Lists: repos[] and users[] inputs (up to 500 targets per run, one start fee). New mode user-search. Rows are saved in batches and enrichment calls run in parallel, so big runs are faster. 60-second request timeout with retries, short rate-limit waits, honest PARTIAL outcome with resume.startPage. Code search without a server token now ends CONFIG_ERROR with a clear message and no charge.
Success-rate remediation: missing required inputs now emit a schema-safe input-error diagnostic row and finish with a warning status instead of marking the run FAILED. This protects public run health while still telling API/MCP users exactly what to fix.
Reliability hardening: 404/422/451/409 now emit a graceful _status + _note record instead of failing the run; canary check at run start aborts cleanly if GitHub API is down; preflight validation surfaces missing inputs before any HTTP call; modeCodeSearch returns a search-error record on 422 syntax issues; modeReadme/modeCommits/modeUser/modeContributors/modeReleases all soft-handle 404s.
Initial release. 11 modes: repo, repo-search, issues, prs, code-search, contributors, releases, readme, commits, user, trending. REST v3 + GraphQL v4. PAT-aware rate limit handling with Retry-After. Premium PPE pricing ($0.005/result, $0.01/deep-result).