One command centre for Bitcoin
GT3 AI answers a single question, continuously: where is Bitcoin likely to go next, and what is the live flow doing right now? It does this by running two engines side by side over the same live market feed.
- Order-flow analytics — reads the live tape and order books across every major venue and distils them into directional pressure, institutional order activity, funding and positioning.
- Probabilistic forecasting — simulates thousands of possible price paths per horizon and returns the full distribution of outcomes, not a single guess.
Everything is derived from public exchange data in real time. There are no black-box promises — every number on the screen traces back to a defined calculation over observable market state, refreshed on a fixed cadence.
A fault-tolerant pipeline
The engine is built as a real-time streaming pipeline. Live data flows in from the exchanges, splits into the two analytical engines, converges into a signal layer, and is served to the browser terminal. Every external fetch is isolated, so a single venue timing out never stalls the system — the cycle simply proceeds with whatever data returned.
How a forecast is reached
Under the hood, GT3 AI runs like a coordinated team of quantitative specialists — each one an automated engine in the pipeline. This is the path every forecast travels, from raw market data to a live, validated call.
Four AI quant engines work in parallel — mathematical models and complex formulas run over the raw data at once.
Two master quant engines weigh those outputs against one another and resolve them into a single decision.
Two further quant engines re-check the signal against live data, keeping it as accurate as possible before it reaches you.
A final formula reconciles the original modelled data with the newest live data, projecting the forecast forward in real time.
A conversational AI research assistant that sits on top of this workflow and answers your questions in plain English — what the current signal means, where the order flow and positioning are pointing, and how the risk sits — grounded in the same live data the engine runs on. Arriving as part of the GT3 AI V2 update, included with your licence.
Every venue, read as one market
Each cycle, the engine fires many independent requests across every venue and both market types — spot and derivatives — then assembles them into a single consolidated view. Because Bitcoin trades in fragments across many books, GT3 AI reads them together rather than trusting any one exchange.
- Tickers — best bid / ask and last price per venue, averaged into a single consolidated spot.
- Order books — level-2 depth, used for imbalance and wall detection near the mid.
- Trade prints — the raw tape, classified buy / sell and sized in USD.
- Derivatives — perpetual books and tape, funding rates, and open interest.
Fault tolerance is a first-class design goal: every fetch is wrapped so an error returns a marker instead of raising. A venue that is slow, rate-limited or down is simply skipped for that cycle, and the analytics degrade gracefully rather than breaking.
Turning the raw tape into pressure
The analytics engine converts the firehose of trades, books and derivatives data into a small set of decision-ready reads. Each is a defined calculation over the live snapshot.
Cumulative Volume Delta
Buy-side minus sell-side USD across the tape, per venue and market — the net aggression behind price, plus VWAP.
Order-book imbalance
Resting bid vs ask liquidity near the mid-price, with detection of the largest bid and ask walls nearby.
Funding regime
Perp funding classified into neutral, long-paying, short-paying or extreme — a read on crowded positioning.
Open-interest shift
Rolling change in open interest per venue, flagging fast build-ups or unwinds of leverage.
Institutional order detection
Individual large institutional-size block prints on Binance spot and futures, surfaced the moment they hit the tape, with a live buy/sell split — retail-size trades filtered out.
The composite pressure score
These individual reads are fused into one directional number — the figure the terminal shows as CF-PRESS. Spot flow, perpetual flow, book imbalance and funding are each scored per venue, weighted by how informative each is, then combined and normalised into a single directional score. The result is labelled from Strong Sell through Balanced to Strong Buy, so the state of the whole market reads at a glance.
The exact weights, thresholds and horizon caps that tune these calculations are proprietary and are not published here. This document describes the mechanism, not the calibration.
Simulating the range of outcomes
GT3 AI does not output a single predicted price. For each horizon it runs a large-scale probabilistic simulation — thousands of independent price paths — and reports the resulting distribution: the median, the confidence bands, and the probability of finishing higher.
The model is deliberately grounded in the statistical behaviour of crypto rather than a naive bell curve. Two design choices matter most: it adapts to the latest volatility conditions rather than averaging them away, and it is built to expect outsized, extreme moves, because Bitcoin produces far more of them than a simple bell-curve model would predict.
Read recent history
Pull the recent candle series for the horizon and convert closes to log returns.
Gauge current volatility
Read how volatile the market is right now, weighting the most recent activity most heavily so the model reacts to regime changes instead of averaging them away.
Set direction, carefully
Estimate a disciplined directional bias with built-in safeguards, capped per horizon so no forecast runs away on a short-lived trend.
Simulate thousands of paths
Project the price forward step by step with extreme-move-aware dynamics scaled by current volatility — thousands of independent paths per horizon.
Extract the distribution
From the simulated endpoints, compute the median path, the confidence bands, the full terminal histogram, the probability of finishing up, and the 95% range.
Given the same market state, the engine reproduces the same forecast — the output is auditable, not a different guess every refresh.
What the engine flags in real time
On top of the continuous reads, the engine watches for discrete events worth a trader's attention and raises them as they happen. Each alert type is debounced with a cooldown so a busy tape doesn't flood the stream.
A band breach is worth calling out: because the forecast produces confidence ranges, the engine can flag the moment price trades outside its own modelled band — an early signal that conditions have shifted faster than the model expected.
Everything on one screen
All of the above is surfaced in a single browser terminal that refreshes on its own. The operator never runs a query — the current state is always on screen.
- Consolidated spot & targets — the cross-venue price with the horizon's median target and expected move.
- Forecast fan & step table — the probability cone per horizon, with a step-by-step table of median and confidence ranges.
- Venue matrix — live price per venue, up/down at a glance.
- Pressure gauge — the composite CF-PRESS score and its buy / sell label.
- Institutional & alert stream — large block prints and events as they fire.
- Simulated paths — a live sample of the model's trajectories behind the bands.
Hosted, licensed, locked
GT3 AI is delivered as a hosted service. Customers sign in through the browser — there is nothing to install, and the engine and its methodology stay server-side and are never shipped to the client.
Encryption & security layers
A license activates on first sign-in and binds to that device's fingerprint; moving machines requires an admin reset. This keeps a single-seat license single-seat without exposing any of the forecasting logic to the end user.
What it is — and isn't
The methodology has been backtested and live-tested, and every figure the terminal shows is a defined, reproducible calculation over observable market data. That transparency is the point: there are no invented accuracy percentages in this document, because the honest output of the engine is a probability distribution, not a promise.
GT3 AI is a quantitative research and monitoring tool. Its forecasts are probabilistic model outputs — not signals to buy or sell, not financial advice, and frequently wrong. Real markets can and do move outside modelled ranges. Backtested and live-tested performance is no guarantee of future results, and trading Bitcoin and its derivatives carries substantial risk of loss.