refactor(measurement): decompose tooling by vocabulary

.importlinter

@@ -1,5 +1,7 @@
 [importlinter]
-root_package = anonymizer
+root_packages =
+    anonymizer
+    measurement_tools
 
 # The NDD adapter layer is a leaf boundary: it must never import the engine
 # sub-workflows that call it.
@@ -45,3 +47,16 @@ source_modules =
     anonymizer.interface.run_telemetry
 forbidden_modules =
     anonymizer.interface.anonymizer
+
+[importlinter:contract:wandb-leaves-do-not-import-wandb-models]
+name = W&B leaves do not import the compatibility facade
+type = forbidden
+source_modules =
+    measurement_tools.wandb_policy
+    measurement_tools.wandb_settings
+    measurement_tools.wandb_metadata
+    measurement_tools.wandb_metrics
+    measurement_tools.wandb_publication
+    measurement_tools.wandb_field_policies
+forbidden_modules =
+    measurement_tools.wandb_models

tests/conftest.py

@@ -42,7 +42,10 @@ def loader(
         assert spec.loader is not None
         module = importlib.util.module_from_spec(spec)
         sys.modules[module_name] = module
-        for import_path in (*additional_paths, path.parent):
+        import_paths = list(additional_paths)
+        if not any(path.is_relative_to(import_path) for import_path in additional_paths):
+            import_paths.append(path.parent)
+        for import_path in import_paths:
             sys.path.insert(0, str(import_path))
         spec.loader.exec_module(module)
         return module

tests/tools/test_measurement_strict_import_publisher.py

@@ -26,6 +26,11 @@
 REPO_ROOT = Path(__file__).resolve().parents[2]
 
 
+def _wandb_publisher_module() -> ModuleType:
+    publisher = importlib.import_module("measurement_tools.wandb_publisher")
+    return sys.modules[publisher.WandbPublisher.__module__]
+
+
 def _write_simple_suite(tmp_path: Path, *, suite_id: str = "base-suite") -> Path:
     return _write_yaml(tmp_path / "suite.yaml", _simple_suite_payload(suite_id=suite_id))
 
@@ -649,7 +654,7 @@ def test_wandb_publisher_uses_explicit_handle_and_finishes_once(
     case = _write_case_measurements(run_benchmarks_tool, tmp_path, _minimal_benchmark_case(run_benchmarks_tool))
     state = _wandb_state()
     fake_wandb = _fake_wandb_module(state)
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     result = setup_module.WandbPublisher().publish(
@@ -756,7 +761,7 @@ def test_wandb_table_opt_in_changes_only_table_payloads(
     run_benchmarks_tool: ModuleType,
 ) -> None:
     case = _write_case_measurements(run_benchmarks_tool, tmp_path, _minimal_benchmark_case(run_benchmarks_tool))
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     states: list[SimpleNamespace] = []
 
     for log_tables in (False, True):
@@ -811,7 +816,7 @@ def failing_init(**kwargs: Any) -> Any:
         return run
 
     fake_wandb.init = failing_init
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     with pytest.raises(RuntimeError, match=f"{failure_stage} failed"):
@@ -844,7 +849,7 @@ def init_with_two_failures(**kwargs: Any) -> Any:
         return run
 
     fake_wandb.init = init_with_two_failures
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     with pytest.raises(ExceptionGroup) as raised:
@@ -876,7 +881,7 @@ def init_with_bad_finish(**kwargs: Any) -> Any:
         return run
 
     fake_wandb.init = init_with_bad_finish
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     with pytest.raises(RuntimeError, match="finish failed"):
@@ -943,7 +948,7 @@ def init_with_changed_id(**kwargs: Any) -> Any:
         return run
 
     fake_wandb.init = init_with_changed_id
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     with pytest.raises(RuntimeError, match="different run identity"):
@@ -967,7 +972,7 @@ def test_wandb_publisher_rejects_ambient_active_run(
     case = _write_case_measurements(run_benchmarks_tool, tmp_path, _minimal_benchmark_case(run_benchmarks_tool))
     state = _wandb_state()
     fake_wandb = _fake_wandb_module(state, active_run=True)
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     with pytest.raises(RuntimeError, match="ambient W&B run"):
@@ -1000,7 +1005,7 @@ def init_with_interrupt(**kwargs: Any) -> Any:
         return run
 
     fake_wandb.init = init_with_interrupt
-    setup_module = sys.modules[run_benchmarks_tool.publish_benchmark_wandb_best_effort.__module__]
+    setup_module = _wandb_publisher_module()
     monkeypatch.setattr(setup_module, "require_wandb", lambda: fake_wandb)
 
     with pytest.raises(KeyboardInterrupt):

tests/tools/test_measurement_wandb_logging.py

@@ -3,10 +3,10 @@
 
 from __future__ import annotations
 
+import importlib
 import json
 import logging
 import os
-import sys
 from pathlib import Path
 from types import ModuleType, SimpleNamespace
 from typing import Any
@@ -608,7 +608,7 @@ def fail_publish(*_args: Any, **_kwargs: Any) -> Any:
 
 
 def test_wandb_outbound_models_structurally_exclude_sensitive_fields(wandb_setup_tool: ModuleType) -> None:
-    models = sys.modules[wandb_setup_tool.WandbPublishPayload.__module__]
+    models = importlib.import_module("measurement_tools.wandb_models")
 
     with pytest.raises(ValidationError, match="Extra inputs"):
         models.WandbHistoryPayload(metrics={}, text="Alice")

tests/tools/test_measurement_wandb_reports.py

@@ -3,6 +3,7 @@
 
 from __future__ import annotations
 
+import importlib
 import json
 import logging
 import os
@@ -514,6 +515,31 @@ def test_wandb_report_markdown_uses_sanitized_fields(wandb_report_tool: ModuleTy
     assert "sk-secret-token" not in markdown
 
 
+def test_wandb_report_save_uses_auth_preflight_fallback(monkeypatch: pytest.MonkeyPatch) -> None:
+    wandb_reports = importlib.import_module("measurement_tools.wandb_reports")
+
+    report = SimpleNamespace(save=lambda **_kwargs: (_ for _ in ()).throw(RuntimeError("relogin required")))
+    fallback_calls: list[tuple[Any, bool]] = []
+
+    def fallback(value: Any, *, draft: bool) -> str:
+        fallback_calls.append((value, draft))
+        return "saved"
+
+    monkeypatch.setattr(wandb_reports, "_save_report_without_project_preflight", fallback)
+
+    assert wandb_reports.save_report(report, draft=True) == "saved"
+    assert fallback_calls == [(report, True)]
+
+
+def test_wandb_report_save_preserves_unrelated_sdk_errors() -> None:
+    wandb_reports = importlib.import_module("measurement_tools.wandb_reports")
+
+    report = SimpleNamespace(save=lambda **_kwargs: (_ for _ in ()).throw(RuntimeError("permission denied")))
+
+    with pytest.raises(RuntimeError, match="permission denied"):
+        wandb_reports.save_report(report, draft=False)
+
+
 def test_wandb_report_views_parse_v1_and_v2_with_explicit_axes(wandb_report_tool: ModuleType) -> None:
     summary, v1_config = _wandb_report_fixture()
     path = wandb_report_tool.WandbRunPath(entity="entity", project="project", run_id="run-a")