feat!: Introduce new summary section for data metrics

diffly/cli.py

@@ -12,18 +12,10 @@
 
 from ._compat import typer
 from ._utils import ABS_TOL_DEFAULT, ABS_TOL_TEMPORAL_DEFAULT, REL_TOL_DEFAULT
-from .metrics import Metric, MetricFn
-from .metrics.change import DEFAULT_CHANGE_METRICS
-from .metrics.data import DEFAULT_DATA_METRICS
+from .metrics import DEFAULT_METRICS
 
 app = typer.Typer()
 
-#: All metric presets selectable via ``--metric``, combining the change and data sets.
-AVAILABLE_METRICS: dict[str, MetricFn | Metric] = {
-    **DEFAULT_CHANGE_METRICS,
-    **DEFAULT_DATA_METRICS,
-}
-
 
 @app.command()
 def main(
@@ -147,8 +139,9 @@ def main(
         list[str],
         typer.Option(
             help=(
-                "Metric presets to display per column. Repeatable. "
-                f"Available: {', '.join(AVAILABLE_METRICS)}."
+                "Metric presets to display. Repeatable. Change metrics appear as "
+                "extra columns in the Columns table; data metrics appear in the Data "
+                f"Inspection section. Available: {', '.join(DEFAULT_METRICS)}."
             )
         ),
     ] = [],
@@ -162,11 +155,11 @@ def main(
         hidden_column = [*hidden_column, *hidden_columns]
 
     for name in metric:
-        if name not in AVAILABLE_METRICS:
+        if name not in DEFAULT_METRICS:
             raise typer.BadParameter(
-                f"Unknown metric: {name!r}. Available: {', '.join(AVAILABLE_METRICS)}."
+                f"Unknown metric: {name!r}. Available: {', '.join(DEFAULT_METRICS)}."
             )
-    metrics = {name: AVAILABLE_METRICS[name] for name in metric}
+    metrics = {name: DEFAULT_METRICS[name] for name in metric}
 
     comparison = compare_frames(
         pl.scan_parquet(left),

diffly/comparison.py

@@ -4,10 +4,11 @@
 from __future__ import annotations
 
 import datetime as dt
+import inspect
 import warnings
 from collections.abc import Iterable, Mapping, Sequence
 from functools import cached_property
-from typing import TYPE_CHECKING, Literal, Self, overload
+from typing import TYPE_CHECKING, Literal, Self, cast, overload
 
 import polars as pl
 from polars.schema import Schema as PolarsSchema
@@ -25,12 +26,16 @@
     lazy_len,
     make_and_validate_mapping,
 )
-from .metrics import Metric, MetricFn, _make_numeric_metric
+from .metrics._common import Metric, MetricFn
+from .metrics.change import ChangeMetric
+from .metrics.data import DataMetric
 
 if TYPE_CHECKING:  # pragma: no cover
     # NOTE: We cannot import at runtime as we're otherwise running into circular
     #  imports. We're importing again below where we need `Summary` for more than
     #  type annotations.
+    from .metrics.change import ChangeMetricFn
+    from .metrics.data import DataMetricFn
     from .summary import Summary
 
 
@@ -951,16 +956,14 @@ def summary(
             hidden_columns: Columns for which no values are printed, e.g. because they
                 contain sensitive information.
             metrics: Optional mapping from display label to a metric. A value may be a
-                callable ``(left_expr, right_expr) -> pl.Expr`` or a
-                :class:`~diffly.metrics.Metric`. Each callable receives two
-                :class:`polars.Expr` referring to the left and right values of a single
-                column across all joined rows, and must return a scalar aggregation
-                expression. Bare callables are only computed for numerical columns; wrap
-                one in a :class:`~diffly.metrics.Metric` with a column selector to target
-                other column types (e.g. ``Metric(fn, selector=cs.all())``).
-                See :doc:`/api/metrics` for the full list of presets and the
-                :data:`~diffly.metrics.MetricFn` type. When ``None`` (default), no metrics
-                are computed; presets are not applied automatically. Prefer short labels —
+                :class:`~diffly.metrics.change.ChangeMetric`, a
+                :class:`~diffly.metrics.data.DataMetric`, or a bare callable resolved
+                by its arity (two arguments → change metric on numerical columns, one
+                argument → data metric on all columns). To target other column types,
+                construct the metric explicitly with a column selector
+                (e.g. ``ChangeMetric(fn, selector=cs.numeric())``). See :doc:`/api/metrics`
+                for the full list of presets. When ``None`` (default), no metrics are
+                computed; presets are not applied automatically. Prefer short labels —
                 the summary has a fixed width and many or long labels degrade rendering.
 
