Every forecast the M3I research brain has made on AI infrastructure — memory, custom silicon, networking, power and the macro regime — timestamped when made, then graded against what the tape actually did. The Audit Trail covers the geopolitical record; this page covers the AI one.
The brain reasons under three inputs — Jensen Huang's demand model (compute is revenue; the binding constraints are chips, memory, power and lithography), the Visser / 22V AI Macro Nexus regime read (weekly technical scores, absolute exhaustion and a consolidation playbook across a 100-name AI-infrastructure universe), and the Mosaic Method — the week's earnings prints and five tracked expert podcasts fused into one forecast, where convergence across differently-incentivized voices is the strongest signal. It does not score on valuation multiples.
Every call it makes carries a probability and a score-by date and is logged before the outcome is known. When the horizon expires — or an event tests the call early — it is graded Right, Partial or Wrong against end-of-day price data and public facts. Probabilities are updated weekly with a one-line reason; the original figure stays on the record. The misses are analysed in more depth than the hits, because they are where the calibration comes from.
What is published here. Index, theme and macro calls only — the semiconductor index, sector relatives, rates, memory pricing, Bitcoin, the crypto-rails index and its verticals, policy and financing events. Single-name calls and the per-name breakout-entry flags are graded on the same rules internally and reported here only as a tally, so this page never doubles as a stock list. The full-record hit-rate is shown alongside the published one for honesty; both are computed the same way.
Read the numbers with the sample size in mind. Under thirty graded calls this is directional, not statistical. The brain's known weakness is timing; its stronger skill is identifying floors and direction.