The mathematical and procedural rules governing how sports journalism claims are verified, resolved, and synthesized into public reliability records. Every score displayed on Sports News AI is deterministically reproducible from the equations below.
Most platforms rank news by engagement. We track whether claims turn out to be true, and publish the receipts. The Accountability Ledger is append-only: claims and resolutions are never deleted or silently edited, preserving an immutable audit trail of modern sports reporting.
Five deliberate design choices, not marketing claims — each one is a page or a feature you can go check right now.
Every score on the reliability desk carries its sample size next to it. Below 10 resolved claims, we show “insufficient record” instead of a number — see §2 below.
We record who broke a story, not who republished it loudest. Event cards credit the outlet that reported first and how far ahead of the pack they were — not just whoever's headline you saw.
Every transfer saga is one page: every claim, who made it, what corroborated or contradicted it, and how it ended. Competitors show you today's article; we show you the whole arc.
A speed lane pushes a bare factual alert within seconds of detection — no generated prose. An evidence lane follows with a grounded, sourced summary once it's actually verified.
Every published sentence traces back to stored evidence, and the interface will show you that evidence. In a market saturated with generated slop, verifiable restraint is the fifth differentiator — and the one the other four exist to protect.
The sports news app that keeps the receipts: we collect reporting from approved sources, group it into events, extract the specific claims each outlet makes, and track whether those claims turn out to be true.
Explicitly not:
Raw percentages (e.g. 2 correct out of 2 = 100%) are fundamentally misleading for evaluating journalists. A reporter with 45 correct out of 50 stories has demonstrated far higher reliability than someone who is 2 for 2. We compute the Wilson score interval lower bound at a 95% confidence level (\(z = 1.95996\)):
This guarantees that a source's score reflects the statistical lower bound of their true accuracy with 95% certainty.
To prevent unfair reputational harm from tiny sample sizes, no score pill is ever rendered for an outlet or reporter with fewer than 10 resolved claims.
“Insufficient record · 3/4”
Displays the raw fraction only. Score color and Wilson lower bound are withheld.
“68% · 24/31”
Public reliability pill unlocked, linking directly to the reporter's itemized dossier.
A core failure of modern sports aggregation is outlets repackaging wire reports and claiming credit when they turn out true. Under Handbook §5.2 rules:
Sports journalism is rarely 100% binary. A reporter who identifies the correct target club and player but is off on the fee should not be scored the same as someone who fabricated an entire transfer. Resolutions evaluate four discrete components:
Player & Club accuracy
Signing vs. denial correctness
Window & contract duration
±15% transfer fee precision
Reporters change beats, lose sources, or improve rigor over time. Past claims are weighted using exponential recency decay with a half-life of 90 days (~one full football season/transfer window):
A story reported 90 days ago carries half the weight of breaking reporting from today; a story from two seasons ago carries 1/16th weight.
All evaluations rely strictly on Authority Rank 1–2 official announcements (clubs, leagues, federations) or multi-outlet consensus. Under no circumstances does an AI model invent or guess the resolution of a claim.