The path to a complete US tax-prep engine. Everything below is either shipped & certified, certifying (built, correctness being locked down), or planned — with the goal of taking raw tax documents all the way to a complete return, ready to file.
✅ means live; tax features are also checked against IRS and state sources. The order is a working plan and can change. Updated 2026-10-30. Select any feature for what it covers and its limits. How to use what is live is in the docs, which list each live feature's status on the page that documents it.
Forms: Form 1040
certified (ordinary-bracket + standard-deduction + preferential-LTCG 0%/15% stacking, all filing statuses).
Forms: Form 1040
independently verified vs IRS Rev. Procs 2019-44 / 2020-45 / 2021-45 / 2022-38 (§1(h)), all filing statuses, primary-source-quarantined oracle. Scoped: the 20% LTCG band, per-year AMT, and CARES/ARPA-era credits (CTC / EITC / UI-exclusion / advance-CTC / RRC) are NOT part of this cert.
Forms: Form 1040, Sch D
Forms: Form 8960
Forms: Form 8959
Forms: Form 1040
Forms: Form 6251
backstop-only: computes AMT from modeled current facts. Standalone AMT inputs (ISO bargain element, private-activity-bond interest, depreciation preferences) are not caller inputs. Form 6251 line 2j (estate/trust Schedule K-1 (Form 1041) box 12 code A) is $0 in the engine; since 2026-10-05 a return with interest, dividends, positive tax-exempt interest, Schedule D or Schedule E input declines unless the caller attests no positive code A and no code B-I amount (founder decision D3). Not detected: a code A K-1 with no income boxes, on a return with none of those inputs. The general Schedule D line-2k channel is left open on purpose because §56(a)(6) bounds it to over-tax.
Forms: Form 6251
FIX SHIPPED: added `mortgageInterestSecuresAmtQualifiedDwelling: z.boolean().optional()`, required whenever `mortgageInterest` is POSITIVE (a $0 amount needs no affirmation — it has a $0 addback by definition). `true` computes exactly as before (a qualified dwelling has a $0 Form 6251 Line 3 addback by definition); `false`/absent declines (`itemized_deduction_needs_review`), naming Form 6251 Line 3 and the houseboat/RV/transient-use exclusion. The field automatically inherits to the state schemas, so the gate fires identically on state returns. CORRECTED 2026-09-01 (see fed-amt-6251-mortgage-principal-residence-prong below): 26 U.S.C. \u00a756(e)(1) has a SEPARATE, disjunctive principal-residence prong (\u00a756(e)(1)(A)) with no dwelling-type restriction, under which a genuine \u00a7121 principal-residence houseboat DOES qualify. The field this note describes (mortgageInterestSecuresAmtQualifiedDwelling) models only the OTHER prong (\u00a756(e)(2)); the sibling leaf below adds the missing prong as a second, independent affirmation.
Forms: Form 6251
§57(a)(7) makes 7% of the EXCLUDED §1202 gain an AMT preference (Form 6251 Line 2h) for qualifying small-business stock acquired before 2010-09-28 (post-2010-09-27 stock is exempt because amended §57(a)(7) reaches only stock acquired on or before 2010-09-27 -- formerly expressed via §1202(a)(4)(C), struck by Pub. L. 119-21 §70431(a)(4) and made retroactive by §70431(a)(6)(B); CORRECTED 2026-09-08). The engine accepts the TAXABLE portion via collectibles28Gain but has no operand for the EXCLUDED portion and hardcodes otherPreferences: 0 (compute-year.ts), so a pre-2010 §1202 return would COMPUTE instead of decline and silently under-tax. FIX SHIPPED (fail-close): added collectibles28NoSec1202ExclusionPreferenceAffirmed: z.boolean().optional() to federal-profile.ts's schema, with two gates in computeFederalPipeline mirroring the pre-existing §1256 orphan-affirmation pattern exactly — (1) a positive collectibles28Gain without the affirmation === true declines (special_rate_gain_unmodeled, naming Form 6251 Line 2h and the §57(a)(7)/§1202(a)(4)(C) mechanics); (2) the affirmation supplied (true OR false) without a positive collectibles28Gain declines as an orphan/contradictory attestation, same contract as the three §1256 affirmations [SUPERSEDED by round 10, 2026-09-06: both the requiring trigger and this orphan check are now keyed on the PRESENCE of any Schedule D input, not on a positive amount — see this leaf's round-10 evidence entry]. true attests EITHER no §1202 exclusion is involved at all (pure §408(m) collectibles), OR every §1202 exclusion on the return arises solely from post-2010-09-27 stock (100%-excluded, no preference) — computes exactly as before (zero engine change to the actual AMT math; the affirmation only gates access, matching the mortgage-dwelling-gate precedent).
Forms: Form 6251
§56(e) qualified housing interest has two DISJUNCTIVE prongs: (A) interest on debt for property that is the taxpayer's §121 principal residence at the time the interest accrues — NO dwelling-type restriction (Pub 523 explicitly lists a houseboat as a possible main home) — or (B) an AMT qualified dwelling (house/apartment/condominium/mobile home not used on a transient basis, §56(e)(2)) that is a §163(h)(5) qualified residence. mortgageInterestSecuresSec121PrincipalResidence (prong A) and mortgageInterestSecuresAmtQualifiedDwelling (prong B) are independent affirmations; the gate passes when EITHER prong is affirmed true, declines when NEITHER is, with decline prose naming both prongs and their governing statutory citations. Full adjudication: .cert/disputes/fed-amt-6251-2026-08-29.md. CORRECTED/EXTENDED 2026-09-02: §56(e)(3) (pre-1982 indebtedness) is a THIRD, still-unmodeled qualifying category, disclosed (over-decline only). SEPARATELY, a live under-tax was found and FIXED: the refinancing-limitation doc-comment's cost-basis test was wrong (repealed pre-1988 rule) and wrongly AMT-scoped (§163(h)(3)(B)(i) imposes the identical cap for regular tax) -- new affirmation mortgageInterestExcludesExcessRefinancingProceeds fences it, gated in the regular-tax deduction section, orthogonal to this leaf's own AMT two-prong gate.
Forms: Form 6251
Declines unless the caller attests no AMT difference: a Form 6251 line-2m attestation for any Schedule E rental net (gain, zero or loss), and a carryover attestation for any return with a Schedule D input, including a carryover already netted into the gains. An AMT-only capital-loss carryover must be sent as a $0 regular carryover field, which declines. The AMT difference itself is never computed.
Forms: Sch A
independently verified 2026-07-01; medical and cash charitable are the fed-itemized-medical-charitable-cash item
Forms: Form 1040
canBeClaimedAsDependent is required on every request; true applies the year's Rev. Proc. amounts, and CA/NY/IL decline it. Declines: possible kiddie tax (income lines other than wages and Schedule C, losses as $0, plus Schedule C net of half SE tax, over twice the floor), a dependent claiming dependents, a joint or head-of-household dependent, a positive Schedule C without scheduleCPersonalServicesCapitalNotMaterialAffirmed.
Forms: Form 1040
filerAge65OrOlder/filerBlind, and the spouse's on a joint return, add the year's Rev. Proc. amount after the dependent limit. MFS spouse boxes are not applied.
Forms: Form 1040
spouseItemizesDeductions is required on MFS: yes zeroes the standard deduction, unknown declines.
Forms: Form 1040, Sch 1-A
Taken with seniorDeductionEligibilityAffirmed and a 65+ field. Reduces the QBI base and the §68 base, added back for the AMT and the capital-loss carryover.
Forms: Sch A
Preparer-attested profile inputs only; a medical or charitable document on the doc-ingest path still declines. From 2026, a cash return with a non-cash gift or carryover deductible below its value declines.
Forms: Sch 1, Sch A
Winnings go into AGI; losses itemized up to winnings, from 2026 90% of losses (§165(d), OBBBA §70114). One attestation covers both, figured wager by wager.
Forms: Form 4952, Sch A, Form 6251
Deducted up to interest plus nonqualified dividends; the excess carries forward. Current-year short-term items and any tax-exempt interest decline; capital-loss carryovers are admitted. The AMT carryforward is taken as 0.
Forms: Sch A, Form 8283
Worksheet 2 lines 8-10. From 2026 this engine deducts no non-cash gift when gifts exceed 60% of AGI.
Forms: Form 4684, Form 1040
Per item, less $500 per entry; deducted outside Schedule A, itemizing or not (MFS and property partly owned outside the return decline), and not added back for the AMT. The declaration is checked against a FEMA major-disaster snapshot.
