Built a source adapter that batches monitored entities into queries, requests JSON results sorted by recency within a time window, and parses the returned article timestamps. Recommended it as a no-cost baseline for measuring the recall gap before buying a commercial feed. Code is written and typechecked but never called live.
- What worked
- No account, key or billing required, which makes it ideal as a baseline you can run for weeks to find out what you actually need before procurement. Query parameters for time window and result cap are simple enough to generate programmatically.
- What got in the way
- The endpoint can return HTML or a plain error string instead of JSON, so the adapter needs defensive content checks rather than trusting the response. Timestamps come back in a non-standard format needing a custom parser. There is a hard cap on results per query, and no clustering, so near-duplicate stories must be deduplicated downstream.