What we sell
Software access. A monthly or annual seat on the desk. Not a fund. Not a managed account. Not a broker.
Who we are
A desk you can audit.
CryptoVest Ventures, LLC builds CryptoVest Ventures: a predictive crypto research desk. We read live reference prices, volume, crowd votes, and Fear and Greed, then publish a directional pressure score. Every call lists the features that produced it. Misses sit next to hits.
The company is led by Mark Powers, Founder & CEO. The product goal is simple: tools that used to sit on institutional terminals, offered with language a newcomer can actually use.
What we do not sell
Live exchange execution, guaranteed returns, or equities/commodities/forex. Those stay labeled until they are real.
Why subscribe
Because the why is on every signal, the miss log is public, and paper fills at live prints with the SIMULATED banner on.
Desk charter
Not on faith.
The gentleman mark is a reminder that this is a research desk, not a call group. Mark Powers will not sell live execution, guaranteed returns, or managed money. Misses stay on the public ledger with the same weight as hits. The SIMULATED banner does not come off paper mode.
- Entity: CryptoVest Ventures, LLC. Product: CryptoVest Ventures. Domain: cryptovestventures.com.
- Contact: use the Terms page. Counsel has not signed off on legal copy.
- Crypto only in this release. Signals are probabilistic.
Method note
cv-pressure-v1.2 Method Note
Dated 2026-09-21 · last trained 2026-08-28 · production model cv-pressure-v1.2
This note describes the live pipeline. It is not a research paper, not a guarantee, and not investment advice. Weights below are the ones in production. If they change, the version string on the desk changes.
Features and why they are in the model
| Weight | Feature | Why it is here |
|---|---|---|
| 0.20 | 7-day return | Medium-horizon drift. A name that has already moved hard still informs pressure, with a cap so one spike cannot dominate. |
| 0.16 | RSI(14) | Stretched vs mid-range on a 0 to 100 scale. Reads recent average gains versus losses. |
| 0.14 | Volume anomaly | Signed volume z-score. Unusual participation with the move, not raw volume alone. |
| 0.14 | Distance from SMA20 | How far price sits from a short moving average. Trend context without a second model. |
| 0.12 | MACD histogram | Momentum of the moving-average spread, scaled to price so BTC and a smaller name share a unit. |
| 0.12 | Crowd sentiment | CoinGecko community up-vote share when present. Not a licensed social NLP feed. |
| 0.08 | Fear and Greed | alternative.me index. Market-wide mood, weaker weight than name-specific tape. |
| 0.04 | 1-day return | Overnight drift. Small on purpose so one session does not run the desk. |
Weights sum to 1.00. Perpetual funding is shown on the desk as context. It is not a term in this sum.
How pressure and confidence are computed
Each feature is scaled into a bounded reading, then multiplied by its weight. Pressure is the sum of those contributions. The three largest absolute contributions are the "why" on the card.
n_i = clip(scaled feature i, -1.5, 1.5)
pressure = Σ w_i · n_i
direction = up if pressure > 0.18
down if pressure < -0.18
else neutral
confidence = clip( sigmoid( 3.4 * (|pressure| - 0.08) ), 0.05, 0.97 )Confidence is how far pressure sits from a dead band. It is not the chance the next close pays. A 64 percent reading is not a 64 percent probability of profit.
Walk-forward rules
- The check uses daily closes. The model is scored on history up to day t, then compared to the move from t to t+1.
- Up is a hit if the next close is higher. Down is a hit if the next close is lower.
- Neutral is a hit only if the next move is under 1.5 percent. Neutrals count. They are not dropped.
- Hits and misses carry equal weight. The public ledger is not edited after the fact.
- The sample is a trailing window (about the last 45 eligible sessions after a 30-bar minimum), not the entire coin history.
Change log
- cv-pressure-v1.2trained 2026-08-28Current production weights and thresholds above. Prior labels are not restated here.
What the model is not
- Not a neural net and not a black box. Every signal lists feature contributions.
- Not execution-grade market data. Feeds are CoinGecko, Binance as fallback, and alternative.me.
- Not a broker, not custody, not a live order. Paper fills are simulated at live reference prints.
- Not a forecast of magnitude. Direction plus confidence, then the next close is the score.
- Not advice. Predictive signals are probabilistic, not guaranteed. Past performance, whether backtested or live, does not guarantee future results. You can lose money, including the entirety of funds allocated to automated strategies.
Is 51 percent useful?
The blended board often sits near 51 percent. BTC has run higher in the same window. A 51 percent coin-flip with no process is not a desk. A 51 to 64 percent score with a why, a next-day check, and a size rule is a process. Playbooks on Learn tell you to size 2 to 4 percent of paper equity, treat a tight cluster as one bet, and keep misses on the tape. If you need a 90 percent call group, this product is the wrong seat.