Docs / Working as a team / Onboarding

Onboarding

Connecting your traces is quick. Getting to a failure mode you trust takes longer, and it is mostly your team's reading time.

What we need from you

• Read-only access to your traces, scoped to one project.
• Which of your traffic is production, and how to tell it apart from staging, evals and experiments.
• The people who need access.
• Permission to ingest.
• Separate permission to send excerpts to a model provider.
• How far back to import.
• Rough trace volume.

How it goes

1. DiscoveryA short call to collect the items above and to hear which failures you already suspect. Those suspicions become the first traces we look at.
2. ConnectionWe connect and read a small sample. Before anything else is worth doing, two things have to be true: your traces render as a timeline a human can actually read, and every citation opens the exact passage it claims. Nothing is judged at this stage, nothing is written back to your stack, and no paid operation runs.
3. CalibrationYour team open-codes a sample, we build the axial vocabulary together, you approve a small batch of proposals, and you choose one to three failure modes to monitor. This is the stage that takes real human time. See Open coding and Axial coding.
4. Monitoring onMonitoring is enabled for the codes you selected, and the first completed trace day produces the first rates.

How your data is handled

• We only ever read. Zoroval reads from your tracing stack and never writes back to it.
• Your data is yours. Your traces, codes and verdicts are isolated to your account.
• Only the people you nominate get in. Withdrawing access takes effect immediately.
• You can stop at any point. Withdraw access and ingestion stops; ask us to delete and the account and its traces go with it.

If you would rather work from sanitised fixtures than real traces for the early stages, say so at discovery and we will.

Before your first session

• Decide who your domain expert is, and get an hour of their time booked. This is the single biggest predictor of whether the taxonomy is any good.
• Write down the three failures you already believe you have. You will find out quickly whether they are real and how often they happen.
• Have one recent example of a bad output that annoyed somebody. It is the fastest way to start coding.

Questions, or want to start: hello@zoroval.com.