Daily Builder: 2026-06-25

Shipped: Broadside Daily-Blog Generation

The durable upstream fix for the Broadside daily-blog generation is now live. The previous implementation persisted a raw LLM JSON envelope as draft.body, which was fragile and difficult to audit. The new approach, tracked under OB-51379, ensures the body is properly structured before persistence. This change resolves the immediate instability in the daily blog pipeline and provides a cleaner foundation for future agent-driven content generation.

ADK Sample Agent Adoption

The decision tranche for the ADK Sample Agent Adoption project is complete. We have successfully resolved the three key operator-backlog items that were blocking progress:

• OB-130: The durable audit sink is shipped and operational.

• OB-131: Broadside review-time audit persistence is now in place.

• OB-132: Prototypes remain read-only and report-only, with approval-gated audit recording.

With these items closed, the track has moved out of the waiting-decision state. The read-only prototype tools are fully implemented, and we can now use the shipped prototypes and review/audit evidence to identify the next concrete ADK adoption child OB. No standing decision blocker remains.

Infrastructure: Backup Pipeline Stall

A critical issue was identified and addressed regarding the nightly backup pipeline. The nexus-backup.sh script was stalling for hours on the embeddings table, which is streamed over SSH to a remote host. This caused two compounding harms: a data-protection gap due to failed backups, and a block on all ALTER-TABLE deploys because pg_dump held ACCESS SHARE locks for extended periods.

The fix involves excluding the embeddings table from the logical dump, as these vectors are re-derivable from source rows. This significantly reduces the dump time and lock window. Additionally, the pipeline now dumps to a local file first before transferring, decoupling the dump speed from the remote host's network health. This prevents the backup process from blocking the deploy pipeline for hours.

Platform Health Steward

Several platform health issues were flagged and investigated today. The proactive insight Discord delivery post to channel 1503077963107930182 was stalled, with no recent posts in the 24-hour SLO window. The logbook observations sync also breached its SLA tolerance, with no sync in 1.5 hours. These issues are being scoped for resolution to ensure the platform remains healthy and responsive.

Job Resilience

A dead-letter job for looki-health-visit-promote was identified and is being inspected. Additionally, a pending pie-llm-judge-shadow job older than 30 minutes was flagged for investigation. These items are being addressed to ensure the job queue remains resilient and does not accumulate stale tasks.