Early Public Beta
Tantoryn AI
Evidence-driven AI/ML trading intelligence for crypto futures markets.
Tantoryn AI researches, tests and observes automated trading intelligence in the open. Structured market-candidate generation, independent ML evaluation and a dedicated risk gate work together — and every claim on this site is backed by current evidence, not a promise.
What Tantoryn AI Is
A research-driven system, not a signal shop.
Tantoryn AI applies machine learning to the analysis of crypto futures markets through a disciplined, staged pipeline. No output reaches a user before it has passed through structured market analysis, independent ML evaluation, and a dedicated risk and lifecycle gate.
The system currently runs in Early Public Beta: it observes live markets and executes the full pipeline end to end, but every trade is virtual. Real-money order execution is disabled. The current objective is to prove the pipeline's quality with evidence — not to claim a finished profitable product.
How the System Is Structured
A four-stage decision pipeline, built so that no single component can act alone.
Market Data
Continuous ingestion of multi-timeframe market data across tracked instruments forms the shared factual basis for every downstream decision.
Candidate Generator
A dedicated layer identifies formalized market candidates from that data — structured conditions worth evaluating, not yet trade decisions.
ML Decision Layer
Machine-learning models evaluate each candidate. Models can score and rank; they cannot open a trade or bypass risk control on their own.
Risk / Lifecycle Engine
An independent gate applies risk checks, exposure limits and lifecycle rules before anything is allowed to become a signal.
Signal / Virtual Trade Observation
Only after every stage agrees is a virtual trade recorded and observed — delivered publicly through Telegram.
Why Tantoryn
Six principles that shape every engineering and research decision.
Candidate-first intelligence
ML never invents a trade idea from nothing — it evaluates candidates that a separate, formalized layer already produced.
ML does not control risk
Risk and lifecycle management sit in an independent engine that ML cannot override.
Evidence before promotion
No model or method advances toward Production status without documented, reproducible evidence.
Research and Production are separate
A research result is never treated as a Production claim by default; promotion is a distinct, deliberate decision.
Failed experiments are evidence
Rejected methods and negative results are recorded and published — they are part of how the system improves.
Controlled disclosure
We publish process, methodology and status in the open, while proprietary logic and thresholds stay closed by default.
Research Integrity
Financial machine learning fails quietly — usually through leakage, not bad luck. Our process is built to catch that.
- Chronological validation. Data is always split and evaluated in time order; nothing from the future leaks into training or selection.
- Causal reconstruction. Features and labels are rebuilt to reflect only information that would have been available at decision time.
- Leakage control. Preprocessing such as scaling is fitted only on training data, with explicit checks against shared-fit shortcuts.
- Frozen evaluation rules. Evaluation contracts are fixed before a result is seen, so the rules cannot be adjusted after the fact.
- Robustness testing. Accepted methods are stressed against shifts, noise and regime change before being trusted further.
- Independent audit. Material results are checked by a separate reviewer who did not build the thing being reviewed.
Proof of Work
What is verifiably running today — no invented performance figures.
Live market runtime
The pipeline runs continuously against live market data.
Virtual trade execution
Trades are simulated and tracked end to end; no real orders are placed.
Lifecycle tracking
Open positions are tracked through their full virtual lifecycle, not just at entry.
Telegram delivery
Observed activity and reporting reach the public Telegram channel.
Reporting
Users can request virtual trading history and performance summaries through Telegram.
Research & audit workflow
Independent review is standard practice: most material model or method changes go through documented research and a separate independent audit before final approval, with a limited set of scope-based exceptions.
Real-money orders disabled
Real-money order execution remains hard-disabled at the runtime level.
Evidence, Not Claims
The Live Reports Are the Trust Asset
Everything above is a description of a process. This is the observable result of that process, updated live, from the same history the runtime itself produces.
Open Live ReportsFollow Tantoryn AI as It's Built
The project is public by design. Here is how to follow it.