Signall
10 OpenEnv cognitive primitives that train generalist agents.
Core trick
Ten OpenEnv environments each isolate one cognitive primitive, and Unsloth GRPO trains on live HuggingFace Space reward signals.
Tags
Description
10 OpenEnv cognitive primitives that train generalist agents. LLM fine-tuned via Unsloth + GRPO on live HuggingFace Space reward signals.
README excerpt
# [~] Signall — Cognitive Primitives > *cognition with rails* **OpenEnv Agentic Training Platform** — 10 RL environments that isolate the cognitive primitives underlying every economic activity. Train any agent on the building blocks of intelligence. The skills transfer to anything. Built for the [OpenEnv Hackathon SF](https://cerebralvalley.ai/e/openenv-hackathon-sf) — March 2026. --- ## The Thesis Every business activity — trading, negotiating, allocating capital, detecting fraud, managing risk — decomposes into a small set of cognitive primitives. Current RL trains agents on final tasks. We train them on the building blocks. An agent that scores 90%+ across all 10 primitives has the cognitive foundation to operate autonomously in any economic context. ## Line S — Embarcadero to Lands End | Station | Environment | Cognitive Primitive | Economic Application | |---|---|---|---| | Embarcadero | The Bandit | Exploration vs exploitation | Portfolio allocation & market testing | | Powell St | The Sequence | Pattern recognition | Trend prediction & time series | | Fog Basin | The Map | Planning under uncertainty | Logistics & route optimization | | Montgomery | The Auction | Resource allocation | Capital deployment & pricing | | Civic Center | The Tower | Dependency resolution | Project execution & task sequencing | | Nob Hill | The Signal | Attention & filtering | Fraud detection & anomaly recognition | | Market St | The Negotiation | Theory of mind | Deal-making & count