How the daily run works — from market close to a one-line verdict
I'm Andrea Pilotti, and I built this to measure the risk in my own AI and semiconductor holdings. Every trading day the same fourteen steps run in the same order, governed by a single written methodology. This page describes the process end to end: what happens, in what sequence, and why each step exists. Thresholds, weights and calibration values are proprietary and are not published — everything else is here.
What the system estimates
Three questions with fixed definitions and fixed horizons, re-estimated every trading day. Fixed definitions are the point: a forecast whose target moves cannot be scored later.
Probability that the semiconductor complex falls 20% or more from its 52-week peak inside a rolling 30-day window.
The same measure at a deeper level and a longer horizon: the scenario in which a correction becomes a sector bear market.
Probability that the decline stops being a sector event and becomes a market-wide one, with AI as the epicentre.
What the output is
A risk regime, an alert level and calibrated probabilities, published the same way whether the reading is calm or severe.
What it is not
Not a recommendation. The framework produces risk information — never trade calls, position sizing or timing. Estimates are kept calibrated rather than conservative: inflating probabilities "to be safe" would destroy the record's ability to show whether the system works.
Fourteen steps, in the same order, every day
Click any step to open its detail — or expand all. The sequence is not a description of good intentions: it is the order the run is actually executed in, and each step is designed to make a specific shortcut impossible.
The six commitments the process enforces
Each one exists because of a specific way daily research goes wrong.
Weekly, Saturday, monthly
Some checks cannot be done well in a daily window. They have their own cadence.
What this page deliberately omits
The architecture above is described in full, because the discipline is the architecture. The parameters are what make it work in practice — and they are the part that takes longest to build and least time to copy.
- Signal thresholds and trigger conditions — the specific values that move each of the nine signals from one state to the next.
- Ledger weights and base rates — the adjustment magnitudes, and the conditioned historical rates the estimates start from.
- The family-count mapping — the exact rule translating signal states into a regime.
- Escalation and de-escalation parameters — the numeric definition of each alert trigger and each recovery condition.
- Series, proxies and endpoints — the data vendors are named on the home page; the specific instruments and derived measures are not.
- Universe constituents — the nine groups are described; the individual names are not.
- Calibration parameters.
The read is published every trading day
The regime, the alert level and the probabilities are updated with each run, and the reasoning behind each read is published alongside it.
Not investment advice. I am not a financial advisor and I am not licensed. I publish independent quantitative research on AI-related equity market risk, starting from the risk in my own exposure. Nothing on this page is a recommendation to buy, sell or hold any security, and nothing here constitutes investment, legal or tax advice. Estimates are probabilistic, may be wrong, and the past behaviour of any indicator is not indicative of future results. Any investment decision remains the reader's responsibility, ideally taken with a licensed adviser.