Benchmark report
Task verified
4m 9s elapsed
The saved attempt met the documented success criterion.
100/ 100
Agent-Ready- SInstall
- SAuth
- SExecute
- SDocs
- SCurrent
Sifting the 173KB query-grammar reference markdown (sed/grep over sections) to confirm industry, locations.any and company_size field syntax before writing the query.
Where the time went
4:09 agent time- Research0:23
- Attempt2:03
- Verify1:43
Judge reason
Saved code /workspace/home/clay-task/company_search.py authenticates with the documented clay-api-key header against https://api.clay.com/public/v0 and uses the documented query-mode flow: POST /search/query-mode with server-side filters 'industry = "Software Development" and locations.any(is_headquarters = true and country_name = "Germany") and company_size = "201-500"', then POST /search/query-mode/{search_id}/run with limit 10. run_output.txt records HTTP 200 for both calls; create_response.json holds search_id search_0to98tkvx9FNFTtiz5s / source_type companies; run_response.json has the documented shape (data, has_more, source_type, period_quota) with exactly 10 company records. Independent re-parse of run_response.json shows every row industry=Software Development, size=201-500, country=Germany, matching the printed names/domains/locations (Gorillas, Parloa, OneFootball, Adjust, exocad, NiCE Cognigy, Maxon, MOIA, JustRelate, CODESYS Group) exactly; filters were server-side in the query string, not post-filtered. The API reports no total, only has_more=true, which the output states. The field names and the 201-500 enum are confirmed in reference_text.md fetched live from GET /search/query-mode/reference (HTTP 200). No saved file or transcript output contains the key; the agent's own leak check used grep -rlF on the private key file and printed only 'no key in any saved file'.
Attempt record
- 0:48Answered from saved key: Clay Public API keyno wait
- —First success verified
Attempt artifacts saved.