How we classify, count, and correct

Every story on Shadowfetch is assigned to one of three lanes. A lane describes framing — whose concerns lead the story, which trade-offs get emphasis, how causes and remedies are characterized. It is never a judgment about accuracy, and it is not an endorsement. Well-reported journalism exists in all three lanes; so does weak journalism.

Classification happens in two steps. First, an editorial assignment: Shadowfetch assigns a desk to explore a particular emphasis or set of questions. The desk label describes the coverage frame, not the truth or worth of a claim.

Second, a story-level check. An editor reads the piece itself and scores its framing: the headline’s emphasis, which voices appear first and last, what context is included or omitted, and whether loaded characterizations are attributed or adopted. If the story-level read differs from the assignment, the story wins. A second editor must independently reach the same lane before the tag is published; when the two disagree, the story goes to a cross-desk review and, if the disagreement survives, it defaults to Center with a note in our internal log. All standing ratings are re-examined quarterly.

The meter on our homepage is deliberately simple arithmetic: each lane’s share is its number of published stories divided by the total, over the window shown. No weighting, no editorial thumb on the scale — if the mix drifts, the meter shows the drift.

The homepage also charts the same calculation day by day across the trailing week. Our target is parity over the week, not in every hour — news breaks unevenly, and forcing an artificial 50-50 on a day when one side of a story simply produced more coverage would itself be a distortion.

A statement qualifies as a Common Ground fact only when it clears two bars. It must be independently reported by at least one Shadowfetch desk in each of the three lanes — not merely repeated from the same underlying record. And it must be stated in checkable terms: a number, a date, a vote tally, a direct quotation — something a reader can verify without trusting us. We list the agreeing Shadowfetch desks next to every fact so you can do exactly that.

Story dossiers keep three things separate. Verified across coverage means differently framed reports agree on a checkable detail. What is still unknownidentifies questions the available evidence cannot yet answer. The full report preserves the original lane's sourcing and framing rather than relabeling every sentence as common ground.

The evidence record also says when a primary document or dataset is referenced but not attached. We do not treat a citation we cannot show as though the reader has inspected it. Terms that may be unfamiliar link to the plain-English glossary.

The paired headlines in Compare Coverage are matched on topic and time window: both stories must address the same underlying news event within the same news cycle. Headlines are reproduced verbatim — we do not paraphrase, trim, or select for maximum outrage. The point is the ordinary, everyday difference in framing, which needs no exaggeration.

When we publish an error of fact, we correct it promptly, mark the corrected passage, and append a note to the story saying what changed and when. Material errors — ones that could have changed a reader’s understanding of the story — are flagged at the top of the article, not buried at the bottom. Clarifications that do not involve an error of fact are labeled as clarifications, because the difference matters.

Think we got something wrong — a fact, or a lane tag? Write to support@shadowfetch.com with the story link and what you believe is incorrect. Every request receives a review and a reply.

Lane classification involves judgment, and judgment can be argued with — that is why the method is public. The Left–Center– Right axis is an American shorthand that fits some stories better than others; a trade dispute maps onto it more cleanly than a flood does. And over short windows, small sample sizes can swing the meter by several points without meaning much. We publish the method, the math, and the underlying counts so you can challenge all three. When a reader’s challenge is right, we change the tag and say so.

Curious how this looks in practice? Search the archive and filter any topic by lane.