MVTRX
SN79A next-generation exchange built specifically for subnet tokens
A C++ market simulator runs thousands of orderbooks, and miners compete as trading agents inside them. Best risk-adjusted strategies win.
// Synthetic markets. Real agents. Risk-managed scoring.
MVTRX, also branded as taos, runs an agent-based simulation of automated trading. Validators host a C++ simulator that maintains many limit order books populated by background agents. Miners plug in as trading agents and compete on risk-adjusted performance over the simulated runs.
The simple version: It's a quant trading sandbox at network scale, where miners run their own strategies against a simulated market and the best risk-adjusted performers earn the most.
Centralized equivalent: The closest analogs are the internal market simulators quant funds build in-house, plus research projects like JP Morgan's synthetic market data work. Public open-source equivalents are rare.
How it works:
- Miners act as trading agents that receive simulation state updates and submit orders, with rewards tied to a Kappa-3 risk-adjusted return measure and a minimum trading volume requirement
- Validators run the C++ simulator that maintains the orderbooks, forward state to miners, process the order instructions miners send back, and score performance
- The problem it solves: Realistic market data for strategy research is locked inside large funds and vendors. Public L3 (market-by-order) datasets barely exist, which makes training risk-aware trading agents difficult outside well-resourced institutions.
- The opportunity: A simulator that produces full L3 books across many parallel realizations, paired with a tournament of trading agents, is a rare combination of synthetic data and adversarial benchmarking.
- The Bittensor advantage: Many independent miners with different strategy assumptions are more useful for testing market microstructure than any single internal team. Validators host the simulator, miners bring the diversity of agents.
- Traction signals: Per the README, simulations target ~40 orderbooks today with the stated aim of scaling to 1,000+. The test branch shipped version 0.4.2 on May 20, 2026 with a scoring and chain-commitment overhaul and a new feature called GenTRX (distributed training of agents). External coverage on May 19 noted that MVTRX shipped a demo video and GenTRX testnet progress. The team is moving on a roughly weekly cadence.
Category: Other: Synthetic Market Simulation and Trading Strategy Research | Centralized Competitor: Two Sigma, Jane Street, and other in-house quant simulators; JP Morgan synthetic data research
The market for AI-driven trading is enormous and almost entirely private. Funds spend heavily on internal simulators because realistic market microstructure is hard to reproduce: you need many interacting agents, a matching engine that behaves like a real exchange, and enough parallel realizations that performance numbers are statistically meaningful. MVTRX is an attempt to do that work in the open, with a Bittensor incentive layer on top so miners compete to be the agents.
Mechanism:
According to the project's README, validators run two components. The first is a C++ simulator, built on the MAXE engine, that maintains the orderbooks and processes every L3 message. The second is a Python validator that proxies between the simulator and the Bittensor network: it forwards state updates to miners, takes their order instructions back to the simulator, and computes scores. Background agents inside the simulator create baseline market conditions; miners and background agents trade against each other on the same matching engine.
Scoring is documented as a weighted sum of risk-adjusted performance measures. Currently the live setting is an intraday Kappa-3 ratio (the README confirms this; a February 2026 commit message records the activation of a 79%/21% kappa/pnl weighting). Miners must also maintain a minimum level of round-trip volume, which prevents passive strategies from gaining incentive. Response latency is part of the simulation too: slower miners face more events between submitting and execution, so realistic slippage applies. New simulation configurations roll out roughly weekly, and miner scores carry over via a rolling window rather than resetting per run.
On-chain numbers reflect a small but still-emitting subnet. The alpha token trades at 0.01066 TAO with a market cap around 47,231 TAO and a TAO pool depth of 11,653 TAO. Current emission share sits at 1.02% and the smoothed EMA is essentially the same, so emissions are stable rather than accelerating. The 30-day price change is +116%, the 7-day is -14%, and net flows over the last seven days are negative at about 1,296 TAO. Under the current Taoflow model, sustained negative net flows compress emission share over time, so the recent week is worth watching.
GitHub is active but concentrated. The public repository at taos-im/sn-79 was last pushed on May 20, 2026, two days before this writeup, with 113 total commits from a single primary contributor. The main branch carries 0.3.x weekly releases through April 15, while the test branch carries the in-progress 0.4.x line including the GenTRX scoring overhaul. Releases follow a clear semver discipline, and commit messages document exactly what changed (book count increases, scoring parameter activations, training infrastructure). Public commentary from May 19, 2026 corroborates that GenTRX testnet work is in flight.
A miner-burn parameter of 79.6% is set on this subnet. That means the bulk of the miner share of emissions is currently routed away from miners. This is unusual and warrants understanding before participating as a miner.
- Concentration: The miner-burn parameter sits at 79.6%, so most of the miner emission share is burned rather than paid out. The Gini coefficient across the top 100 stakers is 0.786, which indicates concentrated stake distribution. The public GitHub also shows one primary contributor across all 113 commits, which means execution depends heavily on a small core team.
- Liquidity: TAO depth in the pool is around 11,653 TAO and the 7-day price change is -14%. Net flows over the last seven days are negative at roughly 1,296 TAO. Under Taoflow, sustained negative net flows would reduce the subnet's share of network emissions over time.
- Execution: Public commentary from May 2026 describes MVTRX as "internal testing kicked off on Bittensor mainnet, mainnet beta opens to the public once DAO smart contract builds are complete." GenTRX, the distributed-training feature in the 0.4.x test branch, is new and not yet on the main branch. The project is shipping, but it is not yet at the fully-public-product stage.
- Competition: Several Bittensor subnets sit adjacent to trading, market data, and financial signals. MVTRX is differentiated by being a simulator and tournament rather than a forecast feed, but it will need to keep proving that synthetic market data is useful to outside consumers, not only to its own miner tournament.
Into the next one.