Know the evidence
Inputs remain bound to origin, timing, transformation and the exact agent version that used them.
Research standard
Make uncertainty visible. Make decisions answerable.
We publish the principles that define credible work while keeping exact models, thresholds, data configurations and operational evidence inside the protected laboratory.
Six research principles
The standard behind every agent.
Forecast before opportunity
The underlying prediction is produced without using the market price as a forecasting feature.
Measure probability quality
A stated probability must be evaluated by region, horizon and scenario — not judged by a few memorable outcomes.
No action is a valid result
Missing evidence, disagreement, unclear definitions or inadequate net value should stop a decision.
Strong objections can veto
Required reviews cannot be replaced by enthusiasm, majority vote or a higher confidence label.
Preserve what was known
Decisions and outcomes are evaluated against the original evidence cutoff, without fitting history after the fact.
A good process can still lose. A bad process can still get lucky.
Short-term outcomes are evidence, not proof. Research quality comes from repeated calibration, realistic costs, preserved decisions and explicit failure analysis.
Operating modes
Research, paper validation and live operation are not interchangeable.
The platform treats each mode as a distinct contract. Passing one stage does not silently authorize the next.
Historical replay
A decision is reconstructed only from information that would have been available at the original cutoff.
Current, simulated execution
Real-time decisions and market conditions are studied using virtual capital and controlled accounting.
Separately governed
Any real-capital operation would require an independent approval, risk budget, access model and execution evidence. It is not publicly offered today.
What we disclose
Transparency without operational leakage.
Public communication should make the philosophy, agent purpose, validation stage and limitations understandable. It should not hand over the exact recipe used to produce or size a decision.
Public
- Agent purpose and maturity
- High-level research principles
- Validation modes and limitations
- Safety and abstention philosophy
Protected
- Exact model and feature configuration
- Decision and sizing thresholds
- Source mappings and operational timing
- Run telemetry, infrastructure and internal APIs
Research access
The public standard is here. The operating evidence stays in the lab.
Approved users can enter the authenticated environment. Public pages deliberately do not request internal APIs or expose research telemetry.