News, social & macro sentiment

The idea

Charts do not move themselves. Earnings, upgrades, exploits, regulatory headlines and plain crowd mood move them — and by the time that information is fully in the price, the move you wanted to catch is over. The sentiment channel exists to measure the information environment alongside the chart, without ever being allowed to shout over it.

Reading the news with language models

Recent headlines and articles about the asset are gathered and passed to large language models with two narrow jobs. The first summarises: what actually happened, stripped of headline hyperbole. The second classifies impact on a fixed tier scale — from market-moving (a delisting, an enforcement action, a surprise earnings miss) down to ambient noise. Classification against a fixed rubric, rather than free-form opinion, is what keeps the model accountable: the tier it assigns is shown to you in the analyst report, and it can be wrong in ways you can see.

Social chatter and the macro backdrop

For crypto assets especially, social momentum is a real, tradable phenomenon — and a notoriously manipulated one. Social activity measures (volume of discussion, engagement trends, creator concentration) are normalised against the asset's own baseline, so “loud” means louder than usual for this asset, not louder than a larger asset's ordinary day. The macro channel works the same way on the slower drumbeat: rates expectations, risk appetite, sector rotation.

The trust boundary

Sentiment is the channel most vulnerable to manipulation — a coordinated shill campaign is cheap; faking a year of price history is not. So the ensemble enforces a trust boundary: sentiment scores are bounded, capped in weight, and cannot flip a forecast that price-based evidence contradicts. In the analyst report you will sometimes see exactly that tension — “social momentum is elevated, but the technical picture does not confirm” — which is the system working as designed, not hedging.

When there is nothing to read

Thinly-covered assets sometimes have no meaningful news flow. The honest answer is a neutral score with low weight — not a hallucinated narrative. Screenshot-mode predictions work the same way: if you confirm a symbol we can research, sentiment runs live; if the instrument is unknown, the channel abstains and the forecast leans on the technical evidence alone, and says so.