How the Support Model was built

Accord’s reference corpus was assembled through systematic review of published New York child-support and maintenance decisions within the study period. The search was designed to identify every decision reporting sufficient information to reconstruct the relevant support calculation. Review focused on the income findings, the applicable statutory caps, the income actually used by the court, the support amounts, and the calculation basis disclosed in each opinion.

Inclusion in a particular estimate depends on data sufficiency. An opinion that reports an award without the underlying financial inputs or calculation cannot provide a reproducible numerical comparison for that estimate. Some Appellate Division decisions affirm or modify awards without disclosing those details. They may remain important legal authority, but they do not supply the numerical evidence required to reconstruct the award. Appellate decisions that provide sufficient information are assessed under the same criteria.

Reviewed decisions are coded into standardized records with supporting findings, quotations, citations and page references. Unreported facts remain unavailable; unresolved or unsuitable outcomes are withheld. Admission is assessed separately for each estimate, so a coded decision does not automatically contribute to every prediction. The resulting corpus supports examination of the relationship between the documented inputs and judicial outcomes within the available published evidence.

Comparisons consider income relative to the relevant caps, number of children, procedural posture, geography, and the assigned judge's decisions where available. Child-support comparisons also consider the payor's income share. Above-cap income is expressed relative to the applicable cap so decisions from different years can be compared.

Multiple decisions from the same action are adjusted so they do not inflate that action's influence. Missing findings remain missing; unsuitable or unresolved outcomes are withheld. Appellate records distinguish figures left standing from figures vacated or changed.

During development, the model's estimates were compared with outcomes in actual New York matrimonial decisions, withholding each tested case from the comparison pool. These development tests are not external validation, and estimates should be read alongside the statutory calculation and your own judgment.

Each result presents the comparable cases, the assumptions used, and the strength of the available evidence. When evidence is sparse, the result says so, and an estimate may be withheld.

Independent second readings check that the record was read and coded correctly. They are not independent external validation of forecasting performance. Verification against archived sources also does not establish a decision's current good-law status.

• Model ranges are empirical scenarios drawn from comparable decisions, not statistical confidence intervals.
• Probability figures are smoothed comparable frequencies, without demonstrated calibration to individual courts.
• Published decisions are a selected sample of all outcomes.
• Historical award diagnostics have reconciliation limits.
• Software regression tests check the calculations work as designed; they are separate from prediction accuracy.

Use the estimates as support for professional judgment and negotiation: to test a proposed position, prepare for mediation, and explain the basis for a number. They are not predictions of what a court will order and not legal advice.