You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Review the completed Merit Probe Evidence Bundle and automated Further-Investment Gate evaluation, then record the final research-investment decision and rationale. The decision must choose replication, bounded diagnosis, or pause while preserving caveats and the distinction between one-dataset evidence and general GraphNAMLSS claims.
Acceptance criteria
A human reviewer inspects per-seed and aggregate metrics, provenance, failed runs, selected checkpoints, predictions, and interpretability artifacts.
The recorded decision is explicitly one of: proceed to the PEMS-BAY Replication Gate, run a bounded diagnostic agenda, or pause further investment.
The rationale identifies the NLL or CRPS signal, relevant guardrails, interpretability status, and material caveats.
A promising result unlocks only PEMS-BAY consideration, not the full Primary Experiment Matrix, Bayesian Graph Streams, structural uncertainty, or application tracks.
A fail or inconclusive result names bounded diagnostic follow-ups rather than uncontrolled architecture expansion.
The decision avoids claiming general GraphNAMLSS superiority from the METR-LA Merit Probe.
Parent
#35
What to build
Type: HITL.
Review the completed Merit Probe Evidence Bundle and automated Further-Investment Gate evaluation, then record the final research-investment decision and rationale. The decision must choose replication, bounded diagnosis, or pause while preserving caveats and the distinction between one-dataset evidence and general GraphNAMLSS claims.
Acceptance criteria
Blocked by