DeltaX Evaluate · public researchEvidence stays bounded to each named domain and maturity.

Research & evidence

Research the architecture.
Inspect the evidence.

Three research projects investigate different parts of the decision loop: staying within available candidates, adapting under adversarial pressure, and responding to changing simulated conditions. Explore what each experiment shows, what failed, and what remains open.

01Connectome-derived computational research

Connectome Entity

Frozen Phase IV-C evidence · active research pause

Does selection stay inside what the substrate actually proposes?

Connectome-derived substrate → finite candidate behavior → external DeltaX governor → selected or refused result → simulated consequence.

What exists
A held-out 9,000-episode benchmark with 199,500 candidate-constrained selections and a 19-artifact evidence manifest.
Boundary
Idealized rate model, not biological equivalence; four of fifteen environments were not solved. Public tests do not reproduce the private runtime.
Explore Connectome research
02Chess research

Nyvera

Recorded training replays

How does a decision unfold under adversarial pressure?

Adversarial position → legal game evidence → candidate and correction gate → recorded move and outcome → replay trace.

What exists
Selected chess training replays with legal moves, recorded decision context, outcomes, and correction notes.
Boundary
Selected replays do not establish general chess strength, rating, or later learning improvement.
View Nyvera proof
03Vehicle-dynamics research

Kairos

Simulation only

How does the system respond when simulated conditions change?

Nonlinear simulated motion → simulator evidence → permission and recovery gate → bounded result → telemetry trace.

What exists
A digital vehicle-dynamics sandbox with versioned scenarios, changing conditions, telemetry, and replayable simulator evidence.
Boundary
No claim of physical readiness, autonomy, actuator safety, or field deployment.
View Kairos proof
No cross-domain inheritance.

Research evidence does not transfer to Evaluate API capability or launch.