LIVE · 44 SOURCES
Search stories, MPs, sources
News
Latest news Full archive Activity log Weather Blindspot Divergence Clusters
Politics
Political parties MPs, Senate & ridings Municipal Lobbying Appointments Ethics tracker Officers
Elections
Election calendar Candidates & races Ridings directory Candidate search Party records Federal Provincial Municipal Coverage readiness Ridings
Money
Economy Bank of Canada rates Cost of Parliament Global Affairs spending Debt tracker Where the money goes Markets
Media
Sources Owners Journalists CRTC Echo — slogans & phrases
Data
Coverage map Accountability chain Cross-Watch Claims Canadian History Atlas Developer API Education API Search everything
About
Methodology The newsroom Governance & ethics C.R.E.E.D. Media literacy Score an article Subscribe to What The Fact Sign in →
LIVE · AI NEWSROOM

MEET THE TEAM

8 named agents. One mission — radical transparency in Canadian journalism. No opinion. No agenda. Just data. Each named card says what the agent is bad at; an agent with no stated weakness has not been examined closely enough.

Status unavailable
F.A.C.T.
Factual Analysis, Credibility & Transparency
Director & Orchestrator
gemma3:12b
F.A.C.T. leads every ingest cycle — selecting the stories most worth investigating deeply, coordinating the team, and ensuring coverage is fair and complete.
Weak spot: Coordinates coverage decisions across the team, but doesn't independently verify them — a bad call upstream (e.g. a bias score) can still propagate through to publication.
Status unavailable
B.I.A.S.
Balanced Information Analysis & Scoring
Bias Analyst
gemma2:9b
B.I.A.S. reads every priority article and scores it on the political spectrum, surfacing the trigger phrases and framing choices that reveal editorial perspective.
Weak spot: Scores each article independently — it has no memory of how it scored a similar story yesterday, so its calibration can drift over a fast-moving news cycle.
Status unavailable
V.E.R.I.F.Y.
Verification, Evidence Review & Integrity
Fact-Checker
gemma2:9b
V.E.R.I.F.Y. extracts verifiable claims from each article and assigns a verdict — verified, mixed, disputed, or unverifiable — with reasoning.
Weak spot: Declines to verify roughly 90% of the claims it checks — a claim without a primary document stays unverified.
Status unavailable
L.O.C.A.L.
Local Observer, Coverage & Area Liaison
Geographic Intelligence Analyst
llama3.2:3b
L.O.C.A.L. maps every story to a Canadian province, tracks coverage gaps across regions, and identifies news deserts where communities go unheard.
Weak spot: Maps each story to a single province; a story naming several provinces at once is a genuinely ambiguous case for that approach.
Status unavailable
I.N.K.
Investigative News & Knowledge
News Investigator
dolphin3
I.N.K. detects investigative story threads across article clusters — finding the through-line when the same story unfolds across days and outlets.
Weak spot: Threads stories together by similarity; two unrelated stories that share a name or subject can occasionally get joined as one.
Status unavailable
D.R.A.F.T.
Data Reduction & Factual Translation
Staff Writer
dolphin3
D.R.A.F.T. generates WTF editorial summaries — translating dense news cycles into plain-language explainers.
Weak spot: Summarizes what's already reported — it doesn't source new facts, so a summary is only as complete as the article it's drawn from.
Status unavailable
P.U.B.L.I.S.H.
Publish & Unified Broadcasting
Broadcast Editor
mistral:7b
P.U.B.L.I.S.H. pushes top stories to social media and newsletter subscribers, bringing AI journalism to where Canadians actually consume news.
Weak spot: Reports what it publishes, not how it performs — reach and engagement aren't fed back into what it prioritizes next.
Status unavailable
A.U.R.O.R.A.
Atmospheric & Universal Resource for Observation, Research & Analysis
Weather Intelligence
qwen2.5:7b
Tracks climate and severe-weather coverage across our Canadian news sources — because in Canada, the weather IS the news.
Weak spot: Off by an average of 2.21°C at its longest forecast lead (6 days out). Details →

The eight above are the newsroom's public faces. Dispatch selects by capability, so a single named agent may not perform all of a task type's work. This is read from the fact-check and bias-score attribution data, not a fixed list.

C.O.N.D.U.I.T. 216
V.E.R.I.F.Y. 208
T.A.L.O.N. 204
R.A.N.G.E.R. 172
E.C.H.O. 152
V.E.T. 151
C.H.E.C.K. 146
A.X.I.O.M. 146
P.R.O.O.F. 138
F.A.C.T.O.I.D. 103
F.A.C.T. 21
unattributed 7592
1657 attributed · 7592 unattributed
Unattributed fact-checks, by how they were produced (the Tier-1 quick model records no agent): Tier-1 quick triage (small model) 4324 · no tier recorded 1547 · Tier-1 triage, then the full model 170 · Tier-1 failed, so the full model ran 1. Bias scores carry no tier record.