Source and encode law
Open rule libraries capture statutes, regulations, and guidance as machine-readable logic with provenance back to source text.
Architecture
From source law to queryable answers.
The system is intentionally layered. Open rule libraries supply the substrate. Cosilico validates, calibrates, serves, and operates the runtime.
Pipeline
Encode the law. Validate the behavior. Calibrate the households. Serve the query.
Validate rule behavior
Encodings are checked with parse validation, reference integrity, test-driven examples, and comparisons against external systems where available.
Build synthetic populations
microplex and py-statmatch create calibrated, privacy-preserving household populations and fused records that can support counterfactual simulation.
Serve queries and compute
Cosilico exposes the runtime through hosted APIs for household calculation, rule lookup, microsimulation, and imputation.
Operating principles
What the system optimizes for
Deterministic answers should beat fluent guesses.
Provenance matters at the rule level, not only in docs prose.
Population simulation requires calibrated households, not only formulas.
Managed hosting should simplify operations, not turn the stack into a black box.
Visible proof
What is already publicly legible
4/4 published test cases matched
Current public validation artifact: 26 USC §1 2018 TCJA rate schedules matched expected outputs.
100% accuracy on that artifact
The published report records full agreement across the tested filing-status schedules.
$0.01 comparison precision
The validation methodology compares outputs to cent-level tolerance rather than using hand-wavy qualitative checks.