[anth-volk]

Fixes #964

Removes the None presets and None checks that #712 put around Stage 12 span and operation blocks so Pyright, reading policyengine-observability's scopes as possibly suppressing an exception, would not report their values as possibly unbound. Since #959, every package uses 3.0.4, whose scopes are annotated as returning Literal[False] from __exit__ and __aexit__ and return False on every path, so each block either assigns its value or its exception reaches the caller.

What is removed, and why each check could not fail

Module and function	Removed	Value assigned in the block
src/modal/v2_app.py run_single_simulation_segment_us	result = None, "Stage 12 segment calculation returned no result"	calculate_segment(...).model_dump(mode="python"), a dict
src/modal/v2_app.py run_single_simulation_us	result = None, "Stage 12 US simulation returned no result"	run_single_simulation(...), typed dict[str, Any]; every path returns a dict or raises
src/modal/v2_app.py run_single_simulation_uk	result = None, "Stage 12 UK simulation returned no result"	the same function
src/modal/v2_app.py coordinate_report	result = None, "Stage 12 report coordinator returned no result"	stage12_runtime.coordinate_report(...), typed dict[str, Any]; every path returns a dict or raises
stage12_runtime/fanout.py Stage12SegmentRunner.run, dispatch span	call = None and _require_call, "Stage 12 segment dispatch returned no call"	the FunctionCall that modal 1.4.2's Function.spawn creates and returns on every path (it raises for a generator function, and the segment function is not one)
stage12_runtime/fanout.py Stage12SegmentRunner.run, wait span	payloads: list[object] | None = None, "Stage 12 segment wait returned no payloads"	_collect(...), typed list[object]; returns a list or raises
stage12_runtime/fanout.py Stage12SegmentRunner.run, merge span	merged: SimulationCalculation | None = None, "Stage 12 segment merge returned no calculation"	merge_segment_results(...), which returns a SimulationCalculation

The four Modal functions and the merge now return from inside their block, because ruff's RET504 rejects assigning a value only to return it. The scope still ends, and each Modal function's runtime still shuts down, before the function returns.

A search of the executor, gateway and entrypoint src directories for any value set to None before a span, operation or other with block and used after it found one other: setup_gcp_credentials sets creds_file = None before an ExitStack block that leaves it unassigned when credentials are already configured or absent. That check is reachable and stays.

Tests

Removing unreachable code changes no behaviour, so there is no failing test to write first. The first commit adds tests that run every block involved, and they pass before and after the removal:

• the four Modal functions with the observability runtime the services build (OpenTelemetry disabled), returning their result and passing an application error to the caller unchanged, never as a "returned no result" RuntimeError;
• the segment runner with that runtime: a successful run, and an error in each of the dispatch, wait and merge blocks; and a merge failure with the existing recording runtime, the one block that had no failure test.

Checks

• uv run --frozen pytest tests/ in projects/policyengine-simulation-executor: 1889 passed, 22 skipped, 3 deselected at both commits.
• python -m scripts.rules pyright --sync projects/policyengine-simulation-executor: no errors at both commits.
• ruff 0.15.22: ruff format --check projects/*/src libs/*/src and ruff check . pass.
• python scripts/run_quality_guards.py, python -m scripts.rules guard --base-ref origin/main and python -m unittest discover -s scripts/rules/tests -t . pass; python -m scripts.rules render leaves no diff.

🤖 Generated with Claude Code