BV-7X has always been a 7-day oracle. Every night at 21:35 UTC, it predicts whether Bitcoin will be higher or lower in a week. That model runs at 62% accuracy across 13 years of walk-forward validated data. It bets real USDC on Polymarket every day.
We’ve spent the last several weeks trying to go shorter. Can we predict 3 days ahead? 1 day?
Short answer: yes, but it required building entirely new models. The 7-day signals have zero predictive power at shorter horizons. Different timeframes are different problems.
The 7-Day Model
The production signal. Live since February 2026. Multiple signals vote on macro regime, momentum, institutional flows, and on-chain valuation. Version 5.6.6 added regime-adaptive thresholds that adjust to the current market environment — crisis, bear, chop, bull, euphoria — adding nearly 2 percentage points to accuracy.
This is the model that bets. Everything below is R&D.
The Search for Shorter Horizons
We tested 38 features across 8 data sources looking for anything with genuine predictive power for 1-day and 3-day BTC returns. Most of what we tried failed:
- Polymarket macro event markets (Fed, war, recession, tariffs) — scraped 50,000 resolved markets. Zero walk-forward improvement.
- Perpetual futures positioning (funding rates, long/short ratios) — coin flip.
- Cross-asset momentum (equities, gold, VIX) — weak, unstable.
- Macro event calendar gating (FOMC, CPI, NFP) — no effect.
- Fast technical oscillators — near-zero signal.
Most of these had strong statistical correlation with returns, but correlation doesn’t mean you can trade it. The gap between “ranks returns correctly” and “predicts direction” is where most quant strategies die.
Two data categories survived. We’re not saying which ones or how we use them.
New: 3-Day and 1-Day Shadow Models
We found something that works. Not the same architecture as the 7-day model — completely different signals, different data sources, different logic. The 3-day model uses a single contrarian override. The 1-day model uses a multi-signal voting system.
Both are now in shadow mode: generating predictions every night alongside the 7-day signal, logging outcomes, tracking accuracy. No bets yet. If they hold up over 30+ days of live data, they graduate to production.
The 1-day and 3-day predictions are visible on the Terminal scorecard for holders with 1B+ $BV7X. You can watch all three horizons call the market differently in real time — and see which ones are right.
The models frequently disagree. A 7-day BUY with a 1-day SELL isn’t a contradiction — it’s two timeframes seeing different forces. The weekly trend can be up while the overnight position is overleveraged.
What Comes Next
If the shadow models prove themselves, they’ll start betting on Polymarket with the same deep-buffer strategy that protects the 7-day model. If they don’t, they stay in shadow indefinitely. The 7-day model is the revenue engine. Everything else earns its spot.
Methodology details for the 1-day and 3-day models are not public. The signal direction, the data sources, and the thresholds are part of the premium oracle product. Hold 1B $BV7X to watch the shadow predictions live.
Watch the Shadow Models
1-day, 3-day, and 7-day predictions on the Terminal scorecard. Updated nightly. 1B $BV7X required for shadow horizon access.
Live Terminal →