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

# Reject a bad release before it reaches production

> Use bounded validation and deterministic Release Guard evidence to keep an active revision safe.

# Make “production remains unchanged” the default failure mode

An update creates an immutable candidate revision. It does not mutate the active revision in place.
Release Guard compares compatible, persisted active and candidate evidence before promotion.

```bash theme={"theme":{"light":"github-light-default","dark":"vesper"}}
infercrane plan deployment.yaml
infercrane apply deployment.yaml
```

Validate with explicit cost and privacy acknowledgement. Validation traffic is bounded by persisted
policy; InferCrane does not shadow or duplicate production requests silently.

```bash theme={"theme":{"light":"github-light-default","dark":"vesper"}}
infercrane rollout policy set qwen-prod \
  --require-compatibility \
  --require-synthetic \
  --validation-max-requests 100 \
  --validation-max-concurrency 4

infercrane rollout validate qwen-prod \
  --requests 100 \
  --concurrency 4 \
  --acknowledge-validation-cost \
  --wait

infercrane rollout inspect qwen-prod
```

```text theme={"theme":{"light":"github-light-default","dark":"vesper"}}
ACTIVE        rev-18
CANDIDATE     rev-19

Metric             Active       Candidate
TTFT p95             221ms           317ms

Guard: REJECT

Reason
TTFT regression +43%; maximum allowed +15%.
```

The values above illustrate the output shape; your result comes from persisted measurements. When
evidence is missing or incomparable, Release Guard returns `WAIT` instead of inventing a result.

## What the decision proves

```mermaid theme={"theme":{"light":"github-light-default","dark":"vesper"}}
flowchart LR
  ACTIVE[Active revision<br/>still serving] --> COMPARE[Persisted comparison]
  CANDIDATE[Candidate revision<br/>not in logical route] --> COMPARE
  POLICY[Snapshotted policy] --> COMPARE
  COMPARE -->|REJECT| CLEAN[Keep active<br/>clean candidate]
  COMPARE -->|ACCEPT| PROMOTE[Publish candidate<br/>drain old generation]
```

Every evaluation records:

* active and candidate revision identity;
* the exact policy snapshot;
* compatible benchmark IDs and workload parameters;
* measurements that were available;
* explicit unavailable measurements;
* deterministic reason codes and timestamp.

Use the persisted decision during an incident or approval review:

```bash theme={"theme":{"light":"github-light-default","dark":"vesper"}}
infercrane explain rollout qwen-prod
infercrane rollout inspect qwen-prod --output json
```

<Tip>
  After an accepted promotion, issue an [Inference Passport](/features/inference-passports) to bind the
  revision to its release evidence in an independently verifiable artifact.
</Tip>