Fix WA property tax exemption crash for households outside Washington

[MaxGhenis]

Status. The code and tests are final. Still pending: the independent review, and the full-population microsim for 2026 and 2027, which is queued behind the shared heavy lock. The runner checks every record: non-WA results are zero, tier is 0–3, and the exemption is between 0 and the property tax. Results will go in a comment.

Problem

A full-population Microsimulation on the default dataset (populace_us_2024) crashes when it computes wa_senior_disabled_property_tax_exemption:

File ".../senior_disabled/income/wa_pte_income_eligible.py", line 24, in formula
    return income <= p.tier_3[county]
ParameterNotFoundError: The parameter 'gov.states.wa.dor.property_tax_exemption.senior_disabled.income.thresholds.tier_3.MARSHALL_COUNTY_SD' was not found in the 2027-01-01 tax and benefit system.

The crash turned up while running the #10103 microsim on 2026-10-11. The tree it ran on was the #10085 head, close to main.

wa_pte_income_eligible and wa_pte_tier index the county-keyed WA income thresholds with every household's county_str. defined_for = StateCode.WA does not stop the formula from running on other households. In policyengine-core 3.32.15, Simulation._calculate runs the formula over the whole population and only applies the defined_for mask afterwards. It skips the formula only when no household passes the mask. So any simulation with at least one WA household and one household elsewhere raised the error. A WA household whose county is UNKNOWN raised it too.

This is not specific to 2027. A 51-household simulation, one household per state plus DC, raised the same error for 2024, 2026 and 2027 on upstream/main (ae3782f). Full runs have not hit it because nothing outside the property_tax_exemption folder uses these variables. The exemption does not feed household_net_income.

Fix

Mask the county before the lookup, the same way md_flat_rate_county_tax, md_applicable_local_tax_rate and in_county_tax do:

county = tax_unit.household("county_str", period)
in_wa = tax_unit.household("state_code_str", period) == "WA"
county = where(in_wa & (county != "UNKNOWN"), county, "ADAMS_COUNTY_WA")

• Households outside WA get Adams County thresholds, and the defined_for mask then sets their result to zero, as before.
• WA households with an UNKNOWN county get Adams County thresholds instead of a crash. Adams County is what the model's county formula (via first_county_in_state) assigns a WA household that gives no county. This matches the MD precedent (issue md_withheld_income_tax: ParameterNotFoundError for MD household with county=UNKNOWN (no UNKNOWN guard) #8975, Allegany County).
• WA households with a real county are unchanged: all 39 WA counties are keys in every tier file, so the guard never fires for them.

Tests

The new vectorized YAML cases put WA and non-WA households in one simulation. A single out-of-state household would not test the guard, because an all-false defined_for mask skips the formula entirely.

• wa_pte_income_eligible Case 12 (2024): King County, South Dakota, and two UNKNOWN-county households at and $1 over the Adams County Tier 3 threshold. Expected [true, false, true, false].
• wa_pte_tier Case 15 (2027): King County at the 2027 Tier 1 threshold after the $7,500 standard deduction, South Dakota, and UNKNOWN falling to Adams County Tier 2. Expected [1, 0, 2].
• wa_senior_disabled_property_tax_exemption Cases 19 and 20: the 2026 and 2027 King County households from Cases 16 and 17, each beside a Marshall County, SD household. Expected [4_320, 0] and [5_340, 0].

All four cases fail on upstream/main with the ParameterNotFoundError above and pass with the fix. All WA YAML tests pass locally (824).

Invariants

• Unchanged for WA households with a real county. county_key == county_str whenever state_code_str == "WA" and the county isn't UNKNOWN. Every WA county in the County enum (39) is a key of tier_1, tier_2 and tier_3.
• Zero outside WA. wa_pte_tier == 0, wa_pte_income_eligible == false and the exemption is zero for non-WA households (the defined_for mask, unchanged).
• No unknown key reaches the lookup when each household's county is in its own state or is UNKNOWN. In that case every key passed to the threshold lookup is a WA county key, so ParameterNotFoundError can't be raised for any household mix. In the default dataset every household's county FIPS matches its state. A household whose county is in a different state from its state_code (say WA with an Oregon county) still raises, as it does in the MD and IN formulas.

Other county-keyed lookups (checked, not changed here)

I probed the other formulas that index a county-keyed parameter, with one household per state and DC in one 2026 simulation, then again with UNKNOWN-county households added:

• Already guarded against other states' counties: MN CCAP, MD local tax, MD CCS, IN county tax, CO CCAP, WI Shares, and the county-group formulas (PA and KS TANF; NY, IL and TN CCAP; VA CCSP).
• vt_renter_credit has the same out-of-state crash. It indexes VT county tables with the county enum and has no guard. Only taxsim_state_property_tax_credit uses it, so full runs don't compute it. It also crashes for tax units larger than 8, which needs a policy decision.
• MN CCAP, WI Shares and MD CCS crash for an in-state household whose county is UNKNOWN. They mask other states' counties but not UNKNOWN. In the default dataset every household has a valid county FIPS that matches its state, so only household input with county: UNKNOWN reaches this.

I'm proposing these as a separate follow-up.

axiom: n/a: microsim crash guard, no change to any WA rule; real-county WA results are unchanged

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (909176a) to head (2228eb7).
⚠️ Report is 404 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main    #10113   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            4         2    -2     
  Lines           76        30   -46     
  Branches         2         0    -2     
=========================================
- Hits            76        30   -46     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.