Forms: Sch A, Form 4684
no input today
Forms: Form 8995
independently verified 2026-07-02; 20% deduction, overall-limit declines; business/PTP-K-1/SSTB QBI stays out of scope
Forms: Sch D, Form 8949
long-term/QDIV preferential bands certified; short-term taxed as ordinary
Forms: Sch D, Form 8949
independently verified (exact bounded label). Noncovered / missing-basis / $0-basis / adjustment-coded boxes → whole-return decline (no tax-number leakage); a broker-reported code-W wash disallowance now computes NARROWLY under two affirmations (every other wash shape still declines); multi-broker computed returns carry a cross-account/cross-spouse wash-sale VERIFY flag. Broad netting, wash-sale-safe multi-broker, state aggregates, and prior-return-derived carryforward chains remain certifying/planned (below).
Forms: Sch D, Form 8949
the covered-clean A/D aggregate path is shipped ✅ above. The cross-account/cross-spouse wash-sale gate is now addressed by the shipped lot/evidence contract's affirmation (2026-07-04). A NARROW broker-reported code-W wash-sale compute (two affirmations, whole-dollar per-entry add-back) is deployed; lot-level detail and broader wash mechanics (replacement-basis linkage) remain planned. RE-RUN round 1, 2026-09-01 (full record: .cert/disputes/fed-schedule-d-broader-2026-09-01-round1.md): Opus's independent primary-source re-derivation CERTIFIED the base netting path and the wash-sale add-back, and confirmed every claimed out-of-scope decline holds — but found a real, over-tax-safe defect in the §1211(b)/Pub 550 Capital Loss Carryover Worksheet: the engine passed the FLOORED-at-zero Form 1040 line 15 value into the worksheet's own line-3/4 limiting test instead of the unfloored (possibly negative) value the worksheet's own line-1 instruction requires, so the test could never bind and the engine silently destroyed up to $3,000/$1,500 of carryforward per year with zero current-year tax benefit for any filer whose deductions exceed their other income. FIXED: pass the unfloored taxable income into the carryover worksheet only (the displayed/taxed Form 1040 line 15 remains floored, unchanged).
Forms: Sch D
Forms: Form 6781, Sch D
On the federal whole-return aggregate scalar surface, a positive IRS-reported section1256NetGain computes only with section1256CleanBrokerReportedAffirmed:true; a negative aggregate computes only on the no-election path when BOTH section1256CleanBrokerReportedAffirmed:true and section1256LossCurrentlyDeductibleAffirmed:true are present. The aggregate splits statutorily 40% short-term / 60% long-term into Schedule D + the §1411 NIIT base; a net of exactly zero and tagged §1256 box/lot inputs still decline. §475(f), §988, straddles/§1092, §1212(c) carryback elections, states, and mixed special-rate + §1256 loss interactions remain excluded. A corrective source revision (fed-section-1256-6781-contract-and-mixed-edge-repair) additionally requires TAX YEAR 2025 ONLY (a certification-evidence boundary, other years decline), a new section1256AggregateIsFinalNoAdjustmentOrDeferralAffirmed:true attestation (no Form 6781 line-4 adjustment or QOF deferral) on BOTH the positive and loss paths, and enforces the mixed special-rate + §1256 exclusion above by decline (any §1256 affirmation supplied without section1256NetGain also declines). Certified 2026-09-01 after 3 rounds that found and fixed two under-taxes (the §1402(i) dealer and §1256(f)(4) limited-partner affirmations).
Forms: Form 8949
Computes Schedule D from clean covered A/D 1099-B box aggregates or lots (digital assets need cleanCoveredNoAdjustmentsAffirmed too), at most one box total per broker, account and box letter, under caller-attested wash-sale completeness covering spouse, cross-account, IRA and a corporation the taxpayer controls; everything else declines. Record: .cert/dispositions/fed-lot-evidence-contract.md.
Forms: Form 8949
The bound — `Math.round(box.washSaleDisallowed ?? 0) > Math.max(0, Math.round(effReportedBasis + equityCompAdd) - Math.round(box.proceeds))` — is scoped to `(box.granularity ?? "box") === "lot"` ONLY (a genuine single atomic transaction, where the entry's own net truly is that one sale's own gain/loss). Box-granularity aggregates (the default, and every pre-existing wash-sale fixture in this repo) are UNAFFECTED and remain an existing, disclosed, unfixed gap for the box case specifically (same over-tax-safe direction, not newly introduced). A no-loss atomic lot (proceeds ≥ basis) with any positive disallowance declines.
Forms: Form 8949
Root cause shared by both 2026-08-29 dispute findings: the shipped wash-overstatement bound INFERRED atomicity from the caller-controlled OPTIONAL granularity label instead of requiring it to be affirmed. FIX SHIPPED, both prongs: (1) OMITTED-GRANULARITY GAP — an overstatement-shaped entry (washSaleDisallowed exceeding the entry's own net) with granularity omitted now DECLINES instead of silently computing the fabricated gain the guard was shipped to close. (2) AMBIGUOUS-LOT-LABEL GAP — the atomic-lot 'impossible disallowance' decline now fires ONLY on an explicit `washLotIsSingleAcquisitionBlockAffirmed:true`; an unaffirmed granularity:'lot' entry showing the overstatement shape still declines (unchanged over-tax-safe outcome) but with ambiguity-honest prose, never the false atomicity assertion. DESIGN DECISION: an affirmed-COMBINED lot (`washLotIsCombinedMultiBlockAffirmed:true`) computes exactly like a box aggregate — same operands, same formula, no new fields needed, since a disclaimed-atomicity lot carries no more information than a box does. Both new affirmations are mutually exclusive (both true → decline) and `washLotIsCombinedMultiBlockAffirmed` contradicts a present `acquiredDate` (one col-(b) date cannot represent a combined multi-block sale) → decline. Either atomicity affirmation on a non-lot entry declines. The acquiredDate-contradiction check this note describes above turns out to be OVER-BROAD: it declines ANY combined-affirmed row carrying an acquiredDate, even when every underlying block genuinely shares that one date — a legitimate case per Form 8949's own instructions (VARIOUS is described as something a multi-purchase block MAY use, not a requirement). Opus flagged this as a non-blocking usability residual; the real Codex CLI DISPUTED on it. Both agree the direction is over-decline-only (no §6694 exposure, never a wrong computed number). Scoped as fed-lot-evidence-contract-same-date-combined-block-fix below.
Forms: Form 8949
The `washLotIsCombinedMultiBlockAffirmed:true` + `acquiredDate`-present contradiction check (added in fed-lot-evidence-contract-wash-atomicity-affirmation, round 3) declined unconditionally on the theory that 'a single Form 8949 col-(b) date cannot honestly represent a combined multi-block sale.' Form 8949's own instructions say a multi-purchase block MAY use VARIOUS in column (b) — not that it MUST — so a combined sale where every underlying block genuinely shares one acquisition date can truthfully carry that real date, and the check falsely rejected it as contradictory. FIX SHIPPED: removed the check entirely (not narrowed) — the decisive fact is that a combined-affirmed row with acquiredDate OMITTED already computed BEFORE this fix (identical formula, falls through to box-style compute), so a caller could always get the SAME result more cheaply by omitting the date; supplying a truthful date therefore never creates new §6694 exposure relative to the already-accepted dateless path. The pre-existing generic term-consistency check (acquiredDate/soldDate cross-checked against the box-letter-implied holding period, independent of the combined/single affirmation) still independently declines a term-inconsistent or unparseable date pair on this same path — acquiredDate was never an operand in the gain arithmetic itself on any path, only ever consumed by that separate check. The mutual-exclusion check (both atomicity affirmations true → decline) is UNCHANGED — this fix touched only the second, disputed check.
Forms: Form 8949
A covered A/D lot with a comp-adjusted basis (broker reported basis + W-2/1099-NEC ordinary-income = corrected basis) computes the Form 8949 code-B adjustment; requires acquired/sold dates, a typed active ordinary-income source ref, and the comp-adjusted-basis affirmation. No comp-created loss; the derivation is exposed but is an internal operand (double-count-safe).