         Returns:
@@ -978,10 +981,7 @@ def summary(
         from .summary import Summary
 
         resolved_metrics = (
-            {
-                label: v if isinstance(v, Metric) else _make_numeric_metric(v)
-                for label, v in metrics.items()
-            }
+            {label: _resolve_metric(v) for label, v in metrics.items()}
             if metrics is not None
             else None
         )
@@ -1239,3 +1239,30 @@ def _list_length_exprs(
             for e in _list_length_exprs(expr.struct[field.name], field.dtype)
         ]
     return []
+
+
+def _resolve_metric(v: MetricFn | Metric) -> Metric:
+    if isinstance(v, Metric):
+        return v
+    # Infer the metric family from the number of required positional parameters: a
+    # single-argument callable describes one side (data), two arguments describe a
+    # change. Ambiguous signatures (variadic or a different arity) are rejected so the
+    # user wraps them explicitly in `DataMetric`/`ChangeMetric`.
+    params = inspect.signature(v).parameters.values()
+    required_positional = [
+        p
+        for p in params
+        if p.kind in (p.POSITIONAL_ONLY, p.POSITIONAL_OR_KEYWORD)
+        and p.default is p.empty
+    ]
+    has_variadic = any(p.kind is p.VAR_POSITIONAL for p in params)
+    if has_variadic or len(required_positional) not in (1, 2):
+        raise ValueError(
+            "Cannot infer the metric family from the callable's signature: expected "
+            "exactly one required positional argument (data metric) or two (change "
+            "metric), but got an ambiguous signature. Wrap it explicitly in "
+            "`DataMetric` or `ChangeMetric`."
+        )
+    if len(required_positional) == 2:
+        return ChangeMetric(fn=cast("ChangeMetricFn", v))
+    return DataMetric(fn=cast("DataMetricFn", v))

diffly/metrics/__init__.py

@@ -1,50 +1,18 @@
 # Copyright (c) QuantCo 2025-2026
 # SPDX-License-Identifier: BSD-3-Clause
-
 """Metrics computed per column when generating a summary.
 
 Two families are provided:
 
-- Metrics in :mod:`~diffly.metrics.change` describe the change between numeric
-  columns itself by aggregating over ``right - left``.
-- Metrics in :mod:`~diffly.metrics.data` describe the left and right datasets
-  individually, explaining how a change affects the data.
+- :class:`~diffly.metrics.change.ChangeMetric`s in :mod:`~diffly.metrics.change` describe the change between
+  numeric columns itself by aggregating over ``right - left``.
+- :class:`~diffly.metrics.data.DataMetric`s in :mod:`~diffly.metrics.data` describe the left and right
+  datasets individually, explaining how a change affects the data.
 """
 
 from __future__ import annotations
 
 from . import change, data
-from ._common import Metric, MetricFn
-from .change import (
-    _make_numeric_metric,
-    max,
-    mean,
-    mean_absolute_deviation,
-    mean_relative_deviation,
-    median,
-    min,
-    quantile,
-    std,
-)
-
-DEFAULT_METRICS: dict[str, MetricFn | Metric] = {
-    **change.DEFAULT_CHANGE_METRICS,
-}
-"""The default preset metrics, consisting of the change default set."""
+from ._common import DEFAULT_METRICS
 
-__all__ = [
-    "DEFAULT_METRICS",
-    "Metric",
-    "MetricFn",
-    "change",
-    "data",
-    "max",
-    "mean",
-    "mean_absolute_deviation",
-    "mean_relative_deviation",
-    "median",
-    "min",
-    "quantile",
-    "std",
-    "_make_numeric_metric",
-]
+__all__ = ["DEFAULT_METRICS", "change", "data"]

diffly/metrics/_common.py

@@ -1,27 +1,17 @@
 # Copyright (c) QuantCo 2025-2026
 # SPDX-License-Identifier: BSD-3-Clause
 
-from __future__ import annotations
+from .change import DEFAULT_CHANGE_METRICS, ChangeMetric, ChangeMetricFn
+from .data import DEFAULT_DATA_METRICS, DataMetric, DataMetricFn
 
-from collections.abc import Callable
-from dataclasses import dataclass
+Metric = ChangeMetric | DataMetric
+"""A change or data metric paired with a column-applicability selector."""
 
