v0.1.0 — Durable asynchronous backend lab

Experimental single-host public reliability lab, not production-ready or distributed-scale proof.

Separate Go API, PostgreSQL durable jobs/transactional outbox/results, RabbitMQ confirmed publication, independent workers, commit-before-ack, bounded retries/dead letters, backpressure and graceful shutdown.

Verified actual helper-process deaths at three boundaries, live broker restart, SQL rollback and duplicates. All PR and post-merge unit, adapter and durable-recovery CI jobs passed. Main CI: https://github.com/ajaykr0905/distributed-scale-validation-lab/actions/runs/37741760446 .

Twelve corrected 1/2/4/8-worker trials plus warm-up: 12,400 accepted = completed = results, 262.73–323.21 jobs/s; observed p95 95.43–151.08 ms includes polling/sampling overhead. This workload does not show linear worker scaling. Corrected raw receipts: artifacts/2026-10-08-durable-benchmark-v2.json and recovery-v2.json. Runtime hashes were not captured during historical trials; later inspection is explicitly limited in binary-inspection-v2.json. Future harnesses enforce hashes/matching clean VCS build metadata.

The 90-second artifacts/2026-10-08-recovery-demo.mp4 is a real timestamped agent-executed terminal replay, not Ajay learner/narration evidence. Its 5.119233-second broker recovery is a separate trial from the v2 receipt at 3.401676166 seconds; both include Compose health/startup overhead.

Clean-clone setup and limitations: README.md; docs/durable-backend.md; docs/benchmark-runbook.md; docs/recovery-demo.md. Local unauthenticated single-tenant API; no authorization, HA/database failover or recovery SLO claim. Existing in-memory 10,000-job demo remains a separate functional baseline.