Forms: Form 8949
ISO computes ONLY a DISQUALIFYING disposition (equityCompIsoDisqualifyingDispositionAffirmed:true, fail-closed) — arithmetically identical to the shipped NSO path; a QUALIFYING ISO disposition stays out of scope (its AMT-side basis can differ from the regular-tax basis, a Form 6251 §56(b)(3) exercise-year-preference adjustment this engine does not model). Every equity-comp lot now REQUIRES granularity:'lot' (a box aggregate — the default — always declines, since a single acquiredDate/soldDate pair can't honestly represent many lots), an evidence-backed date pair (exact ISO yyyy-mm-dd, soldDate strictly after acquiredDate), and soldDate in the SAME calendar year as the return's tax year (a lot sold in a different year doesn't belong on this return). ISO additionally requires acquiredDate/soldDate in the SAME calendar year as EACH OTHER: a same-year exercise-and-sale is disqualifying under §422(a)(1) regardless of grant date (the max intra-year gap, 365 days, can never satisfy either qualifying leg), so it is the ONLY ISO disposition shape the engine computes — a same-affirmed lot with cross-year dates still declines (a contradicted affirmation cannot compute; a cross-year exercise/sale needs Form 6251 line-2k remeasurement, out of scope). ESPP computes for EITHER a qualifying or disqualifying §423 disposition with NO ISO-specific same-year constraint (the soldDate-in-tax-year rule above still applies to ESPP too; the caller-attested equityCompOrdinaryIncome covers whichever figure applies; the engine does not derive the §423 'lesser-of' itself; no AMT preference either way, unlike ISO). A computed ISO-disqualifying lot carries a VERIFY flag naming the §422(c)(2) at-a-loss ordinary-income cap. ISO exercise-and-hold (no disposition) has no path through this lot shape — `proceeds` is a required field — confirmed not a build gap. The claim that no compute case can use equityCompType without granularity:'lot' rests on a CODE INVARIANT, re-derived independently by both 2026-09-05 certification passes: equityCompType is read only inside federal-profile.ts, and every compute-path read sits downstream of the single gate whose eqComputable hard-ANDs eqGranularityOk === ((granularity ?? 'box') === 'lot').
Forms: Form 8949, Form 6251
a QUALIFYING ISO disposition's AMT-side basis divergence (Form 6251 §56(b)(3), a prior-year exercise-year preference this engine does not track) remains out of scope; and the $0-basis broker-attested-only override (a broker-attested-but-not-IRS-reported basis) remains the deferred remainder — per founder direction the $0-basis path requires a caller override or fails closed (never guesses basis).
Forms: Form 8949
builds on the shipped lot/evidence contract (form8949Box G–L reserved)
Forms: Form 8949
builds on the shipped lot/evidence contract (lotId/replacementForLotId reserved)
Forms: Form 8949, Form 1099-DA
The 2025 Schedule D instructions and 2026 Form 1099-DA instructions (Box 1i) both confirm the wash-sale rules under §1091 DO apply to the narrow subset of digital assets that are ALSO stock or securities for tax purposes (tokenized securities) — most crypto remains property and stays exempt. Checked directly against the current compute path: this is NOT a live under-tax vector today (the existing wash gauntlet already fail-closes the instant a caller discloses a washSaleDisallowed amount on any digitalAsset:true box, unconditionally, regardless of whether the asset is a tokenized security) — the actual exposure is that the engine's own decline reasoning and derivation-note prose assert a categorical legal fact that is no longer universally true, which could lead a preparer holding a genuine tokenized security to wrongly conclude no wash-sale disclosure is ever needed for any digital asset. Scoped as its own leaf (not folded into fed-lot-evidence-contract) because it is a digital-asset capital-character gating concern, a different code region and design surface than that leaf's box/lot scope-key contract.
Forms: Sch D, Form 6251
PROMOTED NARROWED 2026-10-05 under the 2-round cap; the claim, findings table and residual risk are in .cert/dispositions/fed-1250-collectibles.md. Computes only when the caller attests no AMT-side Schedule D difference, no §1202 line-2h preference and no capital-loss-carryover AMT difference (the last two for any Schedule D input, even $0); a zero-net schedule must be sent as 0. Round records are cited in evidence_refs.
Forms: Form 6251, Sch D
Finding 1: Form 6251 Line 2h's §57(a)(7) preference is keyed to the §1202 EXCLUDED gain (Form 8949 column (g)), not to collectibles28Gain (Schedule D line 18, the NETTED TAXABLE 28%-rate residue this engine's pre-existing gate keyed on) — the IRS 28% Rate Gain Worksheet's own lines 5/6 (a net current-year short-term loss, or a long-term capital-loss carryover) can drive collectibles28Gain to exactly $0 while a large §1202 exclusion (and therefore a large Line 2h preference) survives intact. Executed fixture: a $1,100,000 pre-2010 QSBS gain, 50% excluded ($500,000 excluded → a $35,000 Line 2h preference), the $500,000 taxable residue fully absorbed to $0 by a $500,000 net short-term loss — computed with NO attestation demanded at all under the old trigger, a confirmed $11,375 AMT under-tax. CORRECTED to the UNION of both triggers (netLongTermCapGain>0 OR collectibles28Gain>0), a genuine strict superset of the pre-PR collectibles28Gain>0 trigger. The engine still doesn't model the excluded-gain operand, it now requires the caller to attest none applies. Finding 2: the round-8 scheduleCNoAmtAdjustmentAffirmed gate triggered only on Math.round(scheduleCIncome)!==0, letting an explicitly-supplied $0 Schedule C net (large gross receipts offset by large deductions, one of which — e.g. depletion, Form 6251 line 2d — differs for AMT even though the two regular-tax figures happen to net to zero) bypass the AMT-preference gate entirely. FIX SHIPPED: widened the trigger to scheduleCIncome!==undefined (the loss branch and the two whitelist affirmations are unaffected — they already handled $0 correctly once reached); the SS-wage-coordination gate (schedule_c_se_needs_social_security_wages) was SPECIFICALLY EXEMPTED at exactly schC===0, since computeScheduleSE's own de-minimis branch (netSEEarnings<$400) returns $0 tax before the OASDI-ceiling coordination that gate protects is ever consulted — demanding socialSecurityWages there would be a spurious decline with zero computational effect.
Forms: Sch D, Form 6251
FIX SHIPPED: `scheduleDTaxWorksheet` now returns its internal line14/line21 (already computed, just not exposed); threaded through `Form1040Result` -> `compute-year.ts` -> `computeAMT` as a grouped `regularScheduleDAnchors` object; `amt.ts`'s special-rate branch now inlines Form 6251 Part III lines 19-34 using the two distinct anchors, fails closed (declines, does not approximate) when they're unavailable. SCOPE NOTE: the separate Part III lines 13-17 cap mismatch was fixed by fed-amt-preferential-income-cap-mismatch (#310).
Forms: Form 6251
One literal Form 6251 Part III for both worksheets. Line 13 is the Qualified Dividends and Capital Gain Tax Worksheet's line 4, no longer capped at taxable income (the cap overstated AMT when dividends plus gain exceeded it); taxable income 0 with §1250/28% gain takes the Schedule D Tax Worksheet's lines 2-13 (the form's statement (2)). Form 6251, its exemption worksheet, and the Qualified Dividends and Capital Gain and Schedule D Tax Worksheets round each line, each rate line on its own (docs/decisions/whole-dollar-line-rounding.md); not_modeled.cents_rounding says where that can differ.
Forms: Sch 3
independently verified 2026-07-02; the Form 1116 safe-harbor above the ceiling (fed-ftc-1116-safe-harbor) and full Form 1116 limitation + carryover (fed-ftc-1116) remain separate slices
Forms: Form 1116
WON'T-FIX 2026-10-06 (founder decision D1): Form 1116 is not computed above the §904(j) de-minimis ceiling ($300 / $600 MFJ); those returns decline for a preparer. The safe-harbor code is switched off, not deleted. Record: .cert/dispositions/fed-ftc-1116-safe-harbor.md.
Forms: Form 1116
The categorization question is UNRESOLVED in primary source -- Pub 514's passive list (Dividends, Interest, Rents, Royalties, Annuities) names no Pensions, nor does its general-category text -- and excluding qualified-plan pension from the bound now rests on IRS Pub 519 sourcing a domestic trust pension's investment-earnings component as U.S.-source, plus an EXECUTED safe-direction argument: Pass A's 43-point differential sweep found 0 cases where excluding pension loosens the bound, the delta being exactly round(max(0, pension)) and never negative. A live over-decline counter-consideration remains recorded (an IRS VITA page describes foreign-service pension as passive category income), scoped to the sibling fed-ftc-1116-safe-harbor-foreign-pension-sourcing leaf. The refuted rationales: .cert/disputes/fed-ftc-1116-safe-harbor-2026-08-29-round3.md, .cert/disputes/fed-ftc-1116-safe-harbor-2026-08-31-round4.md, .cert/disputes/fed-ftc-1116-safe-harbor-annuity-category-citation-fix-2026-09-07.md.
