Department · Operations & Infrastructure
The runbook should know the infrastructure changed.
Runbooks and postmortems checked against the real system, drift flagged before the shift, and context in the incident copilot.
Scenario
How context circulates when ValorBrain is in the flow.
- 01
Checking the payment gateway's failover runbook against the real system finds the documented hostname different from the current ingress in k8s.
- 02
The runbook gets marked as divergent, with the owner and the date of the last check.
- 03
The divergence becomes a review for sre/platform, not a surprise in the middle of an incident.
- 04
A similar incident retrieves the earlier postmortems, with what worked and the source of each decision.
- 05
The incident copilot starts from that history instead of re-discovering the cause.
- 06
What was learned last time shows up in the first answer this time.
- 07
Whoever comes on call receives in the session bootstrap what changed since the last shift: drift under review, a pending handoff, a new decision.
- 08
The handoff between shifts and agents is routed with state, not summarized from memory.
- 09
The shift starts from what changed, not from what the runbook forgot.
Frequently asked questions
We already have an ops wiki and written postmortems.
They stay, as sources. The difference: here documentation is checked against the real system, drift becomes an alert with an owner, and the right postmortem shows up when the similar incident does.
What exactly does a reality probe do?
It compares documentation with the real system and flags divergence for human review, logging owner and date of the check. It points at the problem; fixing it remains the team's job.
What if the earlier postmortem is wrong?
Every record has source, author, and date. Corrections become new versions, with the previous one superseded in history, and conflict between sources is flagged for review.
What proof is there that retrieval works?
Internally we measure LoCoMo R@10 of 96.58% (1986 questions, +2.1 points over the published open-source leader) and an average of 75.5% (binary 80.8%) on BEAM-100K, with an open methodology. Internal retrieval benchmarks; not peer-reviewed and not end-to-end QA. For operations, the everyday proof is drift flagged before the shift, not after.
Next step
Less starting over. More continuity
Bring a real project. In one conversation we map where context gets lost between the people and agents on your team, and where to start.