Market Intelligence
From data to understanding: the market regime on three axes, each with its own evidence and confidence. Never merged into a single label.
Three axes, and never a signal
APEX's market regime is not a label. It is three independent measurements, because merging them would lie by compression.
RISK_ON or RISK_OFF. Each instrument contributes with its own sign: an index up is risk appetite, the VIX up is the opposite, the dollar and yields up are pressure.
HIGH or LOW, and what it was measured with: the VIX if it is in the basket, the series' realized volatility if not. The source travels with the measurement.
TRENDING or RANGING. And TRANSITION and UNCERTAIN exist as legitimate answers: not knowing is a state, not a failure.
Each axis brings its own evidence and its own confidence. With fewer than two measured instruments, the regime is UNCERTAIN with zero confidence and it says how many there are and how many would be needed. An environment is not assumed from one data point.
The regime is context and cannot be turned into buy or sell
The text the model sees says so literally. And it does not stop at intent: there is a test that scans the code of this layer looking for any trading action verb. If someone puts a “buy” in there, it fails.
It seems like an excessive precaution until you think about what would happen without it: a risk-off regime turned into a recommendation is a market signal disguised as context, served to every user at once, on data APEX presents as neutral. It is the kind of drift you do not see coming in a code review.
Gold and oil enter with a zero sign, on purpose. Gold is a safe haven and oil is an input cost: forcing them into a direction so the numbers come out neat would be inventing the very signal this layer exists not to invent.
The market cannot push out your memory
The context the model receives has a budget: not everything fits. The market context and your context have separate budgets, and the market one is applied first and apart.
If they shared a budget, the market would push out the memory — which is last on the list and the first to go — and that would invert the product's priority. APEX is not a data terminal: it is an assistant that knows you. An S&P data point cannot take the place of “two weeks ago you entered before your confirmation on this same asset”.
The regime has no history. It is measured now, and APEX keeps no earlier snapshot of it, so it cannot say “the environment changed regime since your last analysis”. What it does compare between two moments are the prices in the basket. It is a data gap, and it is declared instead of filled in.
The same regime, in all four places
Chart analysis, Decision Analysis, Fundamental Analysis and Quick Check read this layer. There is no second implementation.