Forms: Form 1116
WON'T-FIX 2026-10-06 (founder decision D1): part of the Form 1116 safe harbor, which no longer runs; returns above the §904(j) ceiling decline. Record: .cert/dispositions/fed-ftc-1116-safe-harbor-passive-income-consistency-gate.md.
Forms: Form 1116
WON'T-FIX 2026-10-06 (founder decision D1): part of the Form 1116 safe harbor, which no longer runs; returns above the §904(j) ceiling decline. Record: .cert/dispositions/fed-ftc-1116-safe-harbor-passive-annuity-inclusion.md.
Forms: Form 1116
Codex independently fetched an IRS VITA training page (apps.irs.gov/app/vita/content/0507/0507_02_040.jsp) which describes foreign-service pension income as "pension income (passive category income)" without distinguishing qualified from nonqualified plans, and executed a concrete false-decline fixture: a return with $48,000 of pension income attributable to foreign service plus $2,000 foreign interest, asserting $50,000 total foreign-source gross income, is wrongly DECLINED because declaredPassiveIncome only counts the $2,000 interest. Adjudicated (not resolved) by the orchestrating session: this is a genuinely unresolved, contested area of tax law, and the current blanket exclusion is kept as the SAFE choice — it only ever tightens the bound (over-decline direction), whereas blanket-including the FULL pension amount would risk the opposite failure whenever the pension is NOT actually foreign-service-attributable, which Pub 519 confirms is the common case (a domestic-trust pension's investment-earnings component is U.S.-source by construction; only the employer-contribution portion attributable to foreign service can be foreign-source). This engine currently has no field to isolate that portion. A proper fix requires a new caller-facing field/affirmation, not a blanket compute-path change — scoped here rather than guessed at under time pressure.
Forms: Form 1116, NY IT-201, NY IT-112-C
Independently fetched NY's 2025 IT-201 instructions, IT-112-C-I, and IT-112-C directly (tax.ny.gov): a NY resident who paid income tax to a Canadian province must complete Form IT-112-C and claim a resident credit on IT-201 line 41 -- a credit mechanism wholly independent of, and unrelated to the AGI-neutrality of, the federal foreign tax credit this engine already computes. Executed fixture: $60,000 wages + a $20,000 Canadian-source nonqualified annuity with $5,000 provincial tax withheld, full-year NY resident -- NY computed totalTax $3,796 with no IT-201:41 line at all; the correct resident credit (Form IT-112-C Part 2) is $949, a confirmed $949 NY over-tax. Scoped as its own leaf rather than folded into fed-ftc-1116-safe-harbor directly since it requires a genuine new NY capability surface (Form IT-112-C's own multi-step resident-credit computation), not an attestation fix -- mirrors the fed-pension-5b-ny-orp-portion-limit / fed-pension-5b-ny-nqdc-pension-exclusion precedent from this session's own pension-leaf rounds.
Forms: Form 1116
The prose fixes shipped in the round-4 branch treat the SYMPTOM (eight wrong statements of the rule, now corrected and gated by test/lib/practice/ftc-904j-payee-statement-gate.test.ts). This leaf is the CAUSE. The gate is a prose-level backstop that catches a wrong restatement after it is written; a single shared definition removes the opportunity to write one.
Forms: Form 1116
A downstream consumer application outside this repo maintains its own independently-drifted copy of compute-year.ts wired to a LIVE compute route with the identical unguarded call -- that finding is OUT OF SCOPE for this leaf (a different repo, different deploy cycle) and is tracked separately in this operator's own private cross-project notes, not here.
Forms: Form 1116
Executed fixture: single filer, $25,000 ordinary dividends with qualifiedDividends === ordinaryDividends (100% qualified, i.e. $0 of non-qualified dividends anywhere on the return), asserting foreignSourceGrossIncome: $25,000 with no other passive income source in the fixture (implying all $25,000 of dividends are foreign-source) — yet asserting foreignQualifiedDividendsAndCapGains: $0. Since the return has zero non-qualified dividends anywhere, any subset (including the foreign-source subset) must also be fully qualified, so the true foreign QDI component cannot honestly be less than $25,000; the engine computes anyway using the false $0. Adjudicated (not resolved) by the orchestrating session: real in this specific, mathematically-unambiguous fixture shape, but this engine's schema doesn't cleanly expose which income type(s) compose a given foreignSourceGrossIncome assertion when more than one is present (interest + dividends + capital gain can combine), so a general, always-safe mechanical bound requires a dedicated design pass — building it under time pressure risks either missing real cases (unsafe) or over-declining legitimate mixed-income returns with no primary-source justification for the specific threshold chosen. Scoped here rather than guessed at.
Forms: Form 1116
Surfaced by the orchestrating session's own adjudication during fed-ftc-1116-safe-harbor round 6 (2026-09-01, full record: .cert/disputes/fed-ftc-1116-safe-harbor-2026-09-01-round6.md), not by either verifier directly -- both round 5 and round 6 found real, distinct under-tax vectors that trace to the SAME root cause: foreignQualifiedDividendsAndCapGains is documented and used as a single measure ("net capital gain"), but Form 1116's own instructions use TWO different foreign-capital-gain measures for two different purposes (the $20,000 threshold test vs. Worksheet A's mandatory adjustment), and this engine has never modeled the second one. Round 6's fix (decline on any positive net short-term gain) closes the specific defect found but does so by widening an over-decline rather than building the missing computation -- a real return with, say, a $500 foreign short-term stock gain and an otherwise-clean safe-harbor-eligible return now declines unnecessarily. Scoped here as the proper structural fix, distinct from fed-ftc-1116-safe-harbor-qdi-consistency-bound (which is about verifying a CALLER'S HONESTY against other reported figures, not about the field's own definitional scope). UPDATE (code review, round 6 PR, 2026-09-01): round 6's guard also has a KNOWN, DISCLOSED residual within its own stopgap — it compares the COMBINED (foreign+domestic) net short-term figure, so a real foreign ST gain masked by a larger domestic ST loss (combined net ≤ 0) still evades it silently; independently re-executed and confirmed real (a $3,000 foreign ST gain offset by a $5,000 domestic ST loss still computes with the gain's full amount flowing unadjusted into the numerator). This residual is folded into this leaf's own scope rather than tracked separately.
Forms: Form 1116
everything the de-minimis + safe-harbor slices don't cover: itemized-deduction filers, the adjustment-exception failing (≥$20,000 QDI/cap-gain or income above the 32%-bracket threshold), multi-category income, AMT Form 6251 line 8, and prior/carryover-year FTC
Forms: Schedule 8812
Federal Part I uses each year's Schedule 8812 per-child amount ($2,200 from 2025; 2021 declines), $500 ODC, and the $400,000 MFJ / $200,000 other-status AGI phase-out; it is limited by Form 1040 line 18 less the foreign tax credit, so it can offset the AMT. Positive CTC and ODC counts require both dependent-level and separate filer-level eligibility affirmations. Bounded Part II-A ACTC supports unused CTC, the year's per-child cap, and 15% of line-18a earned income above $2,500, where the exposed line 18a is W-2 wages plus affirmed non-loss single-sole-proprietor Schedule C net minus one-half SE tax. For three or more qualifying children, the whole return declines when the unmodeled Part II-B alternate may increase ACTC. Full Schedule 8812, state child credits, dependent-fact derivation, other earned-income branches, and tax years other than 2025 remain outside this badge.
Forms: Schedule 3-A
Under the draft 2026 Schedule 3-A and proposed Treas. Reg. §1.24-3, the refunded ACTC is paid only on a Yes to lines 7 and 8. Phase 2 (omitted declines) at the earliest of final regulations, the final form, or the 2026 filing season.
Forms: Form 8962
form8962Required:true with premiumTaxCredit computes Parts I-III for 2020 and 2022-2025 (2020 repays no excess, ARPA §9662): line 2a from this return (Worksheet 1-1), the excess advance payment repayment added to tax before credits and the net credit as a payment. 2021, 2026, more than one Form 1095-A, married filing separately, a dependent filer, medical expenses and a tax family other than the filers plus the CTC/ODC counts decline.
Forms: Form 8962
2026 first (no repayment limitation; at or below 400% waits on the 2026 Table 2), then 2021 (line A, the unemployment compensation rule). More than one Form 1095-A, married filing separately, Part IV, Part V, a QSEHRA and anyone enrolled while not lawfully present decline today.
Forms: Form 8863, Form 2441, Form 5695, Form 8936
Forms: Form 8801
Forms: Sch SE
SE tax (§1401/§1402: net×0.9235 → 12.4% OASDI to the $176,100 wage base coordinated with W-2 box-3, then 2.9% Medicare) + the §164(f) half-SE above-the-line deduction + SE→Additional-Medicare (Form 8959) coordination remain the exact bounded claim.
