Evolutionary Test-Time Compute

Adapt → Propose → Cascade → Prune (Θ + decay) → Train → Build primitives

Run an evolution to populate the selection tree.

Evolution parameters

3
3
2
0.08
0.25
24

Ecosystem tools & skills

Ground the run in real evidence before it evolves. Leave everything unselected to let the loop choose its own tool calls.

Metabolic worker

F(t) = F₀·e^(−λΔt)

Background sweep decays unused lineages off the main thread; useful recalls inject fitness.

Awaiting first sweep…

Policy weights (REINFORCE)

decomposew 0.000 · π 16.7%
analogizew 0.000 · π 16.7%
simulatew 0.000 · π 16.7%
constrainw 0.000 · π 16.7%
invertw 0.000 · π 16.7%
synthesizew 0.000 · π 16.7%

Attention memory (top-K)

No adapted state yet.

Built primitives

Primitives crystallize after a run completes.

AI Disclosure: This platform uses generative AI. All outputs are AI-generated, may be inaccurate or incomplete, and are provided for informational purposes only. They do not constitute legal, financial, tax, medical, psychological, engineering, or other professional advice, and no professional relationship is created by use of this service. Verify all outputs with a qualified professional before relying on them. Do not input regulated data (PHI, PII, MNPI, ITAR/EAR, CUI, privileged material) unless your organization has explicitly validated this surface for such use.
Virtual tokens • Educational purposes only
© 2026 Kineteq.ai