Activation Runbook

​

Goal

​

Preconditions

• Source ingestion is current.
• Source embedding coverage is healthy.
• AI Gateway is configured and required for backend model calls.
• PM app final projections are explicitly allowed for the controlled run.
• Daily model budget guardrails are configured before broad background synthesis.

​

Activation Steps

1. Confirm live packet drift.

select
  compiler_version,
  packet_version,
  freshness_status,
  count(*) as rows,
  count(*) filter (where packet_type = 'current') as current_rows,
  count(*) filter (
    where packet_type = 'current'
      and generated_at > now() - interval '36 hours'
  ) as current_fresh_36h,
  max(generated_at) as newest_generated_at
from intelligence_packets
group by compiler_version, packet_version, freshness_status
order by newest_generated_at desc nulls last;

2. Run a dry run for one representative active project.

refresh_project_intelligence(project_id, dry_run=true)

3. Run a controlled write for one project.

refresh_project_intelligence(project_id, dry_run=false)

compiler_version = project_intelligence_synthesis_v1
packet_version = project_intelligence_synthesis_v1:current

4. Verify the project page reads the new packet.

5. Activate the portfolio backstop.

6. Rerun health checks.

npm run rag:verify:source-lifecycle
npm run rag:verify:project-intelligence-live-paths
npm run docs:verify:project-intelligence

​

Stop Conditions

• The dry run reports a model/provider failure.
• The write creates a second current packet instead of updating the one current row.
• The compiler version remains legacy after the write.
• Source citations are fabricated or missing.
• The frontend renders stale or unrelated packet content.

​

Completion Criteria