Forms: Form 2210
Source-certified at b8b210b for the TY2025 federal resident Form 1040 modeled surface. Opt-in via priorYearTax. Returns no_penalty only for: (1) de_minimis when modeled current tax minus withholding is under $1,000; (2) withholding_covers_required when withholding covers the lesser of 90% of modeled current tax or the available prior-year harbor, which requires positive prior-year tax and a full 12-month prior return and uses 100% of prior-year tax, or 110% above $150,000 prior-year AGI ($75,000 MFS), with missing prior-year AGI conservatively using 110%; or (3) prior_year_zero when priorYearTax is zero, the prior return covered 12 months, and the taxpayer was a U.S. citizen or resident throughout that prior year. Returns needs_preparer otherwise. For TY2025 only, penaltyCeiling is emitted only when the complete touched-rate schedule is published and encoded; it is a conservative advisory whole-dollar upper bound, does not change tax, AGI, or refund, and is not an exact or filed Form 1040 line-38 amount. Exact per-installment penalty, Schedule AI, 1040-ES planning, special-category rules, waivers, and state regimes remain outside this badge.
Forms: Form 1040, Sch B
Forms: Form 1040
Claim, findings and residuals: .cert/dispositions/fed-pension-5b.md.
Forms: Form 1040, NY IT-201
Executed fixture: a $60,000 ORP distribution with only $36,000 SUNY/CUNY/NYSED-attributable, falsely affirmed in full, computed totalTax $1,047 versus a correctly-prorated $2,367 -- a confirmed $1,320 under-tax with ZERO runtime protection. FIX SHIPPED (option (a) from this leaf's own next_proof, not option (b) -- the engine still cannot compute a partial exclusion, so it declines rather than guesses): new required affirmation nyGovernmentPensionNotOrpOrEntirelySunyCunyNysedAttributableAffirmed (attests EITHER not-ORP-at-all OR entirely SUNY/CUNY/NYSED-attributable), new decline code ny_pension_orp_portion_unsupported, placed before the private-pension gates block. The honest path (omitting the government-pension claim for a genuinely mixed ORP) now computes over-tax-safe ($4,352) instead of the old wrong $1,047; a determined false affirmation of the new field still isn't caught -- a disclosed honor-system limit, consistent with every other attestation on this leaf.
Forms: Form 1040, NY IT-201
Surfaced by real Codex CLI's round-16 /certify-leaf finding on fed-pension-5b (2026-09-01, full record: .cert/disputes/fed-pension-5b-2026-09-01-round16.md), the companion finding to round-16's own federal-side fix (removing §457(f)/§409A NQDC as a sanctioned pensionDistributionTaxable occupant). Independently fetched NY Tax Law §612(c)(3-a) and NY DTF Advisory Opinion TSB-A-10(1)I, which concluded periodic payments from an employer-sponsored NQDC plan 'qualify as pension income' and that the W-2 box-11 amount 'should be reported to New York State as pension income.' Executed fixture: $250,000 W-2 wages ($210,000 ordinary + $40,000 qualifying NQDC in box 11) -- no IT-201:29 subtraction line at all, a confirmed $2,102 NY over-tax. FIX SHIPPED: new field nyNqdcPensionWagesAmount (a caller-asserted subset of wages, bounded: must not exceed wages) + new required affirmation nyNqdcPensionQualifiesFor20kExclusionAffirmed (attests 59½, periodic payments, nongovernmental plan -- absent/false declines ny_nqdc_pension_exclusion_unsupported; supplied without a positive amount declines as an orphan affirmation). SHARES the SAME $20,000 cap with any line-29 private pension on the same return via a single combined min($20,000, private-pension-eligible + NQDC) computation -- additive independent caps would have double-counted and under-taxed whenever both sources are present. Deliberately NOT mutually exclusive with nyPrivatePensionQualifiesFor20kExclusion (both may genuinely apply; the shared cap, not exclusivity, is the safety mechanism). Government pension (line 26, uncapped) and NQDC (line 29) are independent buckets and coexist correctly on the same return.
Forms: Form 1040, Form 5329
Claim, findings and residuals: .cert/dispositions/fed-pension-5b-early-withdrawal-plan-type-mismatch-gate.md.
Forms: Form 1040, Form 5329
Claim, findings and residuals: .cert/dispositions/fed-pension-5b-sepp-modification-recapture-gate.md.
Forms: Form 1040, Form 5329
SCOPED 2026-08-30, not yet built. Currently the engine declines EVERY plan-to-plan transfer/rollover and every §1035 exchange on a SEPP series, even when the statutory safe harbor genuinely applies — a real, disclosed, over-tax-safe residual (see fed-pension-5b-sepp-modification-recapture-gate above for the full citation and rationale). This is an ACCEPTABLE, CERTIFIABLE terminal state on its own (an over-decline with no §6694 exposure, consistent with this leaf's own already-accepted stops-payments-entirely residual) — building this leaf is a genuine improvement, not a correctness requirement, so it is `later` band, not `now`.
Forms: Form 1040, Form 5329
SHIPPED 2026-08-26 (full record: .cert/disputes/fed-pension-5b-2026-08-26.md for the original finding). REVISED SCOPE: rather than computing the 10%/25%/2.5% additional tax, split the single overloaded `pensionDistributionFullyTaxableAffirmed` into TWO atomic, independently-required affirmations — the existing field now asks ONLY about basis/QCD/rollover/Roth/NUA; a NEW `pensionDistributionNotSubjectToEarlyWithdrawalPenaltyAffirmed` field asks ONLY about age-59½-or-a-full-amount-exception. Either field absent/false → decline (fail-closed), each with its own named reason; the decline happens federally, before CA/NY see the return. Nothing new is computed: actual computation of the 10%/25%/2.5% additional tax, partial-exception apportionment, and the CA analogue remain explicitly out of scope for a future slice.
Forms: Form 1040, Form 8960
FIX SHIPPED: added `pensionDistributionIsNonqualifiedAnnuityNii` — a whole-scalar §6694 affirmation asserting the ENTIRE `pensionDistributionTaxable` amount IS (true) or is NOT (false) a nonqualified commercial annuity's §72 income, net investment income under §1411(c)(1)(A)(i) when true. Materiality gate mirrors the existing §469(g) NIIT-invariance gate EXACTLY (an exact base-invariance test — decline only when the classification would provably change the NIIT owed — not a MAGI-threshold proxy, which would over-decline the common case where other NII already saturates the MAGI excess). A new NY-side contradiction gate declines `pensionDistributionIsNonqualifiedAnnuityNii:true` combined with either NY pension-classification affirmation (both of those attest a qualified-plan-family pension, flatly inconsistent with a nonqualified annuity). DISPUTED 2026-08-27 (full record: .cert/disputes/fed-pension-5b-2026-08-27.md) — this leaf's own primary-source citations (§1411(c)(1)(A)(i)/(c)(5), Treas. Reg. §1.1411-8) independently re-fetched and confirmed correct by both passes (closes the prior code-review should-fix). But Codex found the materiality gate compares UNROUNDED bases while the actual NIIT owed is whole-dollar-rounded — at a boundary where both classifications round to the identical $0 (or any identical) NIIT, the gate still declines unnecessarily (executed fixture: single filer, $200,001 pension, decline reason literally prints '$0 vs $0'). Safe direction (over-decline, not under-tax) but a real precision bug — scoped as fed-pension-5b-niit-gate-whole-dollar-rounding below.
Forms: Form 1040, IT-201
NY IT-201-I (2025) p.10's negative list for the pension/annuity exclusion explicitly EXCLUDES "distributions you received as a result of an annuity contract purchased with your own funds from an insurance company or other financial institution" — a nonqualified commercial annuity. The engine's `pensionDistributionIsNonqualifiedAnnuityNii` field was required ONLY on the federal NIIT-materiality axis — a return below the federal NIIT threshold could affirm `nyPrivatePensionQualifiesFor20kExclusion:true` on a genuine nonqualified annuity, leave the NII field unaffirmed, and the engine granted the full $20,000 NY exclusion to income NY's own instructions say does not qualify. FIX SHIPPED: a new decline gate (`ny_private_pension_exclusion_classification_needed`) requires the NII field to be explicitly `true` or `false` whenever `nyPrivatePensionQualifiesFor20kExclusion:true` is claimed on a positive pension, decoupled from federal NIIT materiality. HONEST FRAMING: this gate forces an explicit classification, it does NOT detect a false affirmation — a caller willing to affirm `false` on a genuine annuity still gets the wrong exclusion, same as every other affirmation-gated fact on this engine. What it closes is the honest-but-unaware case: a preparer below the federal NIIT threshold who never had a reason to think about NII classification is now forced to consciously answer it. Also closed in this leaf: widened `nyPrivatePensionQualifiesFor20kExclusion`'s own attested scope to cover all THREE of IT-201-I's negative-list items (own-funds annuity — mechanically gated by this fix; QDRO nonemployee-spouse distributions and SECURE 2.0 Roth employer contributions — gated by attestation only, no engine field exists for either yet, same as every other engine-invisible fact this project gates by affirmation); confirmed line 26 (government pension) is structurally unaffected (its own doc-comment already excludes any commercial-annuity component, and IT-201-I's own negative list for line 26 is a DIFFERENT, non-overlapping list).
