[noahzuerii]

• PluginManager: fix a race where disabling a layer while enablePlugin() was still resolving its awaits could restart polling and re-emit layerToggled(true), leaking a background fetch loop and a stale websocket subscription for a layer the UI showed as disabled. Fixed with a per-plugin toggle epoch that lets an in-flight enablePlugin detect it was superseded by a later toggle.
• data-query service / region API route: clamp limit with a lower bound of 1 (matching searchEntities' existing clamp), so a negative limit can no longer produce a negative Array.slice bound and return an incorrect entity set.
• favoritesSlice: revert the optimistic add/remove on a failed POST/DELETE and surface an error toast, following the same optimistic-flip-with-rollback pattern already used in alertsSlice.
• dataSlice.getAllEntities: filter by the layer's enabled flag to match its own docstring and PluginManager.getAllEntities' semantics.

Fix: chore/refactor+fix mix -> patch bump per project convention.

Type of Change

• Bug fix
• New feature
• New plugin
• Refactor / code quality
• Documentation
• Performance improvement

[codecov]

Codecov Report

❌ Patch coverage is 96.29630% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/core/plugins/PluginManager.ts	87.50%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[silvertakana]

Thanks for this - a careful read of the state and API layers, and two of these are bugs I am glad to have found.
Where it stands: the branch is ten commits behind main and conflicts, but only on package.json (main has since moved ioredis, size-limit and jsdom). The four code files have no semantic drift against current main, so the fixes apply as written - a rebase is all that is needed.
What I verified against current main:

1. PluginManager enable/disable race - real, and the fix is right. src/core/plugins/PluginManager.ts:220 awaits the local engine manifest before setting enabled = true (L228) and starting polling (L245), while disablePlugin (L257) is synchronous. Toggling a layer off during a slow enable resurrects it: polling restarts and a stale layerToggled(true) is emitted for a layer the user disabled. The per-plugin epoch guard closes it, and because no await sits between the final check and the emit, no interleave survives.
2. Negative limit clamp - real, and the fix is right. src/app/api/v1/entities/region/route.ts:65 passes a negative ?limit= straight through, and src/lib/data-query/service.ts:253 does the same for bounds.limit. A negative limit reaches allMatched.length > effectiveLimit at service.ts:291, where slice(0, -5) returns a size-dependent, arbitrary subset rather than an error. searchEntities already clamps at service.ts:198, so your precedent holds.
3. dataSlice.getAllEntities ignoring the enabled flag - a real defect, but nothing calls it yet. The method contradicts its own docstring at dataSlice.ts:21-22 and its PluginManager twin at PluginManager.ts:388, which is worth fixing on its own terms. Being straight about the impact, though: git grep getAllEntities finds no caller under src/, and the two grids that actually render entities (searchEntities.tsx:23, GlobeView.tsx:97) read entitiesByPlugin directly and apply the enabled flag themselves. This one is latent drift rather than something a user hits today.
4. Favourites optimistic rollback - real and user-visible, the strongest of the four. .catch() on a fetch only sees network failures; a 500 or a 401 resolves, so the optimistic favourite stays on screen while nothing was saved, and a failed delete looks like it worked. Checking res.ok, reverting the optimistic update, and surfacing a toast is the right shape, and the duplicate guard on the re-add is a nice touch.
   showErrorToast is already on the store (FETCH_HEAD:src/core/state/alertsSlice.ts:115: get().showErrorToast?.(result.error ?? "Failed to update alert rule");), so the toast call is real rather than a silent no-op.
   CI on this branch: FAILURE: 1 SKIPPED: 2 SUCCESS: 21 (I approved the workflow runs so they could execute on a fork pull request).
   So: I am happy to take all four. A rebase onto main to clear the package.json conflict is the only thing standing in the way.

[github-actions]

Left for a person rather than auto-merged: not a Dependabot pull request. Patch and minor updates merge themselves once every check has completed green; this one is listed in the weekly dependency digest.