fix(vercel): log drains_unavailable provisioning failures at WARN not ERROR #SUPERLOG

[superlog-app]

Summary

Every user whose Vercel team plan doesn't include drains (Vercel's log/OTLP drain feature is a paid-plan gate) triggers a production incident via a log.error call in the OAuth callback handler. The provisioning correctly detects and handles this as drains_unavailable — rolling back the configuration, redirecting the user to a clear error page — but then logs at ERROR, which fires Superlog incidents that operators cannot act on.

Root Cause

In apps/api/src/vercel.ts, the provisioning catch block called log.error unconditionally before computing the outcome:

log.error({ err: e, project_id: decoded.projectId }, "vercel provisioning failed");
const outcome = e instanceof VercelProvisioningError ? e.outcome : "error";

When Vercel rejects both POST /v1/drains calls with 403 "Drains are not available for team '...'", classifyDrainProvisioningFailure correctly produces outcome: "drains_unavailable". The per-signal WARN logs already record the Vercel API message. But the catch block's unconditional log.error then fires on top of them, creating the incident.

Telemetry confirmed 7 occurrences across 4 distinct Vercel teams since 2026-07-03, every one carrying outcome: "drains_unavailable"; no outcome: "error" events have ever been observed in production.

Remediation

This PR computes the outcome before logging and downgrades to log.warn when the outcome is drains_unavailable (an expected external API condition), while preserving log.error for unexpected failures.

All 19 Vercel-service unit tests pass. The 75 pre-existing test failures in the repo are unrelated to this change (confirmed by baseline before/after comparison).

Related incident: frosted-quokka (0582f35d-b982-4506-ab89-7501409d9fda)

---

Was this PR helpful? Leave feedback — goes straight to the Superlog team.

---

Summary by cubic

Downgrade logging of Vercel drains_unavailable provisioning failures from ERROR to WARN to prevent false Superlog incidents, while keeping unexpected failures as errors.

• Bug Fixes
  • Compute the provisioning outcome before logging.
  • Log warn for drains_unavailable; keep error for other outcomes.

^{Written for commit a3da0b8. Summary will update on new commits.}

[superlog-app]

Observability review

• 1 blocking

[superlog-app]

logs · blocking — Use log.info, not log.warn, for the drains_unavailable branch

Replace log.warn with log.info: drains_unavailable is expected control flow (a Vercel plan gate the operator cannot act on), and the rubric reserves warn for nothing — info is the correct level for expected, non-actionable outcomes. Using warn still risks alert rules that key on severity ≥ warn, defeating the purpose of this fix.

Suggested change

	if (outcome === "drains_unavailable") {
	log.info({ err: e, project_id: decoded.projectId }, "vercel provisioning failed");

Useful? React with 👍 / 👎.