METHODOLOGY
How this desk works
01We are the referee, not another oracle
The world has enough AGI forecasts. It has no one scoring them. We track every named public timeline claim — lab leaders, researchers, prediction markets, our own — timestamp it, and grade it when it resolves. Credibility is earned arithmetically, in public.
02Indicators over opinions
Ten time series, chosen because each one must move if timelines are short and must stall if they are long: task horizons, benchmark deltas, cost curves, compute capex, lab revenue, incident frequency, the AI-share of AI research, insider statements, market odds, and the labor front edge. Every value carries a source and an as-of date.
03Evidence grades on every number
VERIFIED means checked against the cited primary source this cycle. SEED means an initial value pending re-verification — visible on the board, never silently upgraded. If we can't source it, we don't print it.
04Translation, not prediction
Each brief connects evidence to allocation-relevant thresholds: when does agentic labor become an economic line item, which capex commitments confirm or falsify the software story, where does the insider/market spread sit this week. Analysis of what evidence implies — never a recommendation to buy or sell anything.
05Both directions, same rigor
A desk that only ever says 'faster' is marketing. Stall signals — benchmark saturation without deployment, capex cancellations, incident-driven pullbacks — get the same prominence as acceleration signals. Our credibility survives either future; a cheerleader's doesn't.