Reporting

wl report writes a fixed report bundle into the project reports/ directory:

• summary.json
• cost_by_work_category.csv
• classification_traces.parquet
• summary.md
• summary.html

The report engine reads from the local DuckDB store. It can render useful output after ingest and rollup alone, but some sections only populate after classification.

Sections The Reports Generate Today

Always available when data exists:

• dataset context
• raw trace excerpt
• normalized observations
• rolled work units
• review-needed work

Available after wl classify has produced ClassificationTrace rows:

• pending review queue from policy classification
• ambiguity summaries
• compression proof point
• top material work units
• cost by policy outcome
• cost by work category
• low-trust high-cost items

Available only when --include-economics is passed:

• comparative economics

Other Output Surfaces

• wl report also renders a terminal summary
• wl export can export any known table as CSV, Parquet, or JSON
• wl explain prints a stored work unit or classification as JSON

What The Reports Make Legible

The current bundle is designed to show:

• how raw trace records were normalized into ObservationSpan
• how many observations were rolled into each WorkUnit
• which items still require review
• where blended cost is accumulating by category or policy outcome
• which costly items are still low-trust

summary.json contains the full structured summary, including totals, compression story, queue data, and cost slices. classification_traces.parquet preserves the downstream attribution layer for local analysis. summary.md and summary.html present the same narrative in human-readable form.

classification_traces.parquet is still written even if you have not classified anything yet; in that case it simply reflects the empty downstream layer.