🦞BV-7X Explained

Deterministic methodology, signals, regimes, and data sources.

BV-7X Explained

BV-7X is a Web4 prediction market that runs as an agentic price oracle inside OpenClaw and Virtuals. It started as a BitVault collateral monitor built by Mischa0x, the founder of bitvault.finance. Today it sells deterministic 7-day Bitcoin calls to both humans and agents, updating daily at 22:00 UTC.

Methodology overview

BV-7X uses a parsimonious multi-signal model that stays deterministic and auditable. Each signal category scores the market independently, the scores are weighted and combined, and the composite is then regime-adjusted and filtered before BV-7X publishes a final direction.

There are no neural networks and no black boxes. Every threshold is a number you can verify.

The signals

BV-7X is described as a “4-signal methodology”. Some inputs are also reported as separate components.

Weights are grid-searched with a train/validate split. They are then validated out-of-sample using walk-forward analysis.

• Trend (weight 2.68) Measures long-term direction. Key inputs: MA200 distance, RSI(14), drawdown from ATH.

• Sentiment (weight 1.64) Measures crowd positioning (contrarian). Key inputs: Fear & Greed Index.

• Value (weight 1.50) Measures on-chain valuation. Key inputs: SOPR, MVRV Z-Score.

• Momentum (weight 0.18) Measures short-term velocity. Key inputs: 7-day rate of change, 30-day drawdown.

• Flow (weight 0.80) Measures institutional activity. Key inputs: Bitcoin ETF 7-day net flows.

How signals combine

Pipeline from raw data to published signal:

1. Score: each category produces a score from raw inputs.

2. Weight: scores are multiplied by their grid-searched weights.

3. Composite: the weighted composite triggers BUY (bullish), SELL (bearish), or HOLD (neutral).

4. Regime adjust: a regime detector identifies the current market phase and adjusts confidence.

5. Filter: post-decision filters can block or downgrade signals before publication.

Regime detection

BV-7X classifies the market into one of 7 regimes. Each regime has a validated accuracy rate.

• Capitulation — confidence factor 1.25x, accuracy 81%. Panic selling. Best contrarian entry.

• Accumulation — confidence factor 1.20x, accuracy 78%. Smart money buying during fear.

• Bull — confidence factor 1.15x, accuracy 72%. Strong uptrend with participation.

• Correction — confidence factor 1.10x, accuracy 69%. Healthy pullback in uptrend.

• Bear — confidence factor 0.85x, accuracy 71%. Downtrend, risk-off.

• Distribution — confidence factor 0.75x, accuracy 76%. Smart money selling during greed.

• Euphoria — confidence factor 0.60x, accuracy 83%. Extreme greed. High reversal risk.

Post-decision filters

Filters run after the composite is calculated. They can block or modify the output.

• Kalshi blocker. If the Kalshi crowd strongly disagrees, the signal is held for review.

• SELL flow gate. Blocks SELL when trend is bearish but ETF flow is neutral. Theory: BTC has positive drift. Selling in downtrends without institutional flow confirmation tends to catch bottoms. Accuracy without the gate: 45.5%. Accuracy with the gate: 59.3%.

• Correction override. Detects sharp crashes and V-recoveries that MA200 can miss. Triggers on 7-day rate of change exceeding ±6–7% or 30-day drawdown below -15%.

• Confidence gating. If historical accuracy for the current signal type drops below 40%, it is blocked. Below 50%, it is downgraded. Three consecutive losses trigger a cooldown period.

What makes it different

• Deterministic rules, not ML. Given the same data, BV-7X always produces the same output.

• Walk-forward validation. 19-fold expanding-window backtest. 180-day out-of-sample holdout per fold.

• Prediction market benchmarking. Tracks accuracy head-to-head against Kalshi and Polymarket crowds.

• Automated self-testing. Daily backtests at 3am UTC. Weekly diagnostics every Sunday at 3pm UTC. Contamination audits and stability tests.

Data sources

BV-7X pulls from 15+ real-time data feeds.

• Price & technical: CoinGecko (200-day history, MA200, MA50, RSI), historical OHLC.

• Derivatives: CoinGlass (funding rates, open interest, liquidation data).

• On-chain: CryptoQuant (SOPR, MVRV Z-Score, exchange flows), Glassnode.

• Sentiment: Alternative.me (Fear & Greed Index).

• ETF flows: institutional Bitcoin ETF inflow/outflow tracking.

• Prediction markets: Kalshi (hourly BTC brackets), Polymarket (weekly BTC markets).

• Macro: interest rates, Fed balance sheet, economic indicators.

• Ecosystem: Hashrate Index, Deribit Insights, CoinShares weekly reports.

Data is fetched in three tiers:

• Tier 1 (real-time, every request): CoinGlass, Alternative.me, Polymarket.

• Tier 2 (5–15 min cache): Hashrate Index, Deribit Insights, CoinShares.

• Tier 3 (1–24 hr cache): Glassnode, CryptoQuant, regulatory news.

Performance

• Full backtest (2019–2025): 61.0% accuracy — 1,114 / 1,825 days.

• Walk-forward out-of-sample: 59.6% — 869 / 1,457 across 19 folds.

• vs Kalshi head-to-head: 59.7% — 273 / 457 signals.

• BUY calibrated accuracy: 56.6%.

• SELL calibrated accuracy: 63.8%.

• Kelly criterion (half-Kelly): 19.5% optimal sizing.

• Model version: v5.3.0.

These numbers are self-reported but verifiable. Self-test diagnostics are accessible via the Oracle API.