Forms: Form 1040, Form 8960
The NIIT-classification materiality gate (federal-profile.ts) compared UNROUNDED bases while the actual NIIT owed is whole-dollar-rounded (niit.ts) downstream — at a boundary where the bases differ slightly but round to the identical NIIT, the gate over-declined. FIX SHIPPED: exported niitOnBase(base) from niit.ts (the SAME expression computeNIIT uses) and switched the comparison to niitOnBase(baseAsExcluded) !== niitOnBase(baseAsIncluded) — no epsilon, byte-identical to the real computation by construction, so the gate cannot diverge from what actually gets taxed. INDEPENDENT DESIGN REVIEW FOUND A CONFIRMED, EXECUTED TWIN DEFECT: the sibling §469(g) NIIT-invariance gate (fed-schedule-e-disposition-469g-niit-amt-fail-closed-repair) has the IDENTICAL bug (same min(max(0,x),magiExcess) pattern, same missing rounding step) — fixed in this same leaf. This REQUIRED updating a pre-existing certified regression fixture in sec469g-disposition-release.test.ts that pinned the OLD $1-boundary behavior (a $1 raw-base gap that rounds to the identical whole-dollar NIIT now correctly computes instead of declining; the real new boundary, found by sweeping every value, is $13 computes / $14 declines).
Forms: Form 1040, IT-201
NY Form IT-201-I (2025), page 11, line 29, "Beneficiaries": a pension inherited from a decedent does NOT carry a fresh, independent $20,000 exclusion for the beneficiary — it is the DECEDENT's OWN exclusion (qualified by the DECEDENT's age, not the recipient's — a second under-tax vector the independent design review's own primary-source re-fetch surfaced beyond the original finding), net of whatever the decedent already used on the decedent's final return, ALLOCATED among all beneficiaries by inherited-value share if there is more than one. `nyPrivatePensionQualifiesFor20kExclusion`'s prior attested scope was silent on decedent/beneficiary allocation, so a beneficiary who honestly satisfied every existing clause (including their OWN 59½ test) was granted the FULL min($20,000, line 5b) regardless of the decedent's usage or the recipient's actual entitlement. FIX SHIPPED: a new whole-scalar affirmation, `nyPrivatePensionIsEntirelyRecipientsOwnNotInheritedAffirmed`, REQUIRED (explicitly true or false) whenever `nyPrivatePensionQualifiesFor20kExclusion:true` is claimed on a positive pension. `false` or omitted → DECLINE (two new codes: `ny_private_pension_beneficiary_classification_needed` when omitted, `ny_private_pension_beneficiary_allocation_unsupported` when explicitly false; the NII-classification gate still fires first when both fields are omitted) — the engine does not attempt the decedent-usage/beneficiary-allocation math itself, since it cannot see the decedent's prior usage, beneficiary count, or inherited-value fraction, and guessing a fraction risks a NEW under- or over-tax. HONEST FRAMING (consistent with every other affirmation-gated fact on this engine): this closes the honest-but-unaware case, not a dishonest one — a caller willing to falsely affirm `true` on a genuinely inherited pension still gets the wrong exclusion. SCOPE: line 29 ONLY — line 26 (government pension) is unaffected, since IT-201-I's own line-26 instruction already extends the full, uncapped exclusion to "an officer, employee, or beneficiary of an officer or employee," with no allocation or decedent-age test; no spousal carve-out exists in the primary source (a surviving spouse inheriting a pension is a beneficiary under this same rule, absent a source establishing otherwise, so declines the same as any other beneficiary).
Forms: Form 1040, Schedule CA (540)
Adds `caPensionIsUniformedServicesRetiredPayOrDodSbpAffirmed`, a whole-scalar §6694 affirmation: true attests the ENTIRE federal line-5b amount is uniformed-services retirement pay (§17132.9) or a DoD Survivor Benefit Plan annuity (§17132.10); a MIXED 5b (partly military, partly an ordinary pension) must affirm false or omit. When true, excludes min($20,000, line 5b) from CA AGI, gated on federal AGI (a HARD CLIFF confirmed against the actual statutory text, not a phase-out): ≤$250,000 for married_filing_jointly, ≤$125,000 for every other filing status (this engine has no separate qualifying-surviving-spouse status, so a surviving spouse/SBP beneficiary filing single/HOH between $125,000-$250,000 AGI is statutorily entitled but computes $0 here — over-tax-safe under-grant, disclosed via the VERIFY flag, not silent). Absent/false → $0 excluded (NOT a decline — same over-tax-safe default pattern as NY's private-pension exclusion; no new CaDeclineCode was needed). DISCLOSED, NOT MODELED: more than $20,000 of combined qualifying income in one year (both retired pay and a separate SBP annuity, each with its own $20,000 statutory cap, or an MFJ return with two military retirees) is under-excluded by this single-scalar field (over-tax-safe). Federal AGI is rounded to whole dollars before the cliff compare ($124,999.50 is eligible, $125,000.50 is not), and the Tier-2 railroad VERIFY clause still fires when the military exclusion is affirmed.
Forms: Form 1040, Schedule CA (540), IT-225
Both CA (ca-compute.ts) and NY (ny-compute.ts) disclose but do not model the Tier-2 railroad retirement exclusion — a federally-taxable pension amount that is state-EXEMPT under 45 U.S.C. §231m (Schedule CA 540 line 5b col B / IT-225 S-122). Direction is over-tax-safe on both states (a railroad retiree currently pays state tax on income both states would exempt if the engine modeled it), so this is `later` priority, not urgent.
Forms: Form 1040, Schedule CA (540)
`pensionDistributionIsNonqualifiedAnnuityNii:true` (attests the entire pension is a self-funded commercial annuity) and `caPensionIsUniformedServicesRetiredPayOrDodSbpAffirmed:true` (attests the entire pension is federal-government uniformed-services retired pay) are mutually exclusive on the same pensionDistributionTaxable scalar — but CA had no equivalent to the NY sibling gate (pension_exclusion_classification_ambiguous) that already catches this exact contradiction shape on the NY side. FIX SHIPPED: a new decline code, `ca_pension_exclusion_classification_ambiguous`, fires when both fields are explicitly true, placed post-federal-recompute (matching the NY sibling gate's relative position) and NOT conditioned on pensionTaxable > 0 (the contradiction is a fact about the affirmations themselves, confirmed correct by an executed test showing it still declines even at a $0 pension, and above the exclusion's AGI cliff). Hard decline, not a suppressed-to-$0 exclusion — unlike an omitted CA affirmation (over-tax-safe default, not a defect), a contradiction means one of two PRESENT §6694 attestations is knowingly false, and pensionDistributionIsNonqualifiedAnnuityNii also drives the federal §1411 NIIT line, so silently suppressing only the CA piece would still return a federal number computed on a known-suspect attestation. Deliberately does NOT require NII classification unconditionally whenever the CA exclusion is claimed (unlike NY's own unconditional-require pattern) — confirmed via test that NII:true + military:undefined still computes normally, since NII status is only factually incompatible with §17132.9, not an element of its own eligibility test the way it is for NY's line-29.
Forms: Form 1040, IT-201, Schedule CA (540)
Prose and disclosure fixes only; the findings and executed fixtures are in .cert/disputes/fed-pension-5b-2026-08-27-round4.md
Forms: Form 1040, IT-201, Schedule CA (540)
A CA return with a pension or IRA distribution declines unless the caller attests no R&TC §17085 additional tax and no IRA-to-HSA funding distribution; the engine computes neither the federal nor the CA additional tax.
