One fused number per coin, from every engine the desk runs. This page names every input, its weight, how calls are logged, and how they are graded. What is proven and what is still being validated are labeled separately.
Each component is mapped to -100..+100 first, then weighted. Weights sum to 105, verified at page build from solscore.py.
| Component | Weight | What it measures |
|---|---|---|
| patterns | 30 | pattern-engine score (-8..+8) |
| tradingview | 15 | TV gauge 1D/4H ratings |
| etf | 15 | BTC/ETH spot ETF streak (Farside) |
| options | 10 | Deribit put/call OI (BTC/ETH) |
| polymarket | 10 | crowd upside odds vs dip odds (XRP/BTC/ETH) |
| momentum | 5 | 7d price change |
| funding | 5 | perp funding, contrarian (crowded longs = bearish) |
| orderbook | 5 | Kraken L2 bid/ask imbalance |
| coinbase | 5 | Coinbase premium (US bid vs world) |
| social | 5 | Threads crowd net sentiment (idea #160; hourly |
The same cut points everywhere: the site, the YouTube scripts, and the paper trading portfolios.
| Score | Label | What it means |
|---|---|---|
| 60 to 100 | Strong Bull | Composite leaning hard into the move |
| 20 to 60 | Bull | Green, not screaming (YouTube reads this "constructive") |
| -20 to 20 | Neutral | The score does not trust the move |
| -60 to -20 | Bear | Fade the move |
| -100 to -60 | Strong Bear | The composite wants out |
Every hourly briefing publishes one verdict per coin (bullish, bearish, or neutral) with a stance, a conviction level, and an entry price. Each is logged before the outcome is known.
Each call is graded 24 hours after it is made, against live Kraken spot. The 24-hour grading delay is enforced in scorecard.py.
Bullish wins if the 24h forward return exceeds +1.0%, loses below -1.0%, otherwise pushes. Bearish is mirrored. Neutral calls never win or lose: they are reported as pushes.
Each call may carry a kill price. A long dies if the 24h low touches it, a short if the 24h high does. Killed calls count as losses and are tracked in their own bucket.
Win rates are also reported conviction-weighted: high counts triple, medium double, low single. Sizing the score to the stated confidence keeps big claims honest.
Describe what is true, label what is still being tested.
Calls are logged before outcomes, graded on a fixed rule, and published with wins and losses. That is a process claim, and the process is running.
The Pulse Score's predictive value is unproven at any sample size. Paper trading two strategies against the published scores is the live test, shown on the track record page.
to_exchange / off_exchange labels are behavioral proxies from a small verified wallet list. Their accuracy against subsequent price action has not been audited yet.
A 60% win rate on 25 calls can easily be luck. Early numbers are a baseline being built, not a verdict. The longer the ledger runs, the harder it is to fake.
Methodology generated Sep 24, 2026 21:03 MDT from solscore.py and scorecard.py. Weights verified to sum to 105 at build.