# Inspect a local benchmark

Inspect the named MRMS benchmark using local state and the public read contract. The launcher downloads historical source data; it does not deploy to Cloudflare. This checks ingestion and coherent revisions, not property damage or business outcomes.

This procedure exercises the shared foundations of [Storm History](/stormtrust/product/platform-and-data/intelligence/storm-history/) and [Storm Replay](/stormtrust/product/platform-and-data/intelligence/storm-replay/). Their [product acceptance contract](/stormtrust/product/platform-and-data/intelligence/history-replay-contract/#acceptance-scenarios) also requires customer discovery, retained report evidence, access, and time-mode behavior beyond this local benchmark check.

## Prerequisites

Follow the [local development prerequisites](/stormtrust/engineering/operations/local-development/). Read the [revision and retention contract](/stormtrust/product/platform-and-data/live-revisions-and-historical-evaluation/). Allow time and local storage for historical downloads and decoding.

1. Start the shared stack from the repository root.  
```sh  
make dev  
```  
The launcher applies local migrations and requests the definition in `org/apps/data/benchmarks/kansas-city-2026-03-10.json`. Wait for **Historical storm request accepted**. Acceptance means work was queued, not that ingestion finished. The launcher can remain running if this request fails; inspect its failure output before continuing.
2. In another terminal, inspect source status and the event list.  
```sh  
curl --fail --silent --show-error http://stormtrust.localhost:8787/v1/data/status  
curl --fail --silent --show-error http://stormtrust.localhost:8787/v1/storm-events  
```
3. Read the benchmark revisions.  
```sh  
curl --fail --silent --show-error \  
  'http://stormtrust.localhost:8787/v1/storm-events/kansas-city-hail-2026-03-10/revisions?limit=48'  
```  
Confirm that returned revisions belong to the requested event and have the expected observation window and products. A successful HTTP response with no completed revisions is not successful replay evidence.
4. Open the local map and inspect an available historical revision. Compare its source, time, and products with the API response. Do not substitute demonstration properties or a nearby ground report for measured outcomes.
5. Record the event ID, revision IDs, and any source or decoding failures. Stop the launcher with `Ctrl+C` when finished.

Production uses the trusted scheduler and checked-in benchmark registry. Do not replace the local hostname with a production hostname to queue this work; the internal mutation route is not part of the public API.