How we report

What Bearings is

Bearings follows a developing story across news outlets from different perspectives — left, right, center, state media, and international bodies — and breaks each article into individual claims. When multiple outlets independently report the same thing, you can see that. When they disagree, you can see that too, spelled out.

What we don't do

We don't decide what's true. We show who reported what, when, and how much independent agreement exists — and leave the judgment to you. We also don't report causes: when an outlet asserts that one event caused another, we label it as that outlet's assertion, never as a finding. Statistical analysis of causes requires weeks of data, and until a story has enough, we say so plainly: insufficient data.

How claims are grouped

Our approach follows the U.S. Intelligence Community's analytic standards (ICD 203). Every claim is typed: a reported fact (directly observable), an attributed statement (sourced to a named person or document), an asserted cause, or a characterization (interpretive framing). Claims describing the same underlying assertion are grouped, and the count of distinct outlets reporting it becomes its corroboration level.

One honest wrinkle: when several outlets all cite the same wire dispatch or the same anonymous official, that's an echo of one source, not independent confirmation. We flag those cases and count them as a single confirmation.

Perspective labels

Outlet perspective labels follow AllSides media bias categories, with two additions outside the left–right spectrum:

• Left
• Lean left
• Center
• Lean right
• Right
• Iranian state media— outside the left–right spectrum
• Russian state media— outside the left–right spectrum
• Chinese state media— outside the left–right spectrum
• Saudi-aligned media— outside the left–right spectrum
• US government— outside the left–right spectrum
• US state-funded media— outside the left–right spectrum
• Intergovernmental— outside the left–right spectrum
• Fact-checker— outside the left–right spectrum

Cost transparency

Claim extraction and grouping use AI (Anthropic's Claude models). Because trust includes knowing what runs the machine, here's the complete processing bill to date: $7.45.