Market intelligenceThe observation
One market.Many waysto see it.
Seven ways of measuring the same 1,047 days of Bitcoin. Each one describes what already happened. None of them predicts what happens next — and they rarely say the same thing.
Open the market lens01The market lens
Eight lenses. One timeline.
Historical observation
BTC/USDT · daily · 1 Jan 2024 – 12 Nov 2026 · 1,047 candles
Descriptive statistics of history. Not predictions, not signals.
Trend · EMA 20 / 50
54%
Days with EMA 20 above EMA 50
- Trend flips
- 19
- EMA crossovers in the window
- Longest up-state
- 148 days
- from 22 Sept 24
- State at the last bar
- Down
- since 2 Nov 26
Two exponential moving averages smooth the closes. While the 20-day sits above the 50-day the stretch is shaded as an up-state; below it, a down-state.
Does not tell you: Averages lag by construction: a state is only confirmed after the move that created it.
Formula & parameters
EMAₜ = α · closeₜ + (1 − α) · EMAₜ₋₁, α = 2 / (n + 1)
State = sign(EMA20 − EMA50), after a 50-day warm-up.
Lower panel: (EMA fast − EMA slow) / close.
Point-in-time: each day uses only that day and the days before it.
02Agreement & disagreement
Methods disagree. That is the data.
Each lane is one method’s call on every closed hour, written down before the outcome was known and scored afterwards. When the lanes split, the hour was ambiguous to these rules. We keep the split visible instead of averaging it away.
64%
of the 109 hours where two or more methods made a call, at least one said up while another said down (70 hours).
Price
79,221–84,898 USDT
Momentum 24baseline
39 up · 70 down · 59 no call
21% hit · 95% CI 16–26% · 239 resolved
Disagreement
Bar height = share of calls against the majority. Full height = an even split.
Resolved outcome
What the hour actually did: up, down or sideways.
Up · upper half Down · lower half Sideways · centre line
Hit rates cover every recorded evaluation of each method on BTC/USDT 1h in the shadow replay, not only the hours drawn here. Each hour resolves as up, down or sideways, so a directional call in a sideways hour counts as a miss. Rules, baselines and models record calls before outcomes are known and are scored automatically. No validated edge.
Not drawn, because they have no recorded calls: LightGBM (ready); XGBoost (ready); Chronos-2 (needs more memory and compute than this server has — evaluation pending); TimesFM 2.5 (needs more memory and compute than this server has — evaluation pending).
03The data
Frozen.
Unfilled.
Public.
- 01
Point-in-time. Every value on a given day uses that day and the days before it. Nothing is computed with hindsight.
- 02
Descriptive, not predictive. Lenses summarise history. They are not forecasts and not trading signals.
- 03
Disagreement is kept. When methods split, the split is shown. Nothing is blended into a single score.
1,047
Frozen candles
A hashed, versioned snapshot of Binance public spot klines, 1 Jan 2024 – 12 Nov 2026. Re-running a lens gives the same numbers.
0
Days missing in this window
Gaps are preserved, never filled. If the exchange paused, the series would show the hole instead of inventing candles.
0
Signed requests
Public market-data endpoints only. No account access, no exchange keys, no orders.
Research snapshot exported 27 Oct 2026 · historical by design, never refreshed to look recent.