Module baseline 

Named baselines, rustev.baseline/1, and the reproducibility check (spec 029, 3.8 and 3.9).

A baseline is a measurement record of one repeat set: it binds the repeat report and every run document by digest, the dataset, split and role, the evaluator and repeat configurations, the plan, the source, the run seeds, the held-out record, and every cohort’s pooled, interval and band. It is not a quality claim and grants nothing. The reproducibility check compares a new repeat set’s stability metrics with a baseline of the same configuration, within the baseline’s bands.

Structs

Baseline

BaselineCohort

BaselineMetric

Exceedance

One stability metric beyond the baseline’s band.

Enums

BaselineError

Why no baseline was recorded (invalid_input).

Reproducibility

The reproducibility verdict (3.9).

Constants

BASELINE

MAX_NAME

The longest baseline name.

Functions

record

Record a baseline of report (3.8). Every binding is verified against the documents supplied: the repeat set is assembled again from inputs and must equal the report, apart from a held-out count, which only the host’s ledger can state and which is copied as recorded.

reproducibility

Compare a repeat set with a baseline of the same configuration: for every stability metric in every baseline cohort, reproducible when |pooled_run - pooled_baseline| <= band_baseline.

valid_name