Repository navigation
Conversation
…ing scenario_114's Virginia income tax policyengine-us 2.15.17 treats every Medicare-eligible person as enrolled (takes_up_medicare_if_eligible defaults to true) and adds the modeled Part B premium, $2,434.80 for 2026, to medical expenses. No prompt states Medicare enrollment or a premium. A sweep of all 1,984 outputs on the reference system with the premium at 0, and with enrollment off, moves exactly two: scenario_114 federal (10,729.61 -> 11,265.27, already proposed in #191) and Virginia (3,514.15 -> 3,654.15). No SNAP or eligibility output moves by any amount. Adds reference_audit/2026-10-05-medicare-part-b: the sweep, a per-household explanation, the drafted records (one for the Virginia output, and a fallback for the federal output if #191's record is not adopted) and the leaderboard impact under each combination with #191. Changes no reference or published number; the exclusion waits for Max's ruling. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
…ct summary and independent reviews - State the prompt's "assume program take-up" sentence in both records and the README, and why it does not settle Medicare enrollment (the benchmark's DEFAULT_TAKEUP_INPUTS has no Medicare entry; the request asks eligibility only). - Name both ACA coverage tests in the mechanism section. - leaderboard_impact.py recomputes the freeze's legacy impact_summary_by_model.csv in every case (the unchanged copy reproduces the frozen file). The stale-weight check now reports that file, which reorders 10 of 46 models with 2.15.17's weights, so the finding is narrowed to "no change in the analyze payload". - Read #191's records from a sha256-checked copy at its head 8af912a, so this directory does not depend on #191's branch. - Add the four workflow reviewers' reports (verification/independent_reviews.json). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
scenario_114 annotation check, for the release that adopts these records (from #197, which fixes scenario_031's Medicaid text and leaves 114 to this PR, as agreed) The README's "Related findings" says the adopting release rewords both 114 case notes. The row annotations need it too, because excluded outputs keep their row text as description. Recomputed on 2.15.17 +
How the reference medical figure is built:
Full agent report (row-by-row, with suggested wording)
|
…_114 row check Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Shipped in release |
Summary
policyengine-us 2.15.17 treats every Medicare-eligible person as enrolled (
takes_up_medicare_if_eligibledefaults to true) and adds the modeled Part B premium to medical expenses (medical_expense_health_insurance_premiums). For 2026 that premium is the $202.90 monthly standard premium, $2,434.80 a year. The engine's IRMAA lookback reads 2024 MAGI, which no household sets, so no income-related amount applies.No PolicyBench prompt states Medicare enrollment or a Part B premium. The prompt rules say to treat unlisted numbers as 0 and not to infer expenses or health coverage.
This PR adds
reference_audit/2026-10-05-medicare-part-b/. It changes no reference and no published number. The exclusion it proposes changes published scores, so it waits for Max's ruling, alongside #191 (decision d963).What the sweep finds
All 1,984 outputs were recomputed on the reference system (2.15.17 +
latest_final); the baseline reproduces all 1,928 scored references. Each output was then rerun with each person's Part B premium at 0, with enrollment off (two ways), and with IRMAA read on 2026 MAGI.Who is charged. 39 people in 33 households are Medicare-eligible, all by age. The premium is charged to 35 people in 30 households; the Medicare Savings Program pays it for the other 4.
What moves. Exactly two outputs move by more than $1, both scored, both in scenario_114 (VA, age 69, itemizes):
What does not move. No SNAP or eligibility output changes by any amount under any reading. The README explains household by household why the other 29 do not move:
Model answers. No model is within $1 of either frozen value.
Records
proposed_exclusions.jsonholds the Virginia record.Leaderboard impact
From
verification/leaderboard_impact.json, viapolicybench.cli analyzeon scratch copies. The unchanged copy reproduces the published payload.Related findings
Invariants checked
not_enrolledandnot_enrolled_directagree on all 1,984 outputs.no_part_b.Review
Independent reviews (a mechanism trace with hand recomputation, a differential re-implementation, a prompt and answer audit, a completeness critic, and an adversarial Subfleet review) are running against this head. Their reports will be added under
verification/.🤖 Generated with Claude Code