Forms: Form 1040, Schedule CA (540), IT-201
The findings and fixes are in .cert/disputes/fed-pension-5b-2026-08-27-round5.md. Deliberately does NOT include F3 (NY railroad disclosure), F4 (NY line-26 prose), F5 (NY line-29 periodic-only prose, an UNDER-tax finding), or F6 (IRA line-5b-vs-4b presentation) — scoped out per founder decision (fix the 3 high-severity items, stop chasing full round-5 certification); those remain as future work, see the round-5 dispute doc's 'Recommended next leaves' 4-7.
Forms: Form 1040, Form 5329
Certified. IRC §72(t)(2)(A)(vii) grants the IRS-levy exception for distributions from qualified retirement plans only; §72(q)(2) — the parallel exception list governing NONQUALIFIED annuity contracts — has NO levy exception at all, and its complete exception list (fetched from uscode.house.gov: age 59½+, contract-HOLDER's death, disability, substantially-equal-periodic-payments, a §72(e)(5)(D) plan, pre-8/14/1982 investment, a §130(d) funding asset, an immediate annuity contract, or an employer-held post-termination plan) is materially SHORTER than §72(t)(2)'s — no levy, education, homebuyer, birth/adoption, reservist, QDRO, medical, disaster, domestic-abuse, emergency-expense, terminal-illness, or qualified-HSA-funding exception exists under §72(q)(2).
Forms: Form 1040, IT-201
Two pieces, both over-tax direction (safe, not a §6694 exposure): (1) PROSE — nyPrivatePensionQualifiesFor20kExclusion's doc-comment (ny-compute.ts), its OpenAPI description, and the tax-capabilities NY prose block now explicitly name "NOT a government-employer pension including U.S. military retired pay" in the line-29 negative list, alongside the pre-existing nonqualified-annuity/QDRO/SECURE-2.0 items. (2) A VERIFY flag, NY-PENSION-LINE-29-CLASSIFICATION, fires whenever the line-29 exclusion is actually applied, naming the government/military misclassification risk and pointing the caller at line 26's uncapped exclusion instead — the closest available guard given the engine cannot distinguish the two from the affirmations alone. The mechanical gate that retires this finding CLASS (not just this instance) is scoped as its own leaf: state-schema-cross-jurisdiction-exclusivity, below.
Forms: Form 1040, IT-201, Schedule CA (540)
test/state-schema-cross-jurisdiction-exclusivity.test.ts derives CA-only and NY-only field sets directly from CaStateProfileSchema/NyStateProfileSchema's own shapes (never a hand-maintained field list, so it stays true as fields are added to either schema) and asserts, for EVERY field on each side, that the OTHER state's .strict() schema rejects it as an unrecognized key — 2 structural assertions plus one generated per one-sided field, plus a derivation-semantics regression test (43 at HEAD 2026-09-05; it was 39 when written 2026-08-29, and the growth is the mechanism WORKING — three NY-only fields added by PRs #150/#153 were picked up automatically with zero edits to the test file, independently confirmed by both certification passes via git reconstruction). A frozen assertion COUNT is deliberately not restated here: on a leaf whose whole value is auto-derivation, a hardcoded integer in prose goes stale on the next schema field. Deliberately does NOT attempt to automate a full semantic pairwise-contradiction sweep (which pairs of same-schema fields SHOULD be mutually exclusive is a judgment call this mechanical schema-diff can't derive on its own) — the design review performed that sweep by hand and found every existing same-schema contradiction already gated (pension_exclusion_classification_ambiguous for govt×private and nii×govt/private; ca_pension_exclusion_classification_ambiguous for nii×caMilitary), each with its own pre-existing regression test elsewhere in the suite.
Forms: IT-201, Schedule CA (540)
Three parts (all three required — parts 1+2 alone do not catch this defect's actual shape): (1) agiRole (src/lib/practice/tax/state-conformity.ts — NOT state/state-conformity.ts; path corrected 2026-09-05) — a REQUIRED (not optional) field on every CaReturnLine/NyReturnLine literal, enforced by tsc, so a new line cannot be added without classifying it; (2) a runtime footing assertion (test/state-lines-footing-meta-test.test.ts) — exactly one agi_base, exactly one agi_total, no negative agi_addition/agi_subtraction amount, and base + Σadditions − Σsubtractions === total, checked against a baseline + one-operand-on + all-on fixture matrix for BOTH CA and NY; (3) a source-parsed operand-coverage assertion — parts 1+2 alone would NOT have caught this defect's actual shape (an operand added to the stateAGI expression with NO row at all, invisible to any per-line check), so this part extracts the operand identifiers directly from each file's `const stateAGI = ...;` statement and asserts every one is registered against a NAMED fixture that's independently verified to actually exercise it. The missing row itself: NY's IT-201:21+23 (a composite lineId — 414(h) goes on line 21, IRC-125 on line 23, but nyWageAdditions is one caller-supplied scalar summing both; verified against NY DTF, not guessed), agiRole: agi_addition, emitted only when positive.
Forms: Form 1040
CERTIFIED 2026-08-29. DISPUTED via /certify-leaf 2026-08-26 (full record: .cert/disputes/fed-social-security-86-2026-08-26.md). The §86 worksheet itself (0/50/85% tiers, all boundary conditions) is independently confirmed CORRECT by both passes. CA + NY full exclusion (derived from computed line 6b) is confirmed correct, including CA's federal-AGI-keyed exemption-credit phase-out correctly staying SS-inclusive. The NIIT and Additional-Medicare-Tax exclusion claims for Social Security benefits are BOTH independently reconfirmed correct. KNOWN COMPUTE GAP (real, OVER-tax only): `form-1040.ts` floored the pre-SS "other income" operand at $0 before it fed the §86 provisional-income calculation, but Pub 915 Worksheet 1 line 3 includes the capital-gain/loss line and is NOT floored — a net capital loss (up to the §1211(b) $3,000 limit) is a legitimate negative contribution the floor incorrectly suppressed, overstating taxable Social Security. FIX SHIPPED AND RE-CERTIFIED 2026-08-29: removed the floor entirely. The same floor in the AMT (Form 6251 line 1) was removed later in amt.ts.
Forms: Form 1040
FIX SHIPPED: removed the `Math.max(0, ...)` floor entirely (no floor added anywhere else — the design review confirmed `taxableSocialSecurity`'s own `provisional <= base → return 0` branch already correctly handles a negative provisional income for every filing status, including the $0 MFS-lived-with-spouse base, so a post-sum floor would be outcome-inert dead code). Also corrected the stale BREADCRUMB comment in `social-security-86.ts`, which claimed provisional income would only be overstated "when any Schedule 1 Part II adjustment is ever modeled" — the §164(f) half-SE deduction is ALREADY modeled and already correctly subtracted at the `form-1040.ts` call site; the comment now states this and keeps the lockstep warning for any FUTURE Part II adjustment. Tests and the review rounds: see evidence_refs and the dispute record.
Forms: Sch C
Shipped and certified only for the TY2025 federal admission of a positive caller-affirmed final Schedule C line-31 net for one sole proprietor: scheduleCIncome enters Schedule 1 line 3 and federal AGI as ordinary business income when scheduleCIsSingleSolePropFinalNetAffirmed:true and all existing whole-return gates pass. The separately shipped fed-se-tax-sch-se row owns Schedule SE arithmetic, the Section 164(f) half-SE deduction, OASDI wage-base and W-2 box-3 coordination, and Additional Medicare coordination; none receives duplicate capability credit here. Excluded from this row: Section 199A/QBI; CA, NY, or any state conformity; Schedule C losses; raw receipts, expenses, cost-of-goods-sold, or deduction derivation; multiple Schedule Cs; spouse allocation; depreciation, Section 179, or bonus depreciation; self-employed health insurance; home office; farm, clergy, statutory-employee, church-employee, or Section 1402 optional-method cases; broader Schedule C completeness; filing, extraction, live, and deployment claims.
Forms: Sch E
Shipped and certified for the exact 2025 positive final Schedule E line-26 passive-rental net only. Federal compute requires scheduleERentalIsPassiveFinalNet:true and scheduleERentalNotSection162Business:true and flows the net to AGI and the §1411 NIIT base. A 2025 NY full-year-resident direct-rental net conforms only with nyScheduleENoStateBusinessAdjustmentsAffirmed:true. Not certified here: rental losses, dispositions, depreciation or cost-seg, bonus depreciation, §179, QIP, §162/QBI rentals, K-1 income, royalties, California, or NY nonresident/part-year allocation.
