Railway Production Parameters (MVP)

This page is the minimum runbook for clawbridge command-provider deployment on Railway.

Recommended startup mode

Use one-shot command mode:

zene host --protocol v1 --bridge-compat --single-request

Recommended environment baseline

ZENE_MAX_CONCURRENCY=8
ZENE_STDIN_TIMEOUT_MS=10000
ZENE_IDEMPOTENCY_TTL_SEC=600
ZENE_IDEMPOTENCY_MAX_KEYS=50000
ZENE_IDEMPOTENCY_REPLAY_MARKER=true

Timeout budget recommendation:

• request timeout in payload: 30000
• clawbridge timeout: 40000
• keep zene_timeout_ms < clawbridge_timeout_ms

Retry budget recommendation:

• TIMEOUT: retry at most once if budget remains
• PROVIDER_DOWN: retry at most once
• PROVIDER_RATE_LIMIT: retry with short backoff once
• INVALID_REQUEST, PROVIDER_AUTH: no retry

Fault triage order

When a request fails, inspect in this order:

1. stdout contract (ok, error_code, error_message)
2. stderr logs (runtime metrics and diagnostics)
3. process exit code (0/2/3/4)

Runtime metrics to monitor

Minimum metrics emitted in stderr snapshots:

• timeout rate
• cancel success rate
• average cleanup duration in milliseconds

These are enough to detect timeout pressure, cancellation health, and cleanup regressions.