[MaxGhenis]

Answers "how does adopting the rules engine affect expected penalty" as a labeled accounting bound, per state, for all 53 jurisdictions.

Data layer

• analysis/build_engine_scenario.py + app/public/engine_scenario_data.json (186 KB): per-case any_strict/any_broad flags in the exact data.json case order, plus full-precision state shares of official error dollars, for the committed cause classes (strict {17,19,20}: programming, mass change, arithmetic; broad {10,17,19,20,21,22}: + policy misapplied, computer-user, incorrect budgeting).
• Gates (tests/test_engine_scenario.py): exact 1e-9 reconciliation of shares against analysis/cause_shares.json; 2e-6 bound on shares recomputed from the rounded data.json arrays (observed max 5.7e-7, serialization scale); ordering invariant (data.json arrays reproduce exactly from the source CSV under load_cases filters); byte-deterministic double-build; SHA-256 pin in app.js matches the committed artifact.

App layer

• New always-visible "Adopting a verified rules engine" panel: per case, zero the flagged error dollars, re-anchor at official × (1 − share), re-run the Monte Carlo through the existing FY 2026 election + delay machinery (FY 2025 stays locked history — adoption flows only through the simulated FY 2026 measurement). Shows class shares, simulated FY 2026 centers on the tier band, expected FY 2028 bill baseline → broad–strict range, pWin shift, FY 2029 line.
• Delay-clause states get an honest deferral branch: adoption pulls draws below the 13.33% threshold, so billing more often starts in FY 2029 keyed to the lower FY 2026 rate, while deferral keys the first bill to a later, unsimulated year (deferral ≠ forgiveness).
• Zero-share branch (e.g. NY strict = 0%): stated as "strict bound equals the baseline".
• National takeaway: summed expected FY 2028 bills $7.67B/yr → $6.23B (broad) – $7.52B (strict) if every jurisdiction adopted.
• Everywhere labeled: accounting bound under the file's own cause coding, reviewer judgments, not a causal adoption estimate; FY 2024 cause mix assumed for FY 2026.
• Artifact loads eagerly with SHA-256 + schema verification; integrity failure surfaces in the panel (no silent fallback). ASSET_V → 20260812.

Suite: 185 passed.

🤖 Generated with Claude Code