Split persistence and report JSON serialization (+introduce Pydantic)

[Gerrrr]

This PR separates durable JSON serialization from report formatting (a follow-up to #161) and starts the Pydantic migration from #78.

The first commit makes to_json() mean “precise persistence JSON” for the change-point domain objects. Numeric values now stay numeric, qhat is included in serialized change points, and the old rounded=True/False switch is removed from the domain model. Rounded, human-facing JSON output now belongs to Report, alongside the existing log and regressions-only renderers.

The second commit adds a real JSON boundary test for analyzed-series persistence. It verifies that AnalyzedSeries.to_json() survives json.dumps() / json.loads() before being passed back into AnalyzedSeries.from_json().

The third commit converts AnalysisOptions to a Pydantic v2 model. Defaults now live on the model, unknown fields are rejected, assignment is validated, and AnalyzedSeries.from_json() restores options with AnalysisOptions.model_validate(...) instead of manual field copying. This follows the direction of the closed #93, but keeps the scope tighter and uses Pydantic v2 APIs.

[Copilot]

Pull request overview

This PR separates durable persistence JSON from human-facing report JSON, and begins migrating JSON boundaries to Pydantic v2 models (notably AnalysisOptions and AnalyzedSeries).

Changes:

• Make change-point domain to_json() output precise persistence JSON (numeric types preserved, qhat included) and remove the old rounded= switch.
• Move rounded/stringified JSON formatting into Report’s JSON renderer.
• Introduce Pydantic v2 models for serialization (AnalysisOptionsModel, AnalyzedSeriesModel) and add tests for JSON-boundary round-trips and validation behavior.

Reviewed changes

Copilot reviewed 8 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
uv.lock	Adds Pydantic and its transitive dependencies to the locked environment.
pyproject.toml	Declares pydantic>=2,<3 as a runtime dependency.
otava/serialization.py	Introduces Pydantic v2 models that define/validate persistence JSON shapes.
otava/series.py	Migrates AnalysisOptions to Pydantic, updates analyzed-series persistence JSON (typed metrics/data/options), and restores options via model validation.
otava/change_point_divisive/base.py	Removes rounded JSON paths, makes change-point JSON precise, and includes qhat in serialized change points.
otava/report.py	Implements report-specific rounded/stringified JSON rendering for human output.
tests/change_point_classes_test.py	Updates expectations to match precise persistence JSON and qhat inclusion.
tests/series_test.py	Adds JSON-boundary and Pydantic validation tests for options/analyzed-series persistence.
tests/report_test.py	Updates expected report JSON shape/format and refactors some test construction.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 9 changed files in this pull request and generated no new comments.

Suppressed comments (4)

otava/serialization.py:67

• Top-level series attributes are also not guaranteed to be strings: GraphiteImporter stores None for unmatched events, while database importers copy attribute column values without conversion (otava/importer.py:573-575 and 799-801). Pydantic therefore rejects previously JSON-serializable series at this new validation boundary. Model these lists with nullable/JSON-scalar values rather than str only.

    attributes: Dict[str, List[str]]

otava/serialization.py:62

• Postgres and BigQuery importers explicitly build List[float] timestamps with datetime.timestamp() (otava/importer.py:546-560 and 772-786). Timestamps with microseconds therefore contain a fractional part, which Pydantic cannot validate as int, so AnalyzedSeries.to_json() now raises instead of persisting those series. Preserve both integer-indexed and fractional Unix times.

    time: List[int]

otava/serialization.py:54

• Change-point group attributes can be None: GraphiteImporter inserts None when no event matches a timestamp (otava/importer.py:169-180), and those values flow into groups via Series.attributes_at(). This strict str type makes AnalyzedSeries.to_json() fail whenever such a timestamp has a change point. Allow nullable/JSON-scalar attribute values so valid importer output remains persistable.

This issue also appears on line 67 of the same file.

    attributes: Dict[str, str]

otava/series.py:450

• Neither the structured nor legacy metric branch actually restores unit: Metric.__init__ unconditionally assigns self.unit = "" at otava/series.py:55, ignoring the third argument passed here. Consequently from_json() silently loses any persisted unit and the next to_json() emits an empty value. Update the constructor to retain its unit argument and cover the metric round trip.

                    metric_json.get("unit", ""),
                )
            else:
                new_metrics[metric_name] = Metric(None, None, metric_json)