pytyche.visual_confidence

Visual confidence payload for generator verification.

Provides a tested payload function that backs all panels of the visual confidence notebook. The notebook renders from the payload; unit tests assert on the payload directly without notebook rendering.

Public API

  • VisualConfidencePayload — frozen dataclass with required sections.

  • build_visual_confidence_payload(bundle, bootstrap_seed, n_bootstrap=200) -> VisualConfidencePayload.

Functions

build_visual_confidence_payload(bundle, ...)

Build a VisualConfidencePayload from a CalibrationBundle.

Classes

VisualConfidencePayload(invariants, ...)

Typed payload backing all panels of the visual confidence notebook.
class pytyche.visual_confidence.VisualConfidencePayload(invariants, truth_summary, data_summary, recovery)[source]

Bases: object

Typed payload backing all panels of the visual confidence notebook.

All four fields are required — there are no optional sections. Future analyzer/BCF panels extend this type by subclassing or by adding fields to a derived dataclass without breaking existing sections (open/closed principle for the payload contract).

invariants

Name → bool map of generator contract checks.

truth_summary

Population-level truth statistics.

data_summary

Per-variant empirical summaries (variant name → stats).

recovery

Empirical recovery comparison with planted truth and bootstrap SE on the empirical lift.

Parameters:

  • invariants (dict[str, bool])

  • truth_summary (dict[str, object])

  • data_summary (dict[str, dict[str, object]])

  • recovery (dict[str, object])

pytyche.visual_confidence.build_visual_confidence_payload(bundle, bootstrap_seed, n_bootstrap=200)[source]

Build a VisualConfidencePayload from a CalibrationBundle.

Computes all panel data from the bundle:

  • invariants: 5 generator contract checks (all bool).

  • truth_summary: planted effect, components, and per-visitor CATE stats.

  • data_summary: per-variant empirical summaries (n_visitors, rates, revenue).

  • recovery: planted effect, empirical lift, and bootstrap SE on lift.

Parameters:

  • bundle (CalibrationBundle) – CalibrationBundle from generate_v2_core().

  • bootstrap_seed (int) – Seed for the bootstrap RNG — controls SE reproducibility.

  • n_bootstrap (int) – Number of bootstrap resamples. Default 200.

Returns:

Frozen payload with all four required sections populated.

Return type:

VisualConfidencePayload