feat(signals): drop the Signals credits quota gate from scout suggestion scans #94166

Problem

The pre-computed scout suggestions runner (products/signals/backend/scout_harness/suggestions_runner.py, _gate_skip_reason) skips a project when its organization is over the Signals credits quota (is_team_signals_quota_limited). The manual refresh endpoint (scout_suggestions_api.py, refresh) throttles on the same check.

That gate was carried over from the scout scheduler, where it makes sense: a scout run exists to feed signals into the pipeline, and ingestion drops them while the limit binds. A suggestion scan is different. It is an internal run we do as part of self-driving. Its output is a stored batch the scouts tab reads, not signals, and it never bills the customer: Signals credits are charged only per report whose implementation ships a PR (SIGNALS_CREDITS_PER_REPORT_WITH_PR in products/signals/backend/billing.py). Nothing is lost by scanning an over-quota project, and the quota is the moment a project most needs a good "what to turn on next" list for when the limit lifts.

The skip also costs more than it looks. The coordinator stamps last_requested_at when it dispatches the child, and a skipped run never touches the row, so a quota-limited project is held for the full refresh_days (7 days by default) even when its billing period resets the next day. In the first tier-1 wave on 2026-09-03, 3 of 10 planned projects were skipped this way and are now held until 2026-09-10.

Proposal

Drop the Signals credits quota gate from the suggestions path. Keep the AI data processing consent gate and the acting-user gate as they are.

• suggestions_runner.py: remove the quota_limited branch from _gate_skip_reason. The SuggestionRunResult.skip_reason value quota_limited stops being produced; the $scout_suggestions_generated telemetry keeps the field for the remaining reasons.
• scout_suggestions_api.py refresh: remove the Throttled on is_team_signals_quota_limited. The daily refresh cap (SUGGESTIONS_REFRESH_DAILY_CAP) stays as the abuse guard, and the kill switch and blocklist checks stay.
• products/signals/backend/scout_harness/AGENTS.md: the arun_scout_suggestions bullet lists "signals quota" among the gates. Update it.
• Tests: add a case in products/signals/backend/test/test_scout_suggestions.py where the quota check returns limited and the scan still runs, and the matching refresh-endpoint case.

Out of scope here, worth a follow-up: a skipped run should not consume the refresh window. The cleanest shape is for a skip to clear last_requested_at (or to re-plan after failure_cooldown_hours instead of refresh_days), so the remaining skip reasons (ai_data_processing_not_approved, no_active_user) do not hold a project for a week after the condition clears.

Rollout

No flag work: the change ships dark until the next coordinator tick plans a project that would have been skipped. The rows already stamped by a quota skip need one manual nudge after deploy, either POST /api/projects/:id/signals/scout/suggestions/refresh/ or python manage.py run_scout_suggestions --team-id <id>, since the planner will otherwise revisit them only at the end of their refresh window.

Context

• Producer shipped in feat(signals): pre-compute suggested scouts per project #85963. Flag payload knobs and the planner are described in products/signals/backend/scout_harness/AGENTS.md.
• The scout scheduler's quota skip (temporal/agentic/scout_scheduler.py) and the pipeline gates in quota.py are unchanged by this issue.