Forms: Sch E, Form 8582
Shipped and certified only for the TY2025 federal §469(i) active-participation rental-real-estate loss allowance at exact behavior revision afa6c88; immutable authenticated-live r278 verifies the unchanged narrow behavior, not deployment of afa6c88. Requires all six affirmations: active participation, no prior suspended PAL, §465 at-risk, not a real-estate professional, passive final net, and not a §162 business, plus all shared federal-profile admission gates. The current-year allowed loss is capped at $25,000 through $100,000 modified AGI, phases out at 50% to zero at $150,000, and uses pre-loss AGI with the §164(f) half-SE addback for an otherwise admitted Schedule C combination. Only the allowed loss reduces AGI; excess is advisory Form 8582 carryforward text, not a durable carryforward ledger or filed Form 8582. Fails closed for MFS, Social Security interactions, real-estate-professional treatment, missing or false affirmations, and unsupported shared conditions. Not certified: other tax years, full Form 8582 or §465 mechanics, prior/future PAL ledgers, K-1 or non-rental passive activity, §469(g) dispositions, depreciation or cost-seg, §162/§199A rentals, broader Schedule E completeness, or state rental-loss compute.
Forms: Sch E, Form 8582, Form 6198, Form 6251
A SEPARATE mechanism from the §469(i) $25,000 allowance above — when a taxpayer disposes of their ENTIRE interest in a passive activity in a FULLY TAXABLE cash sale to an UNRELATED PARTY, the current-year loss PLUS ALL PRIOR-YEAR §469-suspended losses from that SAME activity are fully deducted, UNCAPPED (no $25k limit, no MAGI phase-out). Routed by asserting ALL SIX disposition-specific §6694 affirmations (entire interest disposed / fully-taxable cash sale — excludes installment §453 / like-kind §1031 / gift-or-death — / unrelated party / disposition-specific at-risk / FIVE facts about the disposition's own character: held more than one year, the year-level §1231 netting across all §1231 property is a net GAIN with no unrecaptured 5-year §1231(c) lookback, only the disposition's own final residual long-term-capital amount after that determination is reflected in netLongTermCapGain (not omitted, no more generic short-term-field routing), any unrecaptured §1250 portion is included WITHIN netLongTermCapGain AND SEPARATELY declared via sec1250UnrecapturedGain (declaring it only within netLongTermCapGain without the separate declaration under-taxes it at 0/15/20 instead of 25%), and NO Form 4797/§1231 ordinary gain or loss of any kind exists (recapture, a YEAR-LEVEL net §1231 loss, or lookback-recharacterized gain — a disposition-level loss within a net-gain year is itself a long-term capital loss, not ordinary) — the engine has no §1231-netting algorithm, no 5-year-lookback state, and no field for ordinary Form 4797 amounts, NOT wages / no AMT-side difference from the regular-tax figures for this activity), plus the shared not-real-estate-professional and not-§162-business affirmations. Fail-closed fences each decline with a NAMED reason: partial/grouped disposition, installment sale or like-kind exchange, related party, insufficient §465 at-risk, married-filing-separately (scoped out this slice), an unaffirmed/unrepresentable disposition gain (incl. a short holding period, a YEAR-LEVEL net §1231 loss, unsupported ordinary recapture, or a nonzero 5-year lookback), and AMT exposure — a same-track computed check at the pre-/post-release configurations declines as a belt-and-suspenders backstop, but the PRIMARY AMT safety mechanism is the required no-AMT-difference attestation (a computed check alone cannot rule out a cross-track AMT divergence Form 6251's own PAL/at-risk worksheet would reveal). NIIT: rather than unconditionally floor the released loss for the §1411 NII base, an exact base-invariance test proves whether the un-modeled Form 8960 same-activity NII deduction (bounded $0 to the full released amount) could change the NIIT owed on THIS return — computes when provably immaterial, declines (naming Form 8960) otherwise; this is a real, correctly-firing decline on some real-world high-NII returns, not a defect. Independent END code review complete; the identified blocker was resolved on the shipped PR #47 slice. FULL ✅ — CERTIFIED 2026-08-26 via /certify-leaf. The two remaining wrong-number paths (misplaced ordinary §1245/§1250 recapture; an omitted separate §1250 declaration) are the SAME accepted caller-attestation trust-model limitation as every sibling affirmation on this route — no mechanical backstop is possible without modeling data this engine's schema doesn't carry — not a defect. Full record: .cert/receipts/fed-schedule-e-disposition-469g-2026-08-26.json.
Forms: Sch E, Form 8582
Forms: Sch K-1, Form 1041
Slice 1 of K-1 passthrough: boxes 1, 2a, 2b, 3 and 4a only; every other box must be attested blank or the return declines.
Forms: Sch K-1, Form 1065, Form 1041
Forms: CA 540
wages + clean-conforming non-wage income (interest/div/cap-gain, affirmation-gated) — independently verified 2026-07-02
Forms: NY IT-201
wages + clean-conforming non-wage income (requires nyWageAdditions + affirmation) — independently verified
Forms: CA 540
NH 2020-2024, TN 2020 and WA 2022-2026 need a not-owed attestation on every return. Business and payroll taxes and the filing requirement are not evaluated.
Federal AGI plus tax-exempt interest, less Social Security and attested qualified retirement income, less the exemption (none above the AGI cliff), at 4.95%. Credits and other subtractions are not modeled.
Federal AGI plus tax-exempt interest, less taxable Social Security, the child deduction and the standard deduction, at the year's flat rate. Itemizing, other deductions and credits are not modeled. In 2026 a child deduction whose federal credit under NC's 2023 Code is $0 declines.
Federal AGI plus tax-exempt interest, less taxable Social Security and the exemption allowance, at the year's flat rate. Retirement deductions beyond state-mi-retirement-standard-deduction decline; other subtractions and credits are not modeled.
Key person born 1946 or later, 2020-2025, with miRetirementFactsAffirmed and birth years: the larger of the Tier 2 or Tier 3 Michigan Standard Deduction and Form 4884 Section D. Born before 1946, 2026, survivor elections, military and railroad decline.
needs research
POST /api/mcp/intake/read: one 1099-B CSV layout, read whole or refused; lots as printed, not routed to compute.
Planned: one document per call in, untrusted envelopes out, nothing kept
Planned after the deterministic formats; pinned model
Intake's dedup is built, advisory, not served
IBKR, Coinbase (totals, no lots) and 1098 parsers built, not served; K-1 not started
POST /api/mcp/tax-compute/what-if, live since July 2026. Building scenarios from a client's history, comparing them and multi-year projection are the caller's.
Forms: Form 1040
POST /api/mcp/tax-compute/draft-pdf. Stops before e-file; signing and filing are the caller's.
Forms: Form 1040
POST /api/mcp/tax-compute/return-lines, and the MCP tool tax_compute_return_lines. Lines with no engine input are assumed_zero.
Forms: Form 8879
POST /api/mcp/tax-compute/efile-readiness, and the MCP tool tax_compute_efile_readiness (no PINs). PINs and prior-year AGI are checked for format only, never returned.
Forms: Form 1040-X
POST /api/mcp/tax-compute/amended-return. The caller sends the original return as filed each call.
Forms: Form 1040-ES
The estimatedTaxNextYear block on /federal, 2026 returns.
independently verified 2026-07-01 40 refs, 0 dangling; ComputeResult + DeclineCode enum live
telemetry event dictionary in progress (S3 of this arc)
Published and binding since 2026-07-01 (Version 1.0). California governing law, tax-mcp.com as contracting entity, no E&O/insurance represented, customer is preparer-of-record, 12-month-fees liability cap ($100 nominal for free/eval) — every fact independently verified against the live page.
Forms: Sch D, Form 8949
present in code (home-sale-exclusion.ts) but referenced only by a tax-model fixture extractor — NOT on the MCP compute path, so it is NOT shipped.
The senior deduction ships separately (fed-obbba-senior-deduction). The tips, overtime and car-loan interest deductions are not computed; the federal endpoint's OBBBA flag names them. For Form 1116: the car-loan deduction the flag names is reported on Form 1116 LINE 4B -- named expressly in the 2025 Instructions for Form 1116 under 'Line 3b' ('The qualified passenger vehicle loan interest deduction reported on Schedule 1-A (Form 1040), line 30, is reported on line 4b'), a line the Form 1116 safe-harbor hardcodes to ZERO. The same instruction expressly EXCLUDES the enhanced senior deduction (Schedule 1-A line 37) from line 3b, and gives line 3b's own example as Schedule 1 (NOT 1-A) Part II adjustments. OPEN QUESTION for a later round, do not assume either way: the instructions address only those two Schedule 1-A items, so where the OBBBA TIPS and OVERTIME deductions belong on Form 1116 is unresolved by primary source.