> ## Documentation Index
> Fetch the complete documentation index at: https://evalgate.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Prepare a review without granting approval

> Use typed suggestions to organize review evidence while preserving human authority, material versions, and cost boundaries.

Review preparation is assistance for the person examining an adjudication item. It is not automatic approval, automatic reference labeling, or automatic promotion.

The current source contains a **Prepare review** action and an evidence panel. Availability depends on the deployed version and operator configuration; an unconfigured environment returns a blocked preparation rather than a fabricated suggestion.

## What happens when you prepare an item

Open the applicable review item and select **Prepare review**. The request carries the material revision you saw and an idempotency key. The server records intent, checks the current material, and decides whether execution is configured and permitted.

The implementation uses a server-controlled TypeSafe model for this assistance purpose. The actual preparation service supplies the platform environment credential and records the payer as the platform; it does not silently turn this assistance action into customer BYOK execution.

The service checks tenant egress policy and obtains a shared budget reservation before attempting the provider call. If configuration, policy, or budget prevents the call, required review remains open.

## Inspect the result and its evidence

The preparation panel distinguishes pending, completed, blocked/failed, and stale material. Inspect the provider, model, payer, source revision, input digest, recorded outcome, and reserved liability.

If the source changed while preparation ran, the returned result is marked stale and its suggestion is not applied to the new material. Prepare the current revision rather than relying on a previous suggestion.

Reported cost and reserved liability are different. The current preparation service does not obtain authoritative billed cost from the typed result; unknown billing remains an explicit liability rather than being reported as zero.

## Record use separately from correctness

The current UI offers **Used suggestion**, **Overruled suggestion**, and **Did not use**. These describe adoption, not whether the suggestion was correct.

Independent verification is a separate operation with an evidence reference and a recorded correctness outcome. Clicking “Used suggestion” must not create an independently verified label.

Likewise, an assistance result does not change the source item's required review, severity, reference labels, or approval authority.

## Review timers are declared activity

The UI can start and pause a review timer and display recorded active time and queue time. It excludes unclosed intervals rather than guessing their duration.

Those records do not by themselves demonstrate time savings. To evaluate assistance, compare an appropriate unassisted workflow, account for correctness, and keep observed activity separate from estimated effort.

## Recovery

For a pending attempt, inspect its persisted status before retrying. For stale material, reopen the current revision. For a blocked result, investigate operator configuration, allowed provider/model, data-transfer policy, and budget. Do not repeatedly submit new keys to work around an unresolved dispatch.

Continue with [Read evaluation evidence](/docs/guides/read-evaluation-evidence) and [Qualification and authorization](/docs/platform/evaluator-qualification).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.