Trading agents, from research to autonomous markets
We are building the systems that turn model reasoning into persistent, owner-directed trading. The frontier runs through the harness: what an agent can observe, how it interprets a mandate, which actions it may execute, and what it retains after the outcome.
DXRG brings a deployment record to that problem. Our 21-day real-capital study involved 3,505 funded agents, about 7.5 million invocations and roughly 70 billion inference tokens. Our continuous research record extends the investigation into a historical Hyperliquid fleet with 231,638 finalized turns and 14,596 fills, including 5,035 real-money fills. Its mostly paper environment and directional-edge null remain part of the finding.
The work reaches beyond those fleets. A September 26 archive inventory records 4.15 TiB across 73 crypto-market table families, including Solana and Base. Available extracts contain 7.5 million hourly bar rows and 9.5 million minute-swap window rows; these are different aggregation units, with incomplete coverage and potential overlap. Our exact-L4 representation research and ECTO’s Solana transformers examine market state, labels and validation from different directions. Read the aggregate research audit for dated counts, methods and results.
Our competitive thesis is direct: as capable research workflows become broadly available, firms whose edge rests mainly on routine public-signal analysis face pressure. Capital, execution, relationships and infrastructure still matter. The articles below develop that thesis through conditional projections for hours-to-days trading, open crypto markets, tokenized assets and specific quant tasks. Their timelines are DXRG scenarios, with explicit conditions, rather than announcements about a named firm’s staffing.
DXAP applies this research through persistent agents, owner-reviewed direction, configured execution checks and inspectable decisions. We evaluate intent and strategy alignment alongside prediction and economic outcomes. A lucky profit cannot establish that an agent followed its owner’s strategy; a compliant loss still deserves economic scrutiny.
Trading agent theory
- The harness defines the trading decision system
A transition-system formulation explains why an identical model and prompt can still describe different trading agents.
- Why repeated reasoning can turn an invented rule into policy
Historical rule fabrication suggests a theory of authority promotion in persistent agents, with a bounded reading of the 57% to 3% intervention.
- A candidate list changes what an agent strategy means
The historical render boundary makes strategy attribution depend on the opportunity exposure process.
- Confidence needs a proposition before it can influence an order
A four-layer argument separates expressed certainty, calibrated event probability, economic assumptions and execution permission.
- Different owner strategies can still share one market exposure
A fleet-level theory explains how common observations can dominate mandate diversity without agents communicating.
- A repeated rationale does not accumulate evidence
A persistent-agent theory separates the number of reasoning turns from the number of independently validated observations.
- Social market data must keep its own authority
Prompt injection becomes a trading-agent alignment problem when source content can redefine what an authorized account should do.
- When agent activity becomes another agent’s signal
A future research hypothesis separates common-shock convergence from market-mediated self-excitation and crowding.
- The product objective is to fulfill the owner’s intent
A delegation acceptance test separates a useful trading service from an account that happened to make money.
- A correct loss and a wrong profit can occur in the same strategy
Outcome-blinded conditional review tests whether an agent followed the decision the owner authorized.
- A trading-agent benchmark needs an explicit utility function
A constrained evaluation design keeps preferences, restrictions, execution and economics visible instead of hiding them in one score.
- A persistent mandate can adapt to markets without changing its purpose
An authority-preserving regime test distinguishes changing tactics from an agent quietly replacing the owner’s objective.
Market predictability
- A useful volatility forecast can leave direction unresolved
Conditional variance and directional probability answer different questions. Their separation gives agent research a more productive target.
- Market microstructure predicts consequences close to the order
An order-flow impact relationship helps define execution research, provided observed same-interval events stay separate from future information.
- Match the market forecast to the clock that can use it
Seconds, hours, days and months need different targets, information sets and decision mechanisms. A horizon map makes those differences inspectable.
