About Spotlito

Technical research built so the prose cannot contradict the data it describes.

The problem

Most market commentary is written first and justified afterwards. A number is quoted because it sounds right, a level is named because it is round, and nothing in the process can catch it when the figure is simply wrong. Add a language model and the failure gets faster, not rarer.

How a piece is made

We inverted the order. Nothing is written until the evidence exists, and nothing is published until the writing has been checked against it.

Ingest. Candles, funding and open interest are pulled from exchange APIs on a schedule, across four timeframes.
Compute. Support and resistance are derived as clusters of historical swing points weighted by touch count, recency and volume — never round numbers. Structure, trend and indicators are calculated from the same candles.
Score. Each asset is scored on that evidence and put through fixed gates: a minimum setup score, a minimum reward-to-risk, a per-asset cooldown, and a hard cap on how much is published in a day.
Write. A language model receives the computed evidence and nothing else. It may not state a number the evidence does not contain.
Verify. A numeric fact-checker reads every figure in the draft back against the evidence packet. A draft carrying an unsupported number is rejected and regenerated.

On a quiet day we publish nothing

The gates are the point. If no asset clears them, no analysis is written that day — a scan that produces nothing is the system working, not failing. A publication that must fill a slot every morning ends up manufacturing conviction it does not have.

What we cover

Around fifty crypto assets on hourly, four-hour, daily and weekly structure, and a small set of US equities and index ETFs on daily and weekly using delayed end-of-day data. Every published piece states its instrument, timeframe and data source.

What this is not