# Darcy flow rates reproduction results

NOTS measured R300/300 = 0.28903131 ± 0.016963418 (population std, ten trials).

Paper Table 1 reference: 0.012 ± 0.001. This is a literature value, not measured data.

All measured trajectories contain 300 queries including one random initial query. Training and acquisition use the same physical pressure scale. The paper main-text flow sign conflicts with Appendix D.1; the latter is used. Published BO/BFO/GP-TS/STO-NTS entries are reference-only, not reruns.

## Verified execution

Case job 975454: COMPLETED, ExitCode 0:0, elapsed 00:53:35, node a01r3n10; stderr empty. All 10 trials, 3000 oracle queries, checkpoints, strict checkpoint inference, and six PNG/PDF figures passed the completion audit. Independent CSV reaggregation agrees with metrics.json.

The measured NOTS result does **not** reproduce the paper numerical value. Repairs addressed local scripts-package shadowing and neuraloperator state-dictionary metadata serialization; the prior scientific protocol was preserved.

Test-set checkpoint inference: MSE 0.03060311265; mean relative L2 0.3399168253; flow-functional MAE 1.425788999. Best checkpoint: seed49, SHA256 2cf139f9cf75a01064fe54b5e8ba0d4e58a4b76a39aa7cf924320acbee4336b2.

See scheduler_status.json and completion_audit.json for actual execution evidence.