For each dependency edge, this fetches all successful releases across all matching resources and then de-dupes in JS (orderBy(desc(job.completedAt)) + if (byResource.has(...)) continue). This can become expensive for long-lived deployments. Prefer doing the “latest per (resource, environment)” selection in SQL (e.g., DISTINCT ON / window function) and/or fetching for all dependencyDeploymentIds in one query instead of one query per edge.

The upstream release lookup does not filter by input.environmentId, so it can pick a successful dependency release from a different environment for the same resource. Since release is keyed by (resourceId, environmentId, deploymentId), this can cause the UI to mark a dependency satisfied/unsatisfied incorrectly. Include an eq(schema.release.environmentId, input.environmentId) condition (or otherwise ensure you’re selecting the release for the same release-target environment).

This performs one SQL query per dependency edge (Promise.all(edges.map(...))). With many dependencies this becomes an N+1 pattern and adds latency. Consider fetching the latest successful release per dependency deployment in a single query (e.g., inArray(release.deploymentId, ...) + grouping / window function / DISTINCT ON) and then mapping results in memory.

The “current version” lookup ignores release.environmentId entirely and de-duplicates only by resourceId. If the same resource exists in multiple environments, this will attach the latest successful release from an arbitrary environment to all targets for that resource, producing incorrect dependency satisfaction per-environment. Include release.environmentId in the query and key the map by (environmentId, resourceId) (or query per target environment).

currentVersion is looked up via byResource?.get(t.resourceId) even though targets are environment-specific. If the same resourceId appears under multiple environments, this will reuse the same current version across environments. Once the upstream query is fixed, use an environment-aware key here as well (e.g., lookup by ${t.environmentId}:${t.resourceId}).

DependencyGroupHeader renders a <Link> (anchor) inside a <button>, which is invalid HTML and can break keyboard/screen-reader interaction. Refactor so there is only one interactive element (e.g., make the header a non-button container with a separate toggle button, or make the whole row a button and move the link outside).

Typo in user-facing text: dependenc is missing characters and will render incorrectly. Update the string to dependencies (and keep the pluralization logic intact).

⚠️ Potential issue | 🟠 Major

#!/bin/bash
# Inspect where DependencyDetail is mounted and whether it can fan out across lists of versions.
rg -n -C3 '\bDependencyDetail\b|deploymentVersions\.dependencies\.useQuery' apps/web/app/routes/ws/deployments --iglob '*.tsx'
rg -n -C3 '\bDeploymentVersion\b' apps/web/app/routes/ws/deployments --iglob '*.tsx'

# Read DependencyDetail.tsx to understand the dialog structure and the query at lines 345-357
cat -n apps/web/app/routes/ws/deployments/_components/environmentversiondecisions/rule-results/DependencyDetail.tsx | sed -n '300,370p'

# Search for Dialog usage patterns with state management
rg -n -B2 -A5 'useState.*open|Dialog.*open' apps/web/app/routes/ws/deployments --iglob '*.tsx' | head -80

# Get the complete DependencyDetail component
cat -n apps/web/app/routes/ws/deployments/_components/environmentversiondecisions/rule-results/DependencyDetail.tsx | head -360

Gate the polling interval on the dialog's open state.

Every mounted DependencyDetail polls every 15 seconds regardless of whether its dialog is open. Since DeploymentVersion renders this per visible version, this creates one background query per version row.

Add dialog state management and conditionally enable the query:

• Track dialog open state with useState
• Pass open={open} onOpenChange={setOpen} to the <Dialog> component
• Gate the query with an enabled condition or conditionally set refetchInterval to false when the dialog is closed

This follows the pattern used elsewhere in the codebase for controlled dialogs (e.g., RedeployDialog, VersionActionsPanel).

Verify each finding against the current code and only fix it if needed.

In
`@apps/web/app/routes/ws/deployments/_components/environmentversiondecisions/rule-results/DependencyDetail.tsx`
around lines 323 - 326, DependencyDetail currently polls via
trpc.deploymentVersions.dependencies.useQuery({ versionId }, { refetchInterval:
15_000 }) even when its dialog is closed; add local dialog state (const [open,
setOpen] = useState(false)), pass open={open} onOpenChange={setOpen} to the
Dialog component, and gate the query by either using the enabled: open option or
set refetchInterval: open ? 15_000 : false so the background polling only runs
while the dialog is open.