Choice
execute | escalate | halt | other
- execute
- escalate
- halt
- other
One closed action for a pair-programmer step. The sidecar picks a label instead of drafting a paragraph you have to parse.
local System One decisions for tools and pair-programming
State + typed questions → Choice / Score / Noul, with the probability named as a probability. Runs on your GPU (Laya champion). Not a chat bot.
Primitives
execute | escalate | halt | other
One closed action for a pair-programmer step. The sidecar picks a label instead of drafting a paragraph you have to parse.
severity
An ordered severity level for the step in front of you. A rubric, not a vibe word from a chat model.
P(safe)
The probability the next action is safe to take. A named probability — not a hedge buried in prose.
Local champion
Laya is the local System One champion. /decide lives on your tailnet. This page does not publish a MagicDNS URL and does not curl a private host for visitors.
Champion stamp
CHAMPION
Cam stamps held-out bench before promote. No public MagicDNS.
How data becomes decisions
Public Hugging Face plus observed pair-programming traces. Honest inputs only — no synthetic champion rows.
Choice / Score / Noul factory: honest HELPER rows, Nimble contrastive flips (≤8 words), then staging JSONL.
Fine-tune, then a held-out bench. Cam stamps before anything is champion. Nothing auto-promotes.
What this is not
This is a decision sidecar, not a chat train control plane. The verb is decide — calibrated probabilities, not a transcript to parse.
LLMVM chat and watch-the-box-train dashboards are lab archive, not this product.
What we refuse
Older WSD / base1b / YaRN / J-Space / Alienware telemetry pages are lab archive. They are not the homepage story.