# Consultant review protocol — synthetic practice

Purpose: test whether a reviewer can distinguish recorded receipts, interpretation and unresolved questions. This is a learning worksheet, not an operational assessment or a readiness score.

1. Choose one fictional case. Read the scenario and exact original records first. Record the IDs and the facts actually present. A source's report is not independently verified truth.
2. Inspect the feed comparison. Confirm the matching producer/service/kind and receipt window. Separate receipt evidence from event time, identity conflicts, absent feeds and absent producers. An empty unsupported input is a rejection, not an outage finding.
3. Write a labeled interpretation, one alternative explanation or uncertainty, and the next evidence question. The inventory itself may be incomplete. Neither an alert nor a recovery signal proves cause, security coverage or restored service.
4. Name a proposed role only if useful. Do not invent an accepted owner, deadline, service target, provider choice or delegated response authority. A follow-up question can remain unassigned.
5. Select “Review noted” or “Needs clarification” only if you performed that review. Optional review minutes are manually reported actual effort; leave blank if unmeasured. No timer, savings estimate or group statistic is computed.
6. Export all notes explicitly. Every field, including drafts and hidden cases, is included. Export errors preserve the fields. Confirm the download. Printing includes every case, source disclosure and current draft note; refresh/close loses in-tab notes. No automatic save/upload or synchronization.
7. A later consultant may compare reviews for unsupported causal claims, missing evidence questions, accidental owner acceptance or source confusion. Retain disagreement and corrections. A second person's agreement is not field validation or client acceptance.

No human trial has been completed yet. Do not claim review-time improvements, causal accuracy, alert reduction, incident resolution, AI performance or operational readiness from eight contract checks. These cases were authored by a separate AI-generation run given the behavioral contract, with expectations frozen at 52b13ade1 before execution. The parent knew implementation, so this is not a fully blinded or representative benchmark. Case 04 does not actually test same-record-ID isolation across producers despite its title. Existing regressions are complementary, not independent cases.

Current worksheet is synthetic-only. No real client/WWT operational content. A future client trial needs the approved client environment, handling boundary, applicable service inventory, reviewer participation and scope; neither this artifact nor its notes authorizes that trial. Keep deployment/model/automated-response decisions separate. The software remains model-free.


## Revisit saved notes

Use the matching version of this lab to open its exported review JSON in the saved-review panel. Compare its declarations with the source records and your separate current worksheet. Recorded next questions and clarification states are discussion prompts, not automatically accepted actions. A source mismatch requires the original lab version; do not edit the hashes to force acceptance. Matching hashes and commit IDs establish consistency only, never authenticated authorship or truth. Printing includes both sets, labeled separately. Export retains only current worksheet drafts; copy any useful conclusion deliberately after reviewing it. Use synthetic material only.