System what it's doing -- health / learning / proof

reads the Auto-API -- paper only -- no $ claimed

autonomous stack -- live status

checking

checking...

polling supervisor feed...

self-improve: READY-but-INERT (human-gated)real-money: DENYin-game CLV: INSUFFICIENT_DATA (accruing)calibration not edge - vs_close UNPROVEN - units only, no $

unavailablesupervisor feed unavailable

how to read this page

calibration
edge vs calibration
vs_close
Brier score
CLV
INERT
units

Parity grid (live status below)

The cross-sport coverage / loop-health grid is tracked below in real time. A present-but-broken cell would be red and fail the gate. See the ParityGrid panel for current status.

Depth plateau (honest)

Deeper data rejects for the close across all sports and tiers. Adding more historical depth does not beat the devigged close -- the honest finding, recorded as a success.

Confirmed corners survivor -- CALIBRATION, not a market edge

One signal (soccer corners) survived the leak-free gate. It is a CALIBRATION improvement, not a profit edge. No $ edge is claimed anywhere.

In-game prior-conditioning is the measured edge

Conditioning the in-game forecast on the pre-game prior is the proven, MEASURED calibration improvement (static -> conditional Brier). Calibration, not currency.

System health + freshness

checkingPOLL

Autonomy status

Producer / scheduler freshness

checking

Last-update age is derived from each sport's own generated_at (stale never reads green). A sport missing from the producer rotation shows as "not scheduled" rather than being hidden. Units only; no $.

self-improve ratchetbuilt and READY, but INERT -- the pipeline flag is human-gated OFF, so it measures only and ships nothing.

Self-improve loop

POLL

Self-improve ratchet

loading…

Parity / health grid

loading…

CLV over time (measurement-only, probability)

in-game calibration gateper-sport, both-direction cross-corpus verdicts. The meta-finding: generic in-game micro-state REJECTS across NBA / MLB / soccer / tennis.