Five datasets: one extracted from the official form itself, one published chart, three built jurisdiction by jurisdiction

Every number in a finding traces back to one of these files, or to a script in the model that reads one of them.

The form’s own calculation logic, extracted from the blank official worksheet

data/extracted/cjd304-xfa.xml is the XFA template of the 2025 CJ-D 304, as published on mass.gov, with its embedded calculation scripts intact. It is not a transcription or a summary; it is the actual field definitions and JavaScript the Trial Court’s own form runs when a field changes, extracted by model/inspect_worksheet.py. Verification status: primary source, executed directly. The model runs these scripts through a harness on six scenarios and diffs the result against the Python reimplementation; the two agree on the final order in all six.

The published support chart used to validate the schedule

data/extracted/guidelines-chart.json holds 1,104 combined-available-income-to-weekly-support pairs, covering $0 to $8,654 a week ($450,008 a year), read from the Commonwealth’s own published 2025 Child Support Guidelines Chart. Verification status: every one of the 1,104 rows was checked against the model’s Table A, with no disagreement greater than one dollar. The check itself is model/test_guidelines.py, 20 of whose checks are this comparison.

Three fifty-one-jurisdiction datasets, one fact pattern held constant throughout

All three datasets use the same fact pattern: a payor at $201,000 a year, the other parent at $570 a week, three children, no child care. That fact pattern is the author’s own order, disclosed on the about page. Comparisons drawn from these files are about that one fact pattern; they are not a claim about jurisdictions generally.

• data/fifty-state/tier-50-2026-09-05.json: the monthly order in each of 50 jurisdictions at that fact pattern, under two custody scenarios: s1 (equal parenting time) and s2 (the other parent has primary custody). Massachusetts reads $4,388.48 (s1) and $4,714.22 (s2). Verification status: tiered. Each row was built from that jurisdiction’s own primary documents, computed twice independently, reconciled where the two disagreed, and checked by an adversarial review pass; Massachusetts was computed blind as a control and reproduced these same two figures. Georgia is not in this file: its enacted formula produces a lower order under primary custody than under equal time, which the state’s own calculator reproduces, so it was held out rather than ranked.

• data/fifty-state/credit-at-122-2026-09-05.json: for each of 51 jurisdictions, whether a parent with the children 122 overnights a year (a third of the year, the equal-parenting fact pattern used elsewhere on this site) receives a formula credit for it. The file’s own tally: 28 yes, 23 no. Massachusetts is among the 23; its Box 2, the roughly-two-thirds-time category, is treated as the one-third case rather than as a credit against a higher default. Verification status: profile text, read from each jurisdiction’s own guidelines by one research pass per jurisdiction, not independently double-computed the way the two dollar-figure datasets were; the basis field on each row says so, and says which few rows were revised after a later adversarial check (for example South Dakota, revised after the review found an abatement provision the first pass missed).

• data/fifty-state/ceilings-2026-09-06.json: for all 51 jurisdictions, where the presumptive schedule stops. 41 jurisdictions state a combined-income ceiling; Massachusetts’s is $450,000 a year, 13th of the 41, against a median of $360,000. The other 10 use a percentage-of-obligor, Melson, or open formula with no combined ceiling and are listed, not ranked, in the same file. Verification status: 39 rows high confidence, 12 medium, each row’s confidence field stating which, and each citation field quoting the source text directly rather than paraphrasing it. No figure in this file was estimated where a document did not state one.

What is not in these files

None of the five datasets contains a fifty-state estimate of typical child care claims, a deviation rate broken out from imputed-income and default cases, or a distribution of orders across real families; those numbers do not exist here, and no finding on this site asserts them.

Check it yourself

Every figure above is read directly out of the linked file; none is retyped from memory.

• The model that reads these files
• The findings built from them
• The repository