Bound and measure provider requests during generation

[Ankit6149]

Advances #27 and #34 with one shared generation execution budget.

Implemented:

• hard server ceiling of 40 provider calls per campaign execution
• worst-case planning before provider spend: strategy + initial destination + quality revision + duplicate repair
• full 12-destination campaign remains supported at a worst-case plan of 37 calls
• callers cannot raise the hard ceiling above 40
• every provider call is accounted centrally before adapter invocation
• request telemetry records provider/model, request kind, destination, max output-token ceiling, status and duration without prompts/source content
• campaign response/package expose planned vs actual request execution
• repair/revision requests are counted separately
• stable safe request-budget failure with reduce-destinations recovery guidance
• regression tests pin ceiling, preflight ordering, accounting, privacy, duration and 12-destination capacity

Still open:

• [P0][Plan 2] SF-006 — Replace destination burst with bounded generation concurrency #27 live queued/generating/revising progress in Studio
• [P0][Plan 2] SF-006 — Replace destination burst with bounded generation concurrency #27 provider-specific concurrency/long-form lane policy
• [P1][Plan 2] SF-017 — Add request, context, concurrency, and cost controls #34 token/cost estimation before spend
• [P1][Plan 2] SF-017 — Add request, context, concurrency, and cost controls #34 per-owner/session/IP request-rate and concurrent-campaign controls

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
signal-flow-studio	Ignored	Preview	Sep 25, 2026 11:20am UTC