Seven stages turn the day's economic news into signals that are written down before the fact and graded afterwards. Each stage has one job, and each exists for a reason.
Stage 1 · Reading the news
A language model reads each article and scores its effect on the four parts of GDP: consumption, investment, government spending and trade. Each score has a sign and a size, and is judged twice: for the coming weeks and for the longer run.
Why: a headline is not a signal. Breaking it into the components of the economy is what lets later stages trace where the effect goes.
Stage 2 · GDP indices
Scores roll up into four indices — trade tension, consumer confidence, investment momentum and fiscal policy — each built to match how its part of the economy behaves: trade flips fast, consumption moves slowly, policy is sticky.
Why: direction and strength are tracked on separate tracks, so a weak move in the same direction is never mistaken for a turning point.
Stage 3 · Propagation
An economic map, drawn from theory, says how a shock moves: down the supply chain, across related sectors and regions, and forward in time as it fades. The whole effect is then scaled by the market regime and by how influential the source is.
Why: spillovers into neighbouring sectors are averaged, not added up, so one story echoing through many sectors cannot inflate itself.
Stage 4 · Regimes
The system classifies the market into one of seven regimes — from risk-on and overheating to stagflation and crisis — and watches for the fingerprints of a regime change before it happens.
Why: an energy shock helps industrials in a demand boom and hurts them in a supply squeeze. There is no signal without context, so the regime sets every threshold downstream.
Stage 5 · Signals
Sector rotation says which sectors to favour or avoid. Cascade alerts say which part of the economy is likely to move next after a break in another. Predictions are logged with an explicit, machine-checkable success condition.
Why: writing the success condition down before the fact is what makes every call gradeable afterwards.
Stage 6 · Conviction
Each signal starts neutral and earns or loses conviction from independent evidence: how big the move is, how many sources agree, whether it fits the regime, whether the largest companies in the sector confirm it, and how often this kind of signal has been right before.
Why: no single factor can carry a signal to high conviction on its own.
Stage 7 · Feedback
Every call is resolved against what actually happened, and misses are sorted into wrong direction, wrong size or wrong timing. Those grades recalibrate the earlier stages — cascade odds, sentiment strength, which sources lead.
Why: the system learns from its own record, never from market prices. That keeps its view of the economy independent of the market's, so a gap between the two still means something.
Design rules that hold everywhere in the pipeline.
Market prices never flow back into news sentiment. Disagreement between the two is a signal, not an error to correct.
Every base rate is conditioned on the regime and the type of shock. Nothing is assumed to hold in all weather.
Severity, share of GDP, sentiment, price and conviction each keep their own scale and are never added together.
A pattern only counts if there is an economic reason it should persist across regimes.