[MaxGhenis]

The optional graph extra pinned microcosm-frame and microcosm-graph to 3ff92b0a, the head of PolicyEngine/microcosm#873, which is not on microcosm main. Under ARCH-1 (Max, 2026-10-15), #873 retires once dynamics pins a main commit (microcosm#1285, the release contract). This PR repins both packages to microcosm main c28b4537.

Validation

By a GPT-6.1 Sol build lane, 20261015-061437-md-repin-microcosm:

• Install. Fresh venvs on Python 3.13.15 and 3.14.7, each with uv pip install -e '.[dev,graph]'. Both packages' direct_url.json confirms the exact commit, and require_graph() passes.
• Tests. The CI graph-integration tests, which cover every test importing microcosm.*, give 155 passed with no skips on each Python version.
• Tier counts. Unchanged, and the tier-policy test passes against full collection.
• Reproduction.
  • canonical.py is byte-identical between the two commits.
  • keyed_uniform draws are byte-identical between 3ff92b0a and main on 315 cases. Main changes only temporal NumPy coordinates and signed float zero; dynamics passes integer person/year/draw coordinates and string process names.
• Hashes. Implementation and node hashes do change: runtime.py:642 and trajectory.py:573 hash keyed_uniform's whole defining module via source_hash. No committed graph reproduction pin depends on them, and the frozen candidate spec hashes in tests/test_m6_engine_refit.py are unchanged. No runs/*.json, gates.yaml, Python or test file changes.

Docs

docs/population-graph.md now:

• names the main pin;
• drops the "core change must be reviewed first" caveat, since the code is on main;
• rewords the receipt note for main's executor;
• lists tests/test_graph_trajectory_accounting.py in the test command CI already runs.

Dynamics needs no #873-only feature, so microcosm#873 can close once this merges.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
social-security-model	Ready	Preview	Oct 15, 2026 12:24pm UTC

[MaxGhenis]

Merged at reviewed head f34888d (merge commit 2c0eb63) by the dynamics coordinator. Gates: gh pr checks exit 0 (graph-integration 3.13 and 3.14, all four pytest shards), MERGEABLE/CLEAN, not a draft, no change requests.

Independent review: Subfleet 20261015-082342-md541-review (hard tier, served by Opus 5.5 with a shell), APPROVE, with no blocker and no major finding. It checked:

• the installed microcosm-frame and microcosm-graph byte-match c28b4537;
• 155 passed with no skips on 3.13 and 3.14;
• keyed_uniform gives byte-identical draws on 169,164 dynamics-shaped keys;
• mortality, 5-year and accounted trajectories match end to end across the two pins on 3 seeds;
• no committed hash pins the moved node keys;
• the API surface used at 3ff92b0a is all still present on main.

One minor finding is left for a follow-up: the comment at src/populace_dynamics/graph/trajectory.py:140-141 still describes the old pin's gate behaviour. Behaviour is unaffected.

Master moved from 22e3f82 to e9ad042 after CI ran. None of those 63 commits touch pyproject.toml, docs/population-graph.md, src/populace_dynamics/graph, tests/test_graph_* or the workflows. This PR adds no tests, so tier counts are unaffected.

Next: closing microcosm#873, which microcosm main (#1285) supersedes.