[MaxGhenis]

This PR fixes test_spm_system.py, which fails with PolicyEngine/policyengine-core#552, and tightens a provenance test in test_social_security_real_core.py, which passed on both cores before. Both files pass with core #552 and with core's current master. The PR changes no model code.

• test_spm_system.py. After subsampling, the holder test calculates is_household_head and requires a stored period. The reform gives that input a yearly definition without a formula. Core Add guaranteed income / cash assistance variable #552 returns its root default without caching when carry-over finds no source input. The test now supplies a yearly input after each subsample, and keeps its assertions for formula caching, yearly storage, reform policy bindings and baseline isolation.
• test_social_security_real_core.py. The dataset rebuild provenance test stored its 2023 derived entry with put_in_cache. Before build_from_dataset runs, the adapter records _derived_cached_returns, which lists the derived and fast-only returns in the simulation's fast cache and skips stored inputs. With the old fixture, the 2023 entry never reached that list on either core. The test now supplies a 2022 input at construction. The 2023 read carries it over, and the test asserts that this return reaches _derived_cached_returns.

Tests

At 862156d871, whose parent 0cc7e65cd3 merges main at 67d4dde6cb:

• test_social_security_real_core.py and test_social_security_known_periods.py: 134 passed with core Add guaranteed income / cash assistance variable #552 merged into core's master (c5eef1efd9). With core master (9fd76e1ce3), 129 passed and 5 skipped; the 5 skipped cases need core Add guaranteed income / cash assistance variable #552's branch argument.
• The new _derived_cached_returns assertion fails on both cores with the old put_in_cache fixture, and passes on both with the 2022 input.
• Two mutants of policyengine_us/tools/social_security_inputs.py, under core Add guaranteed income / cash assistance variable #552: dropping the branch argument from the physical get_known_periods call, or from the _public_periods call. Each fails all four cases of test_known_periods_lists_only_sources_readable_by_requested_branch.
• test_spm_system.py: 77 passed with core Add guaranteed income / cash assistance variable #552 and 77 passed with core master.
• Core Add guaranteed income / cash assistance variable #552's own CI doesn't run this file: it smoke-tests a pinned PE-US release. So before core Add guaranteed income / cash assistance variable #552 merges, a separate run of 38 PE-US core test files, this one included, checks core Add guaranteed income / cash assistance variable #552 against this head.
• CI: run 42401924872 passed all 35 jobs at this head.

impact: none (test only)
axiom: n/a: test-only change

🤖 Generated with Claude Code

[MaxGhenis]

US + core hub merge audit, us#10671 at 862156d8710b2870fa3c8c06c855e1e2a361deae (squash): Supply yearly input in subsampling holder test.

• What it changes: it fixes test_spm_system.py, which fails with Keep branch period reads, dumps, and disk deletion consistent policyengine-core#552, and tightens a provenance test in test_social_security_real_core.py. Both files pass with core Add guaranteed income / cash assistance variable #552 and with core's current master. It changes no model code.
• Review: An Opus 5.5 reviewer approved this head in round 1 with no P1 or P2 findings. Both changed files pass on core master (9fd76e1ce3) and on core Add guaranteed income / cash assistance variable #552 merged into master (c5eef1efd9), and each change now catches a mutant that the old test let through. The body now follows the review's P3 notes on its wording. One P3 stays: the changelog fragment describes only the SPM change. It describes that change accurately, and fixing it would rerun CI on a test-only PR.
• Gates checked live at merge:
  • gh pr checks exits 0;
  • the latest Pull request run at the head concluded success (42401924872, 35 of 35 jobs);
  • MERGEABLE;
  • not a draft;
  • no reviewer's latest review requests changes;
  • the semantic-overlap guard;
  • the merge pins the head.
• Impact: none; the PR changes only tests.
• Axiom: n/a; the PR changes only tests.

🤖 Generated with Claude Code