Research Method
How the engine works
Every asset gets three independent trend readings, recomputed continuously from exchange price data. Here is exactly what each published number means — and the rules we never break.
Three horizons, three algorithms
- Short-term — computed on fine-grained intraday prices; trend length is measured in hours.
- Mid-term and Long-term — computed on daily-resolution prices with different lookback structures; lengths are measured in days.
The tiers are independent readings, not nested windows. Each algorithm hunts for the strongest defensible trend at its own resolution, so a long-term trend can be younger than a mid-term one. Do not assume containment.
What each field means
| Direction | up, down,
turning up/down (an early reversal reading), and
up/down · weakening (trend intact, momentum fading). |
|---|---|
| Stability (0–100) | How uninterrupted the move has been: the share of movement in the trend's direction measured across segmented steps of the trend. 100 is a straight line; near 0 is chop. |
| Length / Since | How long the current trend has been alive, and its start date — a verifiable timestamp, on the record. |
| Change | Percent move from the trend's start to now. |
| Max pullback | The single worst adverse move the trend has survived — the honesty metric. A +80% trend that inflicted a 40% drawdown on the way is a different animal than one that did not. |
| Deviation | Distance of the current price from that horizon's own moving baseline (64-hour, 64-day and 256-day averages for short, mid and long respectively). |
Rules we never break
- No trend, no number. If an algorithm finds no defensible trend, the cell is blank. We never dress a null up as a zero.
- Stale data does not ship. Symbols whose price feed goes quiet are pulled from the table and listed as excluded — visibly.
- The archive is immutable. Every published snapshot is preserved as published. No retroactive edits, no silent deletions.
- Unknown values do not leak. If the engine emits a state our public vocabulary does not cover, we publish a blank, not a guess.
Freshness
The report shows the timestamp of the data it was produced from. If publication falls unusually far behind, the page says so with a visible data delayed flag rather than pretending.
This page describes published outputs. Model internals, training data and signal construction are proprietary. See also the disclaimer.