Where every number comes from

NO CODE CHANGE · note refreshed in data/audit-fixes-6. The engine's flat 22% models the NON-cadre blended employee rate (~22%); but this app's target persona is a Financial Analyst / FP&A professional, who in France is almost always CADRE · and the cadre blended employee rate (AGIRC-ARRCO tranche-2 + APEC + the usual CSG/CRDS) is commonly cited around ~25%. So for our typical user the 22% likely UNDERSTATES social contributions (and overstates net) by roughly the 5% shown: swapping 25% at the 3 benchmark points shifts modelled net by −4.7% to −5.9% (see 'France parigi · 22% flat SS vs ~25% cadre blended' in audit/validation-results.json). This does NOT affect the income-tax barème validation (which isolates the tax computation from a given taxable base, and is 0.000%). Left UNCHANGED and FLAGGED pending a decision on whether to model cadre (~25%) as the default given the finance-professional persona, vs keeping the more conservative general-population 22% · a product/positioning call, not a data-correctness one. Deliberately not auto-corrected.

IMPROVED but STILL FLAGGED (data/audit-fixes-6). ATTEMPT: added a standard-deduction model · the missing Berufsauslagen + Versicherungsprämien the audit brief identified · using real ZH single-person values (Berufsauslagenpauschale 3% of salary, min CHF2,000/max CHF4,000; Versicherungsprämienabzug CHF2,600 cantonal / CHF1,800 federal). RESULT vs kantonsvergleich.ch 2026 (single): 60k engine 4,658 vs ref 4,956 = −6.01%; 80k 8,433 vs 8,467 = −0.41%; 120k 17,373 vs 17,407 = −0.20%. Two of three points now reconcile to <0.5% (including the persona's ~CHF108k range → net now accurate); max |dev| dropped from 13.7% to 6.0%. WHY NOT WITHIN 1.5% AT 60k: the pre-fix overstatement was NON-monotonic (+7.2/+13.7/+9.3, worst at 80k, not at the extremes) · a signature no deduction can fit, because at near-equal marginals across the 3 points the 60k point requires a SMALL effective deduction while 80k requires a LARGE one, which is impossible with any coherent forfait+insurance model (confirms the prior 'a flat constant can't reconcile all 3' finding). The residual is therefore a bracket-CURVE divergence, not missing deductions; closing 60k would require reshaping the simplified cantonal thresholds/rates (full re-parameterization, out of scope). Per the brief, a real deduction model was added (strict improvement, real Swiss values · not a forced constant) but the entry stays FLAGGED with the attempt documented. The interactive official calculators (ESTV swisstaxcalculator, zh.ch Steuerrechner, ge.ch calculette) are JS-rendered and could not be driven programmatically in this environment (no browser tool available) · OUTPUT was instead cross-checked against kantonsvergleich.ch, which states it sources 'FTA tax tariffs 2026' with standard Swiss deductions (Berufsauslagenabzug, Versicherungsprämienabzug) included. The engine's simplified brackets apply NO such deductions, which is the primary source of the gap below. Per the audit brief, the simplified cantonal/federal brackets themselves were NOT re-parameterized (out of scope) · see audit/validation-results.json for the full 6-point comparison.

Validated OUTPUT at CHF 60,000/80,000/120,000 gross (single, no confession, no children): engine total_tax = 7,700/13,216/26,739 vs reference 8,025/13,265/25,527 → −4.1%/−0.4%/+4.8% (2 of 3 points >1.5%, but NOT systematically biased · crosses zero, unlike Zurich). audit-fixes-6: the ZH standard-deduction fix was evaluated for Geneva and DELIBERATELY NOT APPLIED · Geneva already UNDERstates at 60k (−4.1%), so adding deductions (which only reduce tax) would worsen it; the zero-crossing profile is a bracket-CURVE divergence, not missing deductions. Brackets left unchanged, TAX-0063 stays FLAGGED. The interactive official calculators (ESTV swisstaxcalculator, zh.ch Steuerrechner, ge.ch calculette) are JS-rendered and could not be driven programmatically in this environment (no browser tool available) · OUTPUT was instead cross-checked against kantonsvergleich.ch, which states it sources 'FTA tax tariffs 2026' with standard Swiss deductions (Berufsauslagenabzug, Versicherungsprämienabzug) included. The engine's simplified brackets apply NO such deductions, which is the primary source of the gap below. Per the audit brief, the simplified cantonal/federal brackets themselves were NOT re-parameterized (out of scope) · see audit/validation-results.json for the full 6-point comparison.

1 item declared, not modelled

1 item declared, not modelled

1 item declared, not modelled

1 item declared, not modelled