Emit PE one-time rebates in srebate and add --net-of-rebates comparison scoring#1070
Merged
Conversation
TAXSIM books one-time state rebates in payout-year siitax (reported in its srebate column); PE books them to the liability year. The same rebate therefore breaks the match twice — once in each year, opposite signs (#1068, convention #716). Emit PE's srebate as a computed difference: state_income_tax with the 16 one-time rebate variables forced to zero (lazy twin sim, set_input before any calculation) minus actual state_income_tax. This is exactly the amount rebates moved siitax — immune to variables that exist but are not wired into state tax or are liability-capped (a variable-sum approach broke 312 records in 2022 via such phantoms). Add --net-of-rebates to compare, scoring state tax as siitax + srebate on both sides; composable with --rel-tolerance, federal untouched. eCPS validation (8k records/yr, at 0.1%-AGI tolerance): 2021 state 75.50% -> 87.08% (+11.58), federal unchanged 90.29% 2022 state 79.30% -> 83.30% (+4.00), federal unchanged 89.53% Sanity: ME 2021 srebate = 850.00 (relief payment), VA 2021 = 250.00. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
_extract_vectorized_results gained rebate_free_sim_factory; the timed_extract test wrapper had a fixed 3-arg signature. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jul 7, 2026
PavelMakarchuk
added a commit
that referenced
this pull request
Jul 7, 2026
…mode (#1072) Regenerate the full-population comparison (111,347 records x 2021-2025) on current main, picking up the July fixes: transfer zeroing (#1049), pass-through QBID routing (#1051), transfers mapping (#1052), MD county-tax parity (#1066), and the srebate column (#1070). Summaries now carry net-of-rebates state metrics (stateMatchPctRelNet / statePctRelNet): the relative-tolerance match with one-time state rebates netted out of siitax on both sides, removing the TAXSIM payout-year vs PolicyEngine liability-year timing convention (#1068). The dashboard gains a third tolerance mode surfacing it; federal metrics are unchanged by rebates. Full-data downloads point at the full-ecps-comparison-2026.07.07 release (flags: --assume-w2-wages; --disable-salt dropped - inert on the eCPS and overstates federal tax for 2025 itemizers, see #1058). Headline (1% of gross income): state 87.5/88.5/92.3/96.1/96.2 -> net of rebates 92.0/91.4/93.3/96.2/95.9; federal 98.1/97.7/97.6/98.7/98.7. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jul 7, 2026
Open
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
TAXSIM books one-time state rebates in payout-year
siitax(reporting them in itssrebateoutput column); PolicyEngine books them to the liability year. The same rebate therefore breaks the comparison twice — once in each year, opposite signs (#1068; convention question #716). This PR makes the emulator emit its ownsrebateand adds a--net-of-rebatesoption tocomparethat scores state tax assiitax + srebateon both sides — removing the timing convention from match scoring without changing either engine's behavior.Design: srebate is a computed difference, not a variable sum
PE's
srebate=state_income_tax(one-time rebates zeroed, lazy twin sim) −state_income_tax(actual) — exactly the amount rebates moved siitax.Why not sum the rebate variables? We tried — it failed validation (2022 state match −2.75 pts): several PE rebate variables compute amounts that never reach
siitax(mt_income_tax_rebate$1,250,sc_2022_rebate$800 — not wired into state tax; GA amounts liability-capped away), creating 312 phantom mismatches. The difference-based design is immune by construction.Validation (8,000-record eCPS, scored at 0.1%-of-AGI tolerance)
Sanity: ME 2021
srebate= $850.00 (relief payment), VA 2021 = $250.00 — to the penny, matching the mirror analysis in #1068.Details
--net-of-rebatesis opt-in; thesrebatecolumn is additive output (idtl 2/5). Composable with--rel-tolerance.srebateis extracted) from the same dataset; rebate variables are zeroed viaset_inputbefore any calculation (formula variables are silently recomputed otherwise — same lesson as Zero PE-imputed transfers via set_input (fixes MA Senior Circuit Breaker, #1031) #1049).tests/test_net_of_rebates.py(6 tests — comparator both-directions, non-blanket-pass guard, ME $850 e2e). Full suite passes.Refs #1068, #716, #1059.
🤖 Generated with Claude Code