Investment Intelligence
Model-assisted insight. Human reviewed.
The intelligence layer surfaces regime, catalyst, sentiment, and learning context so decisions can be reviewed with evidence.
What it watches
- A frozen daily state per sector: macro weather, trend, breadth, news tone, earnings congestion, insider activity
- Nightly analyst opinions per name — direction, conviction, and what would prove them wrong — written before the session, never after
- Every opinion graded at fixed horizons against the name’s own benchmark
- Conviction calibration: whether high-conviction calls actually beat low-conviction ones, published either way
- A weekly letter written from the graded record — thin evidence produces thin conclusions, stated as such
A decision record, end to end
- Written. The analyst states a direction, a conviction from 0\u20135, and a concrete invalidation condition \u2014 frozen before the session opens.
- Executed \u2014 or not. Deterministic rules decide independently. The analyst cannot place, size, or veto anything.
- Graded. At 1, 5 and 10 sessions the opinion is scored against the name\u2019s own benchmark, from broker records.
- Learned from. Only findings with real sample depth become lessons; the rest are labelled insufficient, honestly.
An analyst earns influence the same way a strategy earns capital: by being right, measurably, and surviving review.
Governed
Models operate inside defined limits and supervised workflows.
Explainable
Outputs include rationale, confidence, source quality, and context.
Accountable
Every opinion is scored against what happened. The calibration is published either way.