CourtListener / RECAP search results land directly onto disk in raw JSON format before parsing. A cryptographic SHA-256 digest is generated immediately, establishing an immutable provenance trail for enterprise evaluation.

Every row below represents an actual network landing from the Free Law Project CourtListener RECAP API. Checksums can be validated against the local snapshot JSON files on disk.

Runs the same fetch → snapshot → normalize → upsert pipeline as the CLI ingestion scripts, on demand. Pick a source, adjust the search term if you like, and pull up to 50 more real records — they land in the same store the Docket Explorer reads from and appear immediately, no mock data.

These JSON endpoints power this UI and can be queried directly via curl or integrated into external ETL pipelines:

# 1. Search patent cases in D. Delaware
curl "http://localhost:4100/api/v1/cases/search?ipType=PATENT&court=ded&limit=5"

# 2. Free-text search by party or patent number
curl "http://localhost:4100/api/v1/cases/search?q=patent&limit=10"

# 3. Retrieve immutable raw audit snapshots with SHA-256 hashes
curl "http://localhost:4100/api/v1/snapshots"

# 4. Pipeline volume statistics by IP litigation classification
curl "http://localhost:4100/api/v1/stats"

Every case below was auto-ingested and auto-classified successfully, but landed here because at least one field genuinely cannot be filled in responsibly by code alone: case outcome (who prevailed requires reading the actual order or judgment, not just docket dates), or a patent-in-suit number that shallow caption/snippet regex couldn't find. These are examples of tasks where computational extraction can be supplemented by analyst review before the structured record is finalized. A review submitted here is written back to the structured record with a “Human-Verified” badge, visible in the Docket Explorer.

Same Jaro-Winkler-plus-review pattern Chrisian runs for Harris County party matching, retargeted at name variants across this corpus — e.g. a law firm whose name is punctuated three different ways across filings. A score above 0.90 is a candidate, never an automatic merge.

Live full-text search against CourtListener's Federal Circuit (cafc) docket index, using the most distinctive party name from each terminated patent case. A hit means that party is mentioned in a CAFC docket — not that it's confirmed to be this case's appeal. Full-text search surfaces false positives (a different suit against the same defendant), so nothing here is written back as a link until a human confirms it.

Live lookup against USPTO's Open Data Portal Patent File Wrapper API for each patent-in-suit number (human-confirmed numbers preferred over shallow auto-extraction). Pulls invention title, filing/grant dates, application status, and the full chain-of-title from assignment records — presented as a history, not a single confident "owner," since chain-of-title disputes are a human question. Requires a USPTO ODP API key server-side.

No USPTO_ODP_API_KEY configured on the server — add one to .env to enable this.

Live lookup against USPTO's Trademark Status & Document Retrieval (TSDR) API for each extracted serial/registration number on a US trademark case — a genuinely separate credential system from the Open Data Portal key used for patent enrichment above (confirmed by testing: that key returns 401 here). Pulls the mark text, prosecution status, owner of record, and goods/services classes. Requires a USPTO TSDR API key server-side.

No USPTO_TSDR_API_KEY configured on the server — add one to .env to enable this.