@uwmd/batch

A local, deterministic collection indexer for .uwx.md records. It validates every deal, records a semantic digest, and emits uwmd-collection.json plus a spreadsheet-safe CSV index.

npx @uwmd/batch deals --out batch-output

The index is a projection over canonical deal files, not a storage protocol. A host may import it into any database without changing .uwx.md semantics.

Read-only workflow projections

filterUWMDCollection(index, filters) narrows an existing index by asset class, deal stage, flag, or a named quick-metric comparison. summarizeUWMDCollection(index) returns deterministic asset-class, stage, and flag summaries. projectUnderwritingQueue(index) returns every matching candidate ordered by blocking_flags, then error count, warning count, and path. Invalid candidates stay visible in the index, summary, and unfiltered queue; none of these helpers changes a deal file.

Corpus fact table (--facts)

npx @uwmd/batch deals --out batch-output --facts

Adds uwmd-facts.jsonl — one JSON object per line, one line per JSON fact in every parseable deal — plus uwmd-facts-manifest.json with the counts and the skip list. Each row is a normative block_values row (UW CSV Bundle spec §3, produced by @uwmd/core's flattenEnvelopeBlockValues, never re-implemented here) prefixed with the deal's identity: path, deal_id, asset_class, semantic_digest, and valid. Deals that parse but fail validation are included with valid: false; files that cannot produce an envelope are listed in the manifest's deals_skipped — the fact table never silently drops a deal.

The JSONL loads directly into DuckDB (read_json('batch-output/uwmd-facts.jsonl')), Snowflake, ClickHouse, or Pandas. The durable key for a fact is (semantic_digest, block_ref, scope, pointer). See the data-lake guide for the full pipeline.