Skip to content

Compare IRS filing seasons and select annual target vintages - #1678

Open
MaxGhenis wants to merge 5 commits into
mainfrom
us-irs-refresh-20261118
Open

MaxGhenis wants to merge 5 commits into
mainfrom
us-irs-refresh-20261118

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Changes

Compare IRS filing-season measures at the same processing week across tax years, including itemization rates by adjusted gross income. The report records both cutoff dates, the input digest, units, and missing observations. The annual fiscal compiler excludes partial-season controls and lists them in its exclusion receipt. It retains the four existing TY2024 week-47 earned income tax credit series.

Use the latest complete county source year available for each state's control, subject to the target year. Select the annual IRS district file that matches the source boundaries of the supplied crosswalk. The packaged 118th-to-119th Congress crosswalk supports the 2023 file. The previous crosswalk remains available for the 2022 file.

Verification

Built and validated two Chronicle bundles covering weeks 21 and 30 for tax years 2024 and 2025. The combined input contains 7,888 observations. The reproduction guide lists the three input digests and the comparison command. Regression tests cover same-week reporting, partial-season exclusions, historical source fallback, and conservation of district totals.

Candidate annual registries

The experiment includes a small reconstruction recipe for the annual IRS source overlay. It pins the input commits and digests and checks the generated bytes. Under existing target selection and exclusions, the 2025 registry contains 32,381 controls compared with 33,219 in the base. Its 1,484 non-IRS controls match the base. The 2024 candidate has 32,411 controls compared with 33,249 in the base. The additional county surface reconciles 6,286 child controls over 102 state-by-measure blocks.

The experiment compiles target registries without fitting weights or changing the active production pins. The source dependency is Chronicle #444.

@MaxGhenis
MaxGhenis marked this pull request as ready for review November 18, 2026 16:02
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

Could you show in the description that the change leaves the US target registry unchanged on a feed built to the packaged scope? Two digests would show it, from the registry compiled at main and at this head on one such feed. The same comparison for the two ACS local-area surface tools would cover the county and district changes.

I expect the registries to match, from reading a1fd41cc9 against packages/microcosm-build/src/microcosm/build/us/chronicle_feed_scope.json:

  • compile_us_fiscal_target_registry now drops partial filing-season facts (fiscal_targets.py:1110-1112). The scope lists four filing-season record sets, the week-47 EITC series the filter keeps.
  • translate_congressional_district_facts_to_current_vintage now calls select_irs_district_source first. That function returns the facts as given when the feed has neither irs_soi.ty2023.congressional_district_2023.all_returns nor irs_soi.ty2022.congressional_district_2022.all_returns, and the scope lists one district record set, irs_soi.ty2023.congressional_district_2022.all_returns.

If the digests differ, docs/us-release-build-rule.md asks for a from-scratch build and certification receipt from this tree before the merge.

Five other open PRs regenerate the same four pin files (us-f0-coverage.json, inventory_coverage.py, test_spec_engine_loader.py and test_us_multispine_pool_tool.py): #1667, #1676, #1682, #1684 and #1686. Whichever merges first sends the others back for a re-pin and fresh CI. Since #1676 moves the US engine and prepares the next Route A build, please merge this one only once we've agreed where it goes relative to it. The source dependency, PolicyEngine/chronicle#444, hasn't merged yet either.

@MaxGhenis

Copy link
Copy Markdown
Contributor Author

This one goes after #1676 and re-pins on top of it. Besides the five I listed earlier, these open PRs edit the same four pin files: #1538, #1453, #779, #1085, #1136, #1137, #1156, #1208, #1221, #1275, #1330, #1462, #1564, #1575, #1665 and #1683.

If #1610 lands first, please also show the same comparisons on its feed, at Chronicle 60266a30 in place of 41b2fb2b. The scope in #1610 lists the same four filing-season record sets and the same district record set as main's, so I expect the registries to match there as well.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant