H.R. 3684 — Infrastructure Investment and Jobs Act (IIJA)

How this was built

lxml structural parse of the enrolled XML; Division J appropriations read from appropriations-major/intermediate/small account lead text; authorizations/contract-authority/transfers located by section via the element tree. Every quote is a <=15-word verbatim substring produced by faithful itertext() extraction and re-checked by exact substring search. Represents the 30 largest/most-illustrative items, NOT an exhaustive enumeration of the ~1,742 dollar figures >= $1M in the law. 30 of 30 sampled figures were re-located as exact strings in the source text. Source: GovInfo bulk BILLS collection (keyless); bill-stage='Enrolled-Bill'. Full bill text.

IIJA extends the H.R. 4820 mechanism taxonomy with three new mechanisms that a landmark authorization law requires: 'authorization' (there are authorized to be appropriated $X -- a ceiling on future appropriations, the single most common big-dollar phrasing here with 184 occurrences), 'contract_authority' (Highway-Trust-Fund tables, 'authorized to be appropriated out of the Highway Trust Fund', obligation-limited elsewhere), and 'direct_spending' (the mandatory $90B + $28B general-fund transfers to the HTF in Division H, 'there is hereby appropriated'). Division J alone contains ~91 account line-items carrying dollar figures and is a standard advance-appropriations act structurally identical to the H.R. 4820 PoC. All 30 extracted dollar strings verify as exact substrings of the fetched enrolled text (found_in_raw_source and quote_found_verbatim both true for every item). Revenue/offset provisions (Division H) were reviewed but yield no concrete '$X offset' line-items (they are excise-tax rate/date changes and IRC amendments scored by JCT), so revenue_or_offset has zero extracted items by design, not omission. Counts and dollar sums are over the 30 SAMPLED items only, not the whole law. The $42,450,000,000 BEAD figure appears in BOTH the 'authorization' bucket (Division F Sec 60102) and the 'appropriation' bucket (Division J NTIA) -- these are the SAME dollars and must NOT be added together. Mechanisms must never be summed across categories. Verdict: MIXED — text extraction works for the APPROPRIATIONS portion only, and mechanisms must be kept separate. IIJA is a hybrid law: Divisions A-I are an AUTHORIZATION + CONTRACT-AUTHORITY statute (surface-transportation reauthorization plus energy/water/broadband authorizations), while Division J is a genuine multi-agency ADVANCE APPROPRIATIONS act ('the following sums are appropriated, out of any money in the Treasury not otherwise appropriated, for the fiscal year ending September 30, 2022'). The headline 'new' IIJA dollars that DO exist as clean '$X for Y' lines in the operative text are almost entirely in Division J (BEAD $42.45B, DOE Office of Clean Energy Demonstrations $21.456B, Energy Efficiency & Renewable Energy $16.264B, FCC Affordable Connectivity Fund $14.2B, Army Corps Construction $11.615B, Bureau of Reclamation $8.3B), plus the Highway-Trust-Fund contract-authority tables in Divisions A-C and the $118B general-fund transfer to the HTF in Division H. THE TRAP: the enrolled text contains 184 occurrences of 'authorized to be appropriated $X' across Divisions A-G. These are CEILINGS on FUTURE appropriations, not spending, and at least one (BEAD's $42,450,000,000) is the SAME dollars Division J actually appropriates -- so summing authorizations + appropriations double-counts. A naive 'sum every dollar in the bill' would be badly wrong, exactly as the H.R. 4820 PoC warned. WHAT IS NOT IN THE TEXT: the widely-cited '$1.2 trillion total / ~$550 billion new investment' topline is a CBO/administration construct built from CBO's cost estimate and multi-year authorization+appropriation totals, NOT any single line in the enrolled text. Division H revenue pay-fors (superfund excise taxes, customs user-fee extension, employee-retention-credit termination, digital-asset broker reporting) do NOT resolve to concrete '$X offset' strings -- they are rate/date changes scored by JCT/CBO and were NOT extracted. RECOMMENDATION: for landmark laws use a MIXED source approach -- (1) extract concrete per-account spending line-items from the TEXT where they exist as '$X for Y' (Division J appropriations, Div A-C contract-authority tables, Div H HTF transfers), each tagged with its mechanism and NEVER summed across mechanisms; (2) tag 'authorized to be appropriated' figures as authorizations and disclose them SEPARATELY (like the guarantee-ceiling block in the H.R. 4820 PoC), de-duplicated against the appropriation of the same program; (3) for the site's aggregate 'cost of the law' headline, cite the CBO cost estimate for H.R. 3684 (public/keyless at cbo.gov) rather than reconstructing a total from bill text -- reconstructing it from text is not honestly possible and invites the double-count error. BOTTOM LINE: bill TEXT is the right source for citable per-line-item spending (it exists and every extracted dollar verifies as an exact substring), but the law's TOTAL should be sourced from the CBO score, not summed from the text.