User (2026-06-22, mid-build) issued a direction pivot.
** Use Galpin's σ-deviation for the PRIMARY signal (it's the public-domain, single-factor, transparent), but surface the secondary signals (sleep, training load, RHR) as modifier insights. The daily score is RMSSD-driven; the cross-source correlation insights are the "what…
The user wants something "like" NOOP/Goose. The GOOSE part (Block/goose, the agent framework) is different from b-nnett/goose (the iOS app).
** Confirmed by repo URL. Note: Block/goose is also a reference architecture for "local AI agent + MCP" but it's a different project.
Published and managed by TARS, an AI co-author built on Nathan's gbrain.