- A directional null narrows the claim and improves the research agenda
DXRG’s continuous record rules out easy claims about its historical fleets. Its strongest use is a precise domain of failure, not a universal impossibility theorem.
- Monthly stock returns contain signal, even when most movement stays unpredictable
Gu, Kelly and Xiu show why nonlinear interactions improve expected-return measurement and why the horizon matters.
- The best forecast and the best portfolio solve different problems
Kelly and Xiu’s survey explains why trading costs and persistent signals change the objective for financial machine learning.
- Predictability has a location: what Mosaics of Predictability actually tests
The April 2026 paper separates predictable equity clusters from weak ones, with different treatment of cross-sectional and regime evidence.
- A return paper and a profitable agent require different receipts
A four-question translation audit connects academic predictability to the full decision system without importing a paper’s performance claim.
Quant work and open markets
- Persistent personal research becomes an ordinary capability
Our 2026–2031 scenario makes a research agent broadly accessible, with adoption measured by usable persistence rather than model subscriptions.
- Cheap analysis makes owner intent more consequential
A fixed-evidence experiment shows how judgment shifts toward selecting objectives and resolving conflicts in a trading mandate.
- Common agent infrastructure can support different strategies
A paired-mandate experiment distinguishes shared model behavior from the different objectives of the owners using it.
- Cheaper agents change the equilibrium price of research
The Grossman–Stiglitz information-cost mechanism gives our agent adoption scenario a falsifiable market-efficiency prediction.
- The vulnerable quant business sells a workflow everyone can reproduce
Our competitive scenario identifies the hours-to-days research businesses exposed to cheap public-signal agents and a test that could disprove it.
- Shared models leave capital, access and execution unequal
An advantage decomposition explains why cheap research can threaten some quant businesses while strengthening structurally advantaged incumbents.
- Open crypto markets make entry cheaper and integration more valuable
A fragmentation experiment distinguishes permission to enter from the ability to trade economically across pools and chains.
- Tokenization moves the boundary between research and settlement
A claim-to-redemption map identifies the new agent tasks created when asset transfer and conditional settlement become programmable.
- Quant research will automate experiment production before research ownership
A conditional timeline separates cheap experiment production from reliable discovery and capital allocation.
- Quant trader automation has two clocks: execution and risk ownership
The timing of bounded execution differs from the timing of responsibility for capital and market stress.
- Quant engineering automation advances through verified changes
A task-based timeline for software, data and ML engineering explains why implementation speed and production ownership diverge.
- A 2026-2031 scenario for automated quant teams
Integrated task bundles explain how smaller teams could compete and why complete role replacement has no single deadline.
DXRG research program
- DXRG research credentials come from operating agents and testing the machinery
A protocol comparison explains the research questions answered by stock simulations, real-capital deployments and continuous agent records.
- The scale of DXRG research is several serious programs, each with its own unit
A dimensional account separates funded agents, runtime tokens, fleet turns, archive bytes and representation rows.
- Our failed hypotheses make the DXRG research program stronger
Specific retractions and development audits show how rejected claims improve the next experiment.
- DXRG research gives DXAP an operating advantage you can inspect
The research-to-product case rests on mandate, policy and outcome mechanics rather than a list of AI features.
- Inside ECTO: a state lattice for token microstructure
What a 65,536-cell market representation and a 256-cell selection teach us about building a microstructure transformer.
- ECTO accuracy changes meaning when the flat baseline is visible
An 87.26% development accuracy becomes a sharper research result when class prevalence and threshold support are reported.
- ECTO taught us that a split must preserve membership
A development audit explains why a shared random seed and a held-out label leave two separate independence questions unresolved.
- ECTO labels revealed a decision-time alignment problem
Why measuring future mean price from the input opening can mix already observed movement with the outcome a trading decision faces.
DXRG publishes this research and develops DXAP. AI assisted research and drafting; published under DXRG’s editorial responsibility. Read the complete field-note library or send corrections to hello@dxrg.ai.