. S
    Add up questions 2 to 7S runs from 0 to 18. Question 1, the reporter's own voice, is kept apart because it decides the worst band on its own.
  2. D
    Editorialising in its own voiceQuestion 1 is 2 or more: the narration itself takes a side. Score 8 to 10.
  3. C
    One-sided, no editorial voiceQuestion 3 is 2 or more, and question 2, 5 or 6 is too. Score 6 or 7.
  4. A
    Straight reportingQuestion 1 is 0 and S is at most 1. Score 1 or 2.
  5. B
    Localised faultsAnything else. Score 3 to 5, by S.

Method. Bias types follow the Table of Media Bias Elements and the Media Bias Taxonomy (both CC BY 4.0). The impartiality rules follow the BBC Editorial Guidelines and the Ofcom Broadcasting Code. The score is calculated from the seven answers, never generated.

What this is not. No outlet gets a standing bias rating, and there is no left/right axis, because that axis does not describe Bangladeshi politics. Each analysis covers one article about one event. Members' analytics count how each outlet's headlines were labelled over a chosen period, with the number of headlines behind every share, and also show the average score of an outlet's scored articles: that average describes those articles (mostly each story's lead), not the outlet, and outlets are never ranked by it or by any share.

Known limits. Headlines are grouped into stories automatically, so an unrelated headline sometimes slips in; it is shown faded rather than hidden. Where outlets carry the same wire copy, the comparison has little to show. Human annotators agree on bias in Bangla news only moderately (κ = 0.574 at best), which limits what any automated score can claim.

The seven questions

Rubric 1.0.0. Each answer is the level that fits the article, from 0 to 3.

  1. G1
    Reporter's own voice
    1. Narration is descriptive throughout.
    2. One isolated evaluative word in the reporter's voice.
    3. Recurring evaluative language in the reporter's voice.
    4. The narration itself argues or attacks.
  2. G2
    Headline
    1. Reports the event or the speech act; contested claims are attributed.
    2. Compresses or intensifies, but the claim stays attributed.
    3. Asserts a contested claim in the paper's voice.
    4. Asserts a contested claim and adds the paper's own evaluative words.
  3. G3
    Other side missing
    1. Every party involved responds substantively.
    2. The other side is present, but only as a token line.
    3. The other side is absent, but the article says a response was sought.
    4. The other side is absent, with no sign a response was sought.
  4. G4
    Weight of the unanswered allegation
    1. No allegation against an identifiable party.
    2. Policy criticism or disagreement.
    3. Allegation of misconduct, bad faith or incompetence.
    4. Allegation of crime, violence or grave moral fault.
  5. G5
    Loaded quotes
    1. Quotes are informational.
    2. Some evaluative words in quotes, in proportion to the event.
    3. Quoted invective carries most of the article's judgement.
    4. Quoted invective is also lifted into the headline or pull-quote.
  6. G6
    Unattributed claims
    1. Every disputable claim is attributed.
    2. One or two disputable claims are unattributed.
    3. Several disputable claims are unattributed.
    4. One side's claims are narrated as background fact.
  7. G7
    Missing context
    1. Necessary background is present.
    2. Minor context is missing.
    3. Missing context materially changes the reading.
    4. The omission is the article's central defect.

Reading a roundup

Each headline on a story is labelled one of three ways. A roundup shows the earliest headline of each kind side by side, with the judging words marked.

  1. NeutralReports what happened or who said what.
  2. Stated as factPuts a judgement in the outlet's own voice, such as calling a bill “weak”.
  3. In quotation marksCarries a judgement inside quotation marks, attributed to someone else.

Labels describe the headline only, not the article or the outlet.