Config matrix

The exhaustive, always-current reference for GoldenAnalysis configuration. GoldenAnalysis is read-only metrics + reporting, so its config surface is small: analyze(...) keyword arguments, the regression RegressionPolicy, and a couple of vocabularies. This page consolidates them into one gated reference.

Everything below the line is generated from code (scripts/gen_config_matrix.py) and verified in CI. The guidance above the line is hand-authored. See Overview and Cross-run.

How GoldenAnalysis is configured

Run knobs are analyze(...) keyword arguments (see Runtime options): analyzers (which analyzers to run; None = all frame-compatible), dataset, run_id, generated_at.
Regression gating uses a RegressionPolicy (default_pct plus optional per_metric overrides). A metric’s Direction (higher_better / lower_better / neutral) decides which way a change counts as a regression, compared against a chosen Baseline (previous / rolling_median / last_known_good).
Analyzers are an open registry (frame.summary, match.rates, cluster.distribution, quality.rollup, plus any third-party entry points), so the analyzer set is extensible and not gated here.

Config object reference

Every config object in the pydantic tree(s), generated from the package schema. Nested objects link by name.

`RegressionPolicy`

Per-metric regression thresholds (percent). Falls back to default_pct.

Field	Type	Default	Choices / notes
`default_pct`	float	`10.0`	Percent change from the baseline that counts as a regression for any metric without a per-metric override.
`per_metric`	dict[str, float]	`{}`	Per-metric percent thresholds keyed by metric name, overriding default_pct for those metrics.

Runtime options

Config passed as call keyword arguments (this package has no single config model). Generated from the signature.

`analyze(...)`

Argument	Type	Default	Choices / notes
`df`	pl.DataFrame	required
`analyzers`	Sequence[str] \	None	`None`
`dataset`	str \	None	`None`
`run_id`	str \	None	`None`
`generated_at`	datetime \	None	`None`

CLI

Every command and its options/arguments, generated from the Typer app. choice-typed options list their allowed values.

Command	Option	Type	Default	Notes
`mcp-serve`	`--transport`	text	`'stdio'`	stdio \| http
`mcp-serve`	`--host`	text	`'0.0.0.0'`	Host interface to bind when transport is http.
`mcp-serve`	`--port`	integer	`8300`	HTTP port (A2A convention: Analysis = 8300).
`regressions`	`--dataset`	text	required	Dataset name whose run history to check.
`regressions`	`--history`	path	required	ReportHistory path (.jsonl or .db).
`regressions`	`--baseline`	text	`'rolling_median'`	Baseline strategy to compare against, e.g. rolling_median.
`regressions`	`--window`	integer	`7`	Number of prior runs the baseline is computed over.
`regressions`	`--policy`	text	`None`	Per-metric gates: JSON or “key=pct,key=pct”.
`regressions`	`--fail-on-regression`	boolean	`False`	Exit 1 if any regression is flagged (CI gate).
`report`	`input`	path	required	A .parquet/.csv frame, or a .json AnalysisReport to re-render.
`report`	`--analyzers`	text	`'all'`	Comma-separated analyzer names, or ‘all’.
`report`	`--format`	text	`'markdown'`	markdown \| json
`report`	`--out`	path	`None`	Write the report here instead of (also) printing.
`trend`	`--metric`	text	required	Metric key to trend, e.g. cluster.singleton_ratio.
`trend`	`--dataset`	text	required	Dataset name whose run history to read.
`trend`	`--history`	path	required	ReportHistory path (.jsonl or .db).
`trend`	`--last`	integer	`30`	Number of most recent runs to include in the trend.

Enumerated vocabularies

Allowed values for the str-typed / registry-backed fields above.

Metric direction

Direction — Metric.direction.

Value	Meaning
`higher_better`	A higher value is better; flag a regression on a decrease.
`lower_better`	A lower value is better; flag a regression on an increase.
`neutral`	No preferred direction; never flags a regression.

Regression baseline

Baseline — --baseline.

Value	Meaning
`last_known_good`	Compare against the last run marked good.
`previous`	Compare against the immediately prior run.
`rolling_median`	Compare against the rolling median of recent runs.

Analyzers

_FALLBACK — analyze(analyzers=...).

Value	Meaning
`cluster.distribution`	GoldenMatch clusters: count, singletons, size percentiles, reduction ratio.
`frame.summary`	Generic frame stats: row/column counts, null ratio, duplicate ratio, memory.
`match.rates`	GoldenMatch results: pair count, match rate, threshold, recall estimate, score histogram.
`quality.rollup`	GoldenCheck / GoldenFlow rollup: findings, health score, rows changed, rules fired.

Environment variables (index)

1 GOLDENANALYSIS_* runtime knob(s) read by the package, scanned from source so this is complete. Grouped by area:

NATIVE (1): GOLDENANALYSIS_NATIVE

Get started

Concepts

GoldenMatch

GoldenCheck

GoldenFlow

GoldenPipe

GoldenAnalysis

InferMap

SQL extensions

Reference

Research

How GoldenAnalysis is configured

Config object reference

`RegressionPolicy`

Runtime options

`analyze(...)`

CLI

Enumerated vocabularies

Metric direction

Regression baseline

Analyzers

Environment variables (index)

See also

​How GoldenAnalysis is configured

​Config object reference

​RegressionPolicy

​Runtime options

​analyze(...)

​CLI

​Enumerated vocabularies

​Metric direction

​Regression baseline

​Analyzers

​Environment variables (index)

​See also

How GoldenAnalysis is configured

Config object reference

`RegressionPolicy`

Runtime options

`analyze(...)`

CLI

Enumerated vocabularies

Metric direction

Regression baseline

Analyzers

Environment variables (index)

See also