Submit an async intelligence job

Submit a job for async execution. Request body is { kind, input } discriminated by kind.

Use kind: stats for exact engagement totals over every record matching the query. Matches come from indexed tags, text, themes and products plus semantically similar content; the semantic matches help find the right terms but never cap the count, though semantic discovery is bounded, so a very large topic can return a floor rather than the full population. Incomplete discovery or reconciliation fails the job instead of returning sampled or partial totals. Stats jobs cost a flat $5.00 per call, expose 24h publicly, and may omit window (defaults to all).

Response shapes:

• 200 + { fromCache: true, result, charged } — identical recent job within the per-kind freshness window; cached result returned synchronously. charged is the full rate, except for a patterns/stats answer this account already paid for today, which is 0.
• 202 + { dedup: true } — identical job currently queued/running; reuses the in-flight jobId.
• 202 + { estimated_cost } — new job accepted; the credits were reserved at the estimate. Poll GET /intelligence/jobs/{id}.
• 402 — insufficient credits; body recommends how to fit your budget.

For the per-record-charged kinds, window (in input) is REQUIRED — it selects the slice of the corpus to retrieve over. It does NOT affect price. estimated_cost is an upper bound for those kinds; the final charge is computed from the records actually returned.

Input validation is strict: any undeclared field, type mismatch, or out-of-enum value is rejected with 400 (nothing is silently stripped or coerced).