feat: explicit Brier score variants for 2×2 evaluation matrix

[Polichinel]

Summary

• Replaces Brier_sample/Brier_point with three task-explicit variants: Brier_cls_point, Brier_cls_sample, Brier_rgs_sample
• Fixes broken Brier_cls_sample for MC Dropout probability samples — uses mean(y_pred) instead of mean(y_pred > threshold) which destroyed discrimination
• Sets classification Brier profile default to threshold=0.0 (Pre-Release Note 05: event y > 0)
• Adds Brier_rgs_sample to regression sample membership for count-data binarisation use case
• Updates CIC MetricCatalog.md with new naming convention, test counts, and breaking rename notice

Test plan

• 249 tests passing (net +4 new Brier golden-value and edge-case tests)
• Registry snapshot integrity updated (24 → 25 metrics)
• NativeEvaluator integration tests updated for new metric names
• CIC documentation aligned with code changes

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR refactors Brier score handling in views_evaluation to make the metric names explicit across the (task × prediction-type) matrix, and corrects classification-sample Brier computation for probability-sample inputs (e.g., MC Dropout).

Changes:

• Replaces Brier_sample / Brier_point with Brier_cls_point, Brier_cls_sample, and Brier_rgs_sample across the metric catalog, memberships, profiles, and result dataclasses.
• Fixes Brier_cls_sample to compute event probability as mean(y_pred) (probability samples) rather than mean(y_pred > threshold).
• Updates tests and CIC documentation to reflect the renamed/added metrics and updated registry size.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
views_evaluation/profiles/hydranet_ucdp.py	Updates profile docstring to reflect inheritance behavior (incl. Brier params).
views_evaluation/profiles/base.py	Updates base-profile genome defaults for the new Brier metric variants.
views_evaluation/evaluation/native_metric_calculators.py	Introduces the three explicit native Brier calculators and fixes cls-sample computation.
views_evaluation/evaluation/metrics.py	Renames/adds dataclass fields to match new metric names.
views_evaluation/evaluation/metric_catalog.py	Registers new metric names/functions and updates task/pred-type memberships.
tests/test_native_evaluator.py	Updates evaluator integration tests for renamed Brier metrics and probability-sample semantics.
tests/test_metric_catalog.py	Updates registry snapshot size assertion (24 → 25).
tests/test_metric_calculators.py	Updates imports and expands/renames Brier golden/edge-case tests for new variants.
documentation/CICs/MetricCatalog.md	Updates contract doc with rename note and updated Brier/QS test counts.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

This contract doc explains the Brier metric rename, but other documentation in the repo still references the legacy names (e.g. documentation/standards/physical_architecture_standard.md lists Brier_point/Brier_sample). Please update those references (or add an explicit note/link here) so the documentation set stays internally consistent after the breaking rename.

Suggested change

- Explicit Brier variants added (2026-04-09): `Brier_sample`/`Brier_point` replaced by three task-explicit variants: `Brier_cls_point`, `Brier_cls_sample`, `Brier_rgs_sample`. The `_cls_`/`_rgs_` infix denotes the task context (classification vs. regression). `Brier_rgs_point` is intentionally omitted — a regression point estimate is not a probability. `Brier_cls_sample` averages probability samples (`mean(y_pred)`); `Brier_rgs_sample` binarises count samples (`mean(y_pred > threshold)`). Catalog size: 24 → 25.

- Explicit Brier variants added (2026-04-09): `Brier_sample`/`Brier_point` replaced by three task-explicit canonical variants: `Brier_cls_point`, `Brier_cls_sample`, `Brier_rgs_sample`. The `_cls_`/`_rgs_` infix denotes the task context (classification vs. regression). `Brier_rgs_point` is intentionally omitted — a regression point estimate is not a probability. `Brier_cls_sample` averages probability samples (`mean(y_pred)`); `Brier_rgs_sample` binarises count samples (`mean(y_pred > threshold)`). Legacy-name mapping: `Brier_point` => `Brier_cls_point`; `Brier_sample` => `Brier_cls_sample` for classification-sample contexts and `Brier_rgs_sample` for regression-sample contexts. Note: older repository documents may still mention `Brier_point`/`Brier_sample`; interpret those references using this mapping until all docs are updated. Catalog size: 24 → 25.