The Apex Times / Methodology v1
Apex Coverage Analysis
Our comparisons show reporting found in monitored sources and, when available, external news search. These are bounded searches, not a complete inventory of every outlet covering an event. Political-lean labels are Apex editorial judgments or explicitly labeled experimental AI estimates, not ratings supplied or endorsed by Ground News, AllSides, Ad Fontes, or Media Bias/Fact Check.
How We Find Related Reports
Article searches aim for 10-15 additional reports, prioritizing distinct publishers. Headline and lead context, explicit bill identifiers, locations and a three-day window around the original story help filter candidates. We do not add unrelated reports to meet a quota. External discovery uses the GDELT Project; its first-indexed timestamps are labeled separately from confirmed publication dates. An older article can be indexed later, so matches remain provisional. Search results provide headlines and source links, not proof that we have read the complete articles. Search outages leave saved results and monitored-source matches available.
What the Coverage Bar Counts
Each distinct publisher domain counts once. All matched news outlets, including unrated outlets, remain in the denominator. Official records are shown separately, never assumed to be centrist. The percentage bar combines left and lean-left outlets under Left, and right and lean-right outlets under Right; the logo columns retain all five categories. Percentages are rounded to whole numbers totaling 100 when news outlets are present. Multiple outlets may carry the same wire report or rely on the same announcement; publisher count is not independent corroboration. Missing coverage in our index does not establish that an outlet ignored a story.
Our Political-Lean Rubric
Left and right describe framing relative to the contemporary US policy spectrum. Lean left and lean right indicate a narrower tendency; the outer categories indicate a more consistent tendency in the reviewed sample. Center requires evidence of competing perspectives, not merely a lack of obvious partisan words. Unclear or insufficient evidence stays unrated. Mentioning a party, criticizing a politician, emotional language, and factual accuracy do not by themselves establish political lean. No category is a truth score.
Articles and Outlets Are Different
An article-framing assessment applies only to the stored passage reviewed. Feed snippets and limited excerpts are not complete articles: we cannot infer what the full article omitted. An outlet editorial assessment requires an editor to review at least 12 reports spanning three topics and 14 days within a 90-day window, with exact evidence passages and a written explanation. This is a minimum sample, not a claim of statistically representative measurement. News, opinion, broadcast and other formats may differ.
Automatic AI Estimates
Where available, an AI badge identifies an unreviewed estimate based only on a publisher's sampled headlines and feed snippets. We sample 12-16 distinct reports across at least 14 days within a 90-day window. At least three topics and 12 interpretable passages with exact internal evidence are required for a directional or center estimate. Publisher names and URLs are withheld from the model; it is instructed not to infer reputation or treat a quoted speaker's position as the outlet's. These checks do not establish accuracy or a representative sample. Estimates can be wrong and do not describe every article. Insufficient evidence stays unrated. The explanation includes sample links, model and generation date. Eligible samples are refreshed no more than weekly; estimates expire after 30 days. API limits and missing material can delay updates. Editors' assessments take precedence; a manual withdrawal is not replaced automatically by an AI estimate.
Evidence and Editorial Review
Related headlines are matched automatically and can include different developments. Published comparison points require linked supporting sources; disagreements are not manufactured for balance. AI assistance for story comparisons produces proposals only. An identified editor must check the event matches, evidence, interpretations and limitations before an editorial assessment becomes public. Automatic outlet estimates are a separate, clearly labeled feature without that approval. Provider names and party labels are not a substitute for reading the source material.
Updates and Corrections
Editorial assessments show their review date and sample links, expire after 90 days, and return to pending review when edited. Revision history is retained internally. We do not automatically average article labels into a publisher score. We do not claim that these judgments are free of bias.
Disagree with an assessment or spot an incorrect match? Request a correction, including the report URL and the evidence we should review.