Every Criterica Intelligence outcome model draws from a single normalized master index — the corpus of court and regulatory records that gets built once, cleaned once, and reused across every model rather than assembled fresh for each project. Pass 44 is the latest completed integration cycle against that index: a defined batch of new source records ingested, normalized to the shared schema, deduplicated against everything already in the corpus, and validated before anything trained on it is allowed to move toward production.
The integration discipline matters more than the intake volume. A record does not count toward the corpus until it has passed schema normalization — mapping the source's native format into the fields every downstream model expects — and deduplication against the existing index, so that records already captured through an earlier pass or a different source do not get counted twice and do not silently inflate a model's apparent training population. Passes are numbered precisely so that any model's training data can be traced to the exact integration cycle that produced it.
Pass 44 extended coverage across the corpus's existing federal, state, and international scope rather than opening a new category of source. The corpus is deliberately built breadth-first and depth-second: growing coverage across markets Criterica already serves before adding entirely new jurisdictions, on the view that depth inside a served market compounds a model's value faster than shallow coverage spread across new ones.
Data engineering at this scale is a continuous pipeline, not a series of discrete projects. Pass 45 is already underway, extending the same integration discipline to additional municipal and administrative data sources — the tier of records that sits below state and federal courts but still generates enforceable, verifiable outcomes relevant to the markets Criterica Intelligence serves. Municipal and administrative sources carry their own normalization challenges, since format and disclosure practice vary far more at that level than in federal or state court systems, and Pass 45's scope reflects that additional engineering work.
None of this corpus work is visible directly to an institutional buyer evaluating Criterica Intelligence — what is visible is model performance, and model performance is downstream of exactly this kind of unglamorous, disciplined data engineering. A model is only as good as the population of resolved matters it was trained and validated against, and that population is only as trustworthy as the integration process that built it. Pass 44, like every pass before it, is documented, timestamped, and traceable, so any question about what a given model actually saw during training has a specific, checkable answer.