Handle the six skipped tests

[martinholmer]

Currently, there are six tests that are being skipped.
It seems as if each of these tests should be examined and a determination make about its status:

• If the test adds nothing, then it should be removed.
• If a revised version of the test could add something, then the test should be revised and not skipped.

[donboyd5]

@martinholmer, here is an analysis of the skipped tests from Claude, with my edits, and with my tentative plan. I have started reviewing them. Please weigh in with any thoughts you have. Because expected values are likely to change relatively frequently over the next several weeks as we iterate over improvements to 2022 data, I will put in place a mechanism that allows easy updating of expected values.

1. test_tax_revenue.py#L31
   Reason skipped: Currently, expected income tax revenue in 2025+ is pre-OBBBA (the One Big Beautiful Bill Act changed revenue projections).

What it tests: Compares TMD-computed income tax and payroll tax revenues for 2023–2033 against expected fiscal year revenues from YAML files. This is the only test that validates multi-year forward projections via growfactors.

Value: HIGH. This is a critical end-to-end validation. Without it, there's no automated check that TMD produces reasonable aggregate tax revenue over time. The fix is straightforward — update expected values for post-OBBBA law. Until then, this is a significant gap.

Plan: @donboyd5 will prepare a PR to implement, with proper expected values.

2. test_variable_totals.py:14
   Reason skipped: "See issue Improve test_variable_totals: update expected values and tighten tolerances #410: expected values need updating"

What it tests: Compares weighted totals for every TMD variable against expected values (45% tolerance or $30B); even when the test is run, many variables are intentionally skipped. The loose tolerances and skipped variables make the test of no value at present. Properly implemented, this is a broad regression test across the entire variable set.

Value: HIGH. This is the closest thing to a comprehensive data quality check. The running tests (test_weights, test_population, test_partnership_s_corp_income) only cover a handful of variables. Without this, a variable could be wildly wrong and no test would catch it. Updating the expected values is presumably low-effort.

Plan: @donboyd5 to prepare PR focusing on key variables, with tight tolerances.

3. test_income_tax (test_misc.py:45
   Reason skipped: No reason given.

What it tests: Validates TAXYEAR income tax statistics — total weighted count (194M), high-AGI filer count and tax amount ($911B from >$1M AGI filers), and zero-AGI counts by filing status.

Value: MEDIUM-HIGH. The running tests check population and basic weight sanity, but nothing else validates the distribution of tax liability (e.g., that high-income filers pay roughly the right amount). This tests something no other running test covers — the tax computation results themselves for the base year.

4. test_create_file (test_create_tmd_variables.py:10)
   Reason skipped: No reason given.

What it tests: Calls create_variable_file(write_file=False) — essentially a smoke test that the variable creation pipeline runs without errors.

Value: LOW. The make test target already depends on tmd_files, which runs the full pipeline. If variable creation fails, the build itself fails before tests even start. This test is redundant with the Makefile dependency.

Plan: @donboyd5 to prepare PR to delete the test.

5. test_imputed_variable_distribution ([test_imputed_variables.py:176](vscode-webview://1hrd58l03g91pncs5cb7e06mfunahl960bftbrkf4j6ikst3afrk/tests/test_imputed_variables.py#L176))
   Reason skipped: No reason given.

What it tests: Checks that imputed variables (overtime_income, tip_income, auto_loan_interest) have expected mean and standard deviation.

Value: MEDIUM. The other imputed variable test (test_obbba_deduction_tax_benefits) runs and checks tax benefit impacts, but nothing running validates the statistical properties of the imputations themselves. If imputation drifts (e.g., due to a random seed change or data update), this would catch it.

Plan: @donboyd5 to update expected values.

6. test_tmd_stats (test_tmd_stats.py:10)
   Reason skipped: No reason given.

What it tests: Generates describe() statistics for every TMD variable and diffs them against an expected file.

Value: MEDIUM. Overlaps significantly with test_variable_totals (#2 above) but is more granular — it checks min, max, quartiles, not just weighted totals. However, it's also more brittle (any small change triggers a diff). If #2 were re-enabled, the marginal value of this drops.

Plan: @donboyd5 to fold, as appropriate, elements of this test into test_variable_totals, perhaps renaming that test.

What's left uncovered (gaps)?

The following are important areas with no test coverage that merit exploration:

Growfactors / uprating correctness. No test validates that the growfactors in tmd_growfactors.csv are reasonable or that uprating from 2015→2022 produces sensible distributions. test_tax_revenue partially covers this for out-years, but nothing checks the base-year uprating itself.

Distributional accuracy by income group. Tests check aggregates (total revenue, total population) but nothing validates that income is distributed correctly or plausible across deciles/percentiles. A file where all income was concentrated in one record could pass every running test.

Individual itemized deduction amounts. No test (running or skipped) validates that SALT deductions (E18400, E18500), mortgage interest, charitable contributions, etc. are in a reasonable range.

Comparison to IRS SOI aggregates. The national_targets_pipeline tests exist but are --ignored by make test. There's no running validation that TMD aggregates match published IRS Statistics of Income data.