Changelog¶

All notable changes to mrv-lib are documented here. The project follows Keep a Changelog and Semantic Versioning.

(no changes yet)

Patch release: report packaging, resolution-path wiring, lint, and docs.

Report template now ships in the wheel. The LaTeX template moved from the repo-root templates/ directory into the package at src/mrv/templates/template.tex and is declared in [tool.setuptools.package-data]. report() / generate_report() resolve it via importlib.resources, so it works for pip-installed users (previously the template was not packaged and report generation failed).
report() now renders resolution (Paper 2) result JSON. _render / _expand_assets branch on data["test"]; the resolution case is rendered from overall_mean_ari with the Spearman / factor-set sections omitted. Previously a resolution result.json raised KeyError.
RepValidator._build_json: an exact overall ARI / Spearman of 0.0 now serialises as 0.0 instead of null (truthiness check replaced with an explicit is not None).
RepInvarianceResult.summary() now reports the mean of the finite per-pair Spearman values (falling back to n/a only when none are finite) instead of a hardcoded n/a.
mrv.data.download_yahoo.download skips a symbol/frequency when a required OHLC column is missing (instead of raising KeyError) and includes Volume in the saved CSV when present.

Retired the SR 26-2 / SR 11-7 governance-report symbols. sr26_2_report, generate_sr26_2_report, and the deprecated sr11_7_report / generate_sr11_7_report aliases have been removed; they referenced a template that no longer ships. Use the generic report() / generate_report().
Removed the res subcommand from the CLI. Resolution invariance is labels-first: fit your own regime models at each frequency and call validate_res() from the Python API. The CLI no longer exposes an unimplemented convenience path.

Minor release consolidating the representation-invariance (Paper 1) and resolution-invariance (Paper 2) validators behind a stable public API.

mrv.invariance.rep_invariance_validator(model_fn, admissible_class, returns, K): functional Paper 1 representation-invariance check returning a typed RepInvarianceResult (per-pair ARI, per-pair Spearman ordering, 1/K null, partition and ordering pass flags).
mrv.invariance.res_invariance_validator(model_fn, resolution_set, ...): functional Paper 2 resolution-invariance check returning a typed ResInvarianceResult (cross-frequency ARI matrix, AMI matrix, within-intraday excess, permutation p-value and null CI).
ResolutionSpec dataclass with the Paper 2 four-frequency panel (5m, 15m, 1h, 1d) and intraday subset (5m, 15m, 1h); constants PAPER2_FREQS and PAPER2_INTRADAY_FREQS.
PUBLIC_API.md: semantic-versioning contract and full public-symbol listing.
__all__ declarations across the public modules; Google-style docstrings and type hints on the public surface.
Example notebook section demonstrating rep_invariance_validator alongside the labels-first RepValidator API.

src/mrv/__init__.py: __version__ set to "0.6.0".
tests/test_version.py: version assertion updated to "0.6.0".
CITATION.cff: version set to "0.6.0"; Paper 1 and Paper 2 statuses updated.

Patch release. Backward-compatible additions and stricter input validation.

Rolling-window index off-by-one corrected; window-boundary values shift by one observation step. Callers replaying historical series should re-run.

SR 11-7 references migrated to SR 26-2 / OCC Bulletin 2026-13 throughout the codebase, templates, examples, and documentation. SR 26-2 (Federal Reserve letter) and OCC Bulletin 2026-13, jointly issued 2026-04-17 by the OCC, Federal Reserve, and FDIC, supersede SR 11-7 (Federal Reserve, 2011) as the primary US model risk management guidance. See OCC Bulletin 2026-13.
sr11_7_report (in mrv.pipeline) aliased to the new sr26_2_report with a DeprecationWarning when called.
generate_sr11_7_report (in mrv.validator.report) aliased to the new generate_sr26_2_report with a DeprecationWarning when called.
Version bumped to 0.5.0.

mrv.pipeline.sr26_2_report: canonical replacement for sr11_7_report.
mrv.validator.report.generate_sr26_2_report: canonical replacement for generate_sr11_7_report.

Relicensed from MIT to AGPL-3.0-or-later with a commercial exception. The open-source license is now GNU AGPL-3.0 (full text in LICENSE): free for academic, teaching, and personal use, with AGPL copyleft and network-source obligations. Organizations needing proprietary or closed-source use that cannot meet the AGPL must obtain a separate commercial license; see COMMERCIAL-LICENSE.md. pyproject.toml, CITATION.cff, and README.md updated accordingly.

Resolution Invariance validator (res): multi-frequency regime analysis (5m/15m/1h/1d), cross-frequency ARI/AMI/VI matrices, permutation p-values, event/calm window analysis, time-of-day seasonality, calendar-window robustness, robustness sweeps (K, window scale), and timeline / rolling-ARI visualizations, with an HMM dual-model comparison.
Business impact function (impact_fn): user-defined callback (labels, prices) -> float on both RepValidator and ResValidator; computes a pairwise impact-delta matrix across representations/frequencies.
Disagreement attribution: leave-one-out factor attribution (rep), frequency-pair decomposition (res), and temporal hotspot detection with a per-day ARI timeline.
Continuous monitoring: monitor() with init/incremental modes, persistent monitoring_history.csv, configurable alert thresholds, file-based alerts, and webhook support.
Model validation report: LaTeX template with auto-generated findings and severity classification (Critical/High/Medium/Low/Informational), user-fillable YAML overrides, executive summary with risk rating, and an effective-challenge evidence section.
Findings engine: findings.py with a Finding dataclass, classify_severity(), generate_findings(), overall_risk_rating(), and YAML override merging.

Data infrastructure: IB download for all frequencies (5m/15m/1h/1d), incremental updates, OHLCV validation and resampling.
Factor engine: seven built-in risk factors (volatility, drawdown, max drawdown, VaR, CVaR, realized skew, stability) with a user-extensible registry.
Model registry: GMM and HMM built in, user-extensible via register_model().
Normalization: rolling z-score and min-max.
Representation Invariance validator (rep): pairwise ARI across multiple factor sets, per-asset heatmaps, JSON output.
Pipeline: standard load -> factors -> fit -> validate -> report flow, each step replaceable.
Two-layer verdict: partition stability (ARI >= 0.65) and ordering stability (Spearman >= 0.85).
Report generation: JSON -> LaTeX -> PDF pipeline with template conditionals, dashboard, heatmaps, and remediation plan.
CLI: run, download, report subcommands.
Configuration: single config.yaml with download, logging, factors, normalization, and validator sections.
Logging: console plus optional timestamped file logging.
Security: IB host/port validation, non-positive price detection.