Bitcoin Market Intelligence

What you're seeing: This shows how far Bitcoin has fallen from its highest price ever. For example, -50% means Bitcoin is half the price of its peak. Bigger drawdowns often represent better buying opportunities, but also indicate higher risk. Bitcoin has historically recovered from all drawdowns, but past performance doesn't guarantee future results.

Historical context: Bitcoin has experienced multiple 70%+ drawdowns in its history, but has always eventually reached new all-time highs.

What you're seeing: Mayer Multiple and power-law residual Z-score share one time axis, making it easier to compare two independent valuation regimes. Dashed thresholds mark historical context; they are reference levels, not trading instructions.

What this means: Real-time data about Bitcoin network activity. The mempool is like a waiting room for transactions. When it's crowded, fees go up. This helps you understand current network demand and can be a proxy for market interest in Bitcoin. (Data from local node)

History window: 1462 days (2022-07-10 → 2026-07-10).

Mempool stats source: fallback (4y).

LSTM training window (overlap only): 2878 days (2018-08-23 → 2026-07-09).

What this means: This monitors how accurate our AI predictions have been recently. Think of it like a report card for the model. If accuracy drops, we may need to retrain it with newer data. This helps ensure you're getting reliable forecasts.

What this means: This summarizes how the neural network was run, what evidence supports it, and which model controls are active.

What you're seeing: This shows 100,000 possible price scenarios for the next 180 days based on Bitcoin's historical behavior. The teal center line is the median outcome. The shaded areas show where the simulated price paths cluster:
• Central 50%: the range from p25 to p75
• Central 90%: the range from p05 to p95

This is like a weather forecast - it shows a range of possibilities, not a single prediction.

Remember: These are probabilities, not certainties. Use this to set realistic expectations and plan for multiple outcomes.

Observed history provides context before the forecast boundary. The LSTM line connects only supplied decision horizons; the shaded 50%, 80%, and 95% bands show increasing uncertainty.

Forward-looking evidence: The LSTM forecast is probabilistic scenario evidence, not guaranteed targets.

Note: Holdout intervals are over-conservative (very high coverage with wide ranges) for 60d, 90d, 120d, 150d, 180d. Treat these horizons as scenario planning only, not precise targets.

Each panel starts from the same current-price reference and uses the same USD scale, dates, and forecast horizon. Small multiples keep model paths comparable without overlapping every line in one plot. Source uncertainty intervals remain visible when supplied.

Preferred-horizon evidence combines ensemble weight with sample size, error, interval coverage, and interval width. Low status means the evidence misses at least one confidence gate. Frozen holdout evidence takes precedence over cross-validation, which takes precedence over ordinary walk-forward evidence; one row is retained per participating model.

What this means: This shows how accurate each prediction model has been in recent history by testing them on real past data (36 tests from 2025-05-05 to 2026-01-05). Lower error numbers mean better accuracy.

How to read this table:
• MAE/RMSE: Average dollar error - lower is better (shows typical prediction accuracy)
• MAPE: Error as a percentage of price - lower is better
• PICP 95%: How often actual price fell within prediction range - closer to 95% is better
• Width %: How wide the prediction range is relative to price - narrower is more precise

What this means: This tests the pre-holdout evaluation ensemble on later origins without retraining. The live forecast is then refit on all available data using the same frozen architecture, hyperparameters, epochs, and seeds, so these metrics assess the pipeline design without leaking evaluation outcomes into training.

Origins: 26 · Stride: 7 · Paths/origin: 500 · 2025-07-15 → 2026-01-06

Method: Direct horizon head (frozen walk-forward)

What this means: Raw PICP measures out-of-sample interval coverage. Calibration-fit PICP only shows how closely post-hoc interval scaling matched its target; independent holdout and frozen-origin results determine whether the interval is trustworthy.

Calibration-fit vs Raw: Calibration-fit coverage is descriptive because scaling targeted 95% on those observations. Raw coverage is the independent evidence used for reliability scoring.

Origins: 36 · Stride: 5 · Paths/origin: 1000

What this means: Here's what all the data above tells us in plain English. These are the most important things to know right now for making investment decisions. Read these carefully - they're written specifically to help you understand what to do next.

• • Log growth model suggests significant undervaluation
• • Market in sideways consolidation - accumulation opportunity
• • Holdout intervals are over-conservative (very high coverage with wide ranges) for 60d, 90d, 120d, 150d, 180d. Treat these horizons as scenario planning only, not precise targets.
• • Holdout precision is weak (wide intervals or high CRPS%) for 14d, 30d, 60d, 90d, 120d, 150d, 180d. Use these horizons as broad risk ranges, not precise targets.
• • Trust gate removed 5 internet feature(s); 0 passed walk-forward validation.
• • Cycle: Early Bull (7% complete) - New uptrend potentially starting. Good time to build positions or add exposure.
• • LSTM Scenario (60d): Moderate simulated incidence (49.4%) of a local peak. Watch risk posture.
• • LSTM Scenario Alert (90d): Simulated paths frequently form a local trough first (Aug 20, 67.7%), followed by a local peak (72.0%). Treat this as a volatility scenario, not a precise timing signal.
• • LSTM Scenario Alert (120d): Simulated paths frequently form a local trough first (Aug 30, 81.2%), followed by a local peak (84.3%). Treat this as a volatility scenario, not a precise timing signal.
• • LSTM Scenario Alert (180d): Simulated paths frequently form a local peak first (Sep 05, 95.5%), followed by a local trough (95.5%). Treat this as a volatility scenario, not a precise timing signal.
• • Strategy: Consider dollar-cost averaging (DCA) to manage entry prices and volatility.
• • Holistics: Always combine these model insights with your own research, risk tolerance, and broader market analysis (fundamental & macroeconomic factors).
• • Disclaimer: Past performance and model projections are not guarantees of future results.