Data first
The core of The Agent Bets is original data, aggregated from public venue APIs (Kalshi, Polymarket, Limitless) and public filings. Our dashboards, volume trackers, listing feeds, and resolution trackers are computed from that data. That aggregation is original to us: it cannot exist anywhere else in this form.
Reported pieces
We publish at most five reported pieces per day. Each one starts from a real development or from our own data, is drafted with machine assistance, and passes four automated gates before a human sees it: a rate limit, an originality check, a value-add check that asks what the page adds beyond its sources, and an Ethics Engine score (below). Nothing publishes without human approval. The gate scripts live in the site's seo/ and ethics/ directories.
Ethics Engine
Every reported piece is scored by The Agent Times Ethics Engine, the same rule-based journalism audit that grades The Agent Times' own reporting. The engine checks ten signals: confidence labels, attribution density, a sources section, clickable sources, a key-facts box, loaded language, self-reported labels, word count, fabrication signals, and editorial voice. Publication requires an engine pass or a score of 70/100 or higher. The grade badge on each article links back to the engine.
Confidence labels
Every piece carries a claim-level confidence label next to its dateline, per the engine's first rule: CONFIRMED means verified by a primary source or two or more independent reports; REPORTED means published by a credible outlet but not independently verified by us; ESTIMATED means derived from inference or modeling. Probabilities on this site are timestamped venue prices, never facts, forecasts, or advice.
Bets-specific rules
Two deterministic checks run on every draft before it can even reach the Ethics Engine. First, any probability, percentage, or odds figure must carry a timestamp or an "as of" attribution. Second, no betting-advice language: no picks, no "bet on", no calls to action to wager. This is an industry publication, not a tip sheet.
What automation does and doesn't do
Automation collects data, drafts structure, and runs checks. It does not decide what is true. Verification, source judgment, and the final publish decision are human. Every reported piece carries a provenance line describing how it was verified.
Seed content
Launch content on this site is marked as draft in our repository and is replaced by pipeline-verified reporting as the newsroom ramps up.
Corrections
Mistakes are corrected openly under our corrections policy.