Result Records and Evidence Levels

Every table and figure should be traceable to data, configuration, code revision, seed, hardware, and an explicit evidence level. This prevents a compact mechanism check from being presented as a source-scale reproduction.

Required Record

from silva_networks import SILVAResultRecord

record = SILVAResultRecord(
    family="silva_fno_deq",
    evidence_status="compact-verified",
    dataset="analytic periodic field",
    dataset_version="v1",
    split="seeded four-sample fixture",
    configuration="field-comparison-v1",
    seed=122,
    metrics=(("mse", 0.145), ("residual", 0.02)),
    data_fingerprint="sha256:...",
    code_revision="commit-or-working-tree-revision",
    hardware="CPU; precision=float32",
    deviations=("compact 8 by 8 grid",),
)
assert record.validate() == ()

Promotion Rule

Evidence moves forward only when the next stage's acceptance checks and artifacts exist. A successful subset run remains subset-verified; it does not become source-scale-reproduced because its learning curve looks promising.

Figure and Table Captions

State the family, dataset and split, metric, number of seeds, scale tier, evidence level, and whether uncertainty is across seeds, samples, or batches. Link the machine-readable result and configuration beside the caption whenever the publication surface permits it.

Where to Go Next

Question	Page
Where are the family-specific acceptance stages?	Family Dossiers
How are result records represented in Python?	Research-Depth API
Which comparisons emit measured records?	Cross-Family Comparisons