-import polars as pl
-import polars.selectors as cs
+MetricFn = ChangeMetricFn | DataMetricFn
+"""A bare change or data metric callable, resolved to a :data:`Metric` by arity."""
 
-
-@dataclass(frozen=True)
-class Metric:
-    """A metric function paired with a column-applicability selector."""
-
-    fn: MetricFn
-    selector: cs.Selector
-
-
-MetricFn = Callable[[pl.Expr, pl.Expr], pl.Expr]
-"""A metric function maps ``(left_expr, right_expr)`` to a scalar aggregation
-expression.
-
-The expressions refer to the left-side and right-side values of a single column across
-all joined rows.
-"""
+DEFAULT_METRICS: dict[str, Metric] = {
+    **DEFAULT_CHANGE_METRICS,
+    **DEFAULT_DATA_METRICS,
+}
+"""All preset metrics, combining the change and data default sets."""

diffly/metrics/change.py

@@ -1,21 +1,37 @@
 # Copyright (c) QuantCo 2025-2026
 # SPDX-License-Identifier: BSD-3-Clause
 
-"""Metrics describing the change between numeric columns.
-
-These aggregate over ``right - left`` to characterize the change itself.
-"""
-
 from __future__ import annotations
 
+from collections.abc import Callable
+from dataclasses import dataclass, field
+
 import polars as pl
 import polars.selectors as cs
 
-from ._common import Metric, MetricFn
+ChangeMetricFn = Callable[[pl.Expr, pl.Expr], pl.Expr]
+"""A `ChangeMetricFn` maps a pair of column expressions to a scalar aggregation
+expression."""
+
+
+@dataclass(frozen=True)
+class ChangeMetric:
+    """A metric quantifying the *change* in a column between the two sides of a
+    comparison.
+
+    Change metrics are rendered as extra columns in the "Columns" table, alongside the
+    match rate.
+    """
+
+    fn: ChangeMetricFn
+    """Aggregates over ``right - left`` (e.g. the mean delta) to describe the change
+    itself."""
+
+    selector: cs.Selector = field(default_factory=cs.numeric)
+    """Selects the columns the metric applies to; defaults to numeric columns."""
 
 
-def _make_numeric_metric(fn: MetricFn) -> Metric:
-    return Metric(fn=fn, selector=cs.numeric())
+# ---------------------------------- CHANGE METRICS ---------------------------------- #
 
 
 def mean(left: pl.Expr, right: pl.Expr) -> pl.Expr:
@@ -54,7 +70,7 @@ def mean_relative_deviation(left: pl.Expr, right: pl.Expr) -> pl.Expr:
     return ((right - left) / left).abs().mean()
 
 
-def quantile(q: float) -> MetricFn:
+def quantile(q: float) -> ChangeMetricFn:
     """Factory returning a metric that computes the ``q``-quantile of
     ``right - left``."""
     if not 0 <= q <= 1:
@@ -66,13 +82,13 @@ def _quantile(left: pl.Expr, right: pl.Expr) -> pl.Expr:
     return _quantile
 
 
-DEFAULT_CHANGE_METRICS: dict[str, MetricFn] = {
-    "Mean": mean,
-    "Median": median,
-    "Min": min,
-    "Max": max,
-    "Std": std,
-    "Mean absolute deviation": mean_absolute_deviation,
-    "Mean relative deviation": mean_relative_deviation,
+DEFAULT_CHANGE_METRICS: dict[str, ChangeMetric] = {
+    "Mean": ChangeMetric(fn=mean),
+    "Median": ChangeMetric(fn=median),
+    "Min": ChangeMetric(fn=min),
+    "Max": ChangeMetric(fn=max),
+    "Std": ChangeMetric(fn=std),
+    "Mean absolute deviation": ChangeMetric(fn=mean_absolute_deviation),
+    "Mean relative deviation": ChangeMetric(fn=mean_relative_deviation),
 }
 """Preset metrics describing the change between numeric columns."""