The US Capital Map: an entity-level classification of Form 990 public charities

Entity-level classification of US 501(c)(3) public charities filing the full Form 990, on two axes: bankability (capacity to service a fixed, scheduled, priced claim) and commerciality (a proxy for extractability -- 501(c)(3)s are non-distributing, so measured extractability is approximately zero for all of them). Covers 361,537 organisations across tax years 2012-2023, from the NCCS Core Series. The 283,771 still filing a full Form 990 in 2021 or later are flagged current_filer, and they are the population every headline figure uses: organisations that stopped filing earlier score materially lower, so including them measures survivorship rather than structure. The full panel is published anyway, flagged rather than dropped, so any population defined on a different filing window can be reproduced from this file rather than taken on trust. 990-EZ and 990-PF filers are out of scope -- they file a different form with different fields. One row per organisation, 27 columns. Includes the per-tax-year reconciliation against IRS Statistics of Income Table 1, which agrees on total revenue within -3.7% (TY2021) and -1.4% (TY2022); TY2023 has no published SOI target and is reported for cross-year comparison only. The pipeline is deposited alongside the data under the MIT licence. At the default cutoff, among current filers, the bankable, non-commercial cell is the largest by entity count: 137,793 organisations, 48.6%. This is a replication of the Australian classification (doi.org/10.5281/zenodo.21749614) on the same two axes, the same scoring and the same cutoffs, so the two are directly comparable. They are SEPARATE deposits and neither contains the other's files. Not a rating. It measures financial shape, not merit, need or effectiveness. Read COLUMNS.md before use. CHANGES IN VERSION 2.0.0 - The commerciality axis has been corrected. Surplus margin was weighted 30% in commerciality and 25% in bankability, placing one quantity in both axes. Commerciality is now 100% earned share. - The panel is extended from tax years 2021-2023 to 2012-2023, and headline figures are now computed on current filers rather than on every organisation appearing anywhere in the panel. - At the default cutoff the bankable, non-commercial share moves from 48.4% to 48.6%, bankable self-funding from 25.7% to 23.6%, fragile trading from 7.3% to 10.0%, and fragile non-commercial from 18.6% to 17.8%. - The per-organisation file expands from 20 to 27 columns, adding first_year, last_year, current_filer and a recurring-revenue baseline (rev_median, rev_cv, revenue_spike, capital_event). - The code archive in version 1.0.0 carried the superseded commerciality weights and could not reproduce the data it accompanied; it has been rebuilt and now also includes robustness.py.