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

Inline comments:
In `@src/app/api/route-detour/route.ts`:
- Around line 60-62: The processStation function returns objects of shape { id:
string; detourMin: number } that are serialized and enqueued without runtime
validation; add a Zod schema (e.g., const stationResponseSchema = z.object({ id:
z.string(), detourMin: z.number() })) and use stationResponseSchema.parse(...)
on the object returned from processStation (or right before the enqueue() call
referenced at enqueue() around line 124) to validate the response payload;
ensure the new schema is imported/defined alongside stationSchema/bodySchema and
throw or handle parse errors so only validated objects are enqueued.
- Around line 85-95: The handler currently ignores request.signal and continues
dequeuing/processing detours; update the detour loop to check
request.signal.aborted before calling dequeue and before each expensive call,
and pass that signal into getRouteDuration (e.g., getRouteDuration(..., {
signal: request.signal })) so in-flight work can be aborted; additionally
validate any streamed response objects with your Zod schema (the same schema
used for API responses) before encoding/queueing them to the stream to ensure
only validated objects are emitted (apply these changes around the detour
processing loop and calls to getRouteDuration and the stream encoder).

In `@src/components/home-client.tsx`:
- Around line 280-350: The scheduled flush timer (flushTimer) isn't cleared on
abort/error paths causing stale pending to be merged later; ensure you
clearTimeout(flushTimer) and set flushTimer = null before any early return or
when handling errors (e.g., inside the AbortError branch and the catch block)
and also clear it in the finally/cleanup path after the reader loop so flush()
cannot run after controller.abort() — update usages in the async IIFE that
reference reader.read(), flush(), scheduleFlush(), flushTimer, controller, and
setDetourMap to always clear the timer (and null it) before returning or
rethrowing.

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

Inline comments:
In `@src/components/home-client.tsx`:
- Around line 339-353: At EOF, the normal completion path currently flushes
pending results but doesn't mark eligible stations that were never seen as
failed; update the EOF handling in the same block that clears flushTimer and
calls flush() to backfill any ids in eligibleIds that are not in seen into
pending with a failure value (e.g., -1) before calling scheduleFlush()/flush(),
so any unseen ids are resolved; locate the variables pending, seen, eligibleIds,
scheduleFlush, flushTimer, controller.signal.aborted, and flush() in the
function and add the backfill step (pending[id] = -1; seen.add(id)) for each id
in eligibleIds minus seen before the final flush.

---

Outside diff comments:
In `@src/components/home-client.tsx`:
- Around line 276-278: The abort handler creates an AbortController
(detourAbortRef) and clears the queue but doesn’t cancel in-flight Valhalla
calls because getRouteDuration/processStation are not given the signal; update
calls that trigger network work (calls to getRouteDuration and processStation)
to accept and propagate an AbortSignal (pass controller.signal from the
AbortController created where detourAbortRef.current is set), and ensure
getRouteDuration (and any fetch/axios logic it uses) respects the signal and
throws/returns early on abort so in-flight requests are cancelled; update all
similar call sites (including the blocks around the other noted ranges) to pass
the signal through.

---

Nitpick comments:
In `@src/components/home-client.tsx`:
- Around line 14-16: The constant DETOUR_FLUSH_MS uses SCREAMING_SNAKE_CASE but
must be renamed to camelCase; rename the identifier DETOUR_FLUSH_MS to
detourFlushMs in src/components/home-client.tsx and update every usage/reference
in the same module (and any tests or other modules that import or reference it)
to the new name, preserving the numeric value (150) and keeping the comment
intact; ensure TypeScript/ESLint passes after renaming.

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

Inline comments:
In `@src/lib/valhalla.ts`:
- Around line 139-141: The abort/timeout logic using signal ?
AbortSignal.any([signal, AbortSignal.timeout(5000)]) : AbortSignal.timeout(5000)
can cause callers to receive an AbortError (reject) instead of the function
returning null as declared; update the network/streaming call that uses this
signal to catch aborts and normalize them to null: wrap the await that uses this
signal (the fetch/stream call in this module that references the signal /
AbortSignal.any / AbortSignal.timeout) in a try/catch, detect an abort via
error.name === "AbortError" (or error instanceof DOMException where name ===
"AbortError"), return null for that case, and rethrow other errors so behavior
matches the declared return contract. Ensure the signal construction remains the
same but that AbortErrors are handled locally and converted to null.
- Around line 139-141: The project uses AbortSignal.any (and
AbortSignal.timeout) in src/lib/valhalla.ts which requires Node ≥19; add an
engines.node entry to package.json (e.g., "node": ">=19.0.0" or ">=22.0.0" to
match CI) so local dev reproduces the CI runtime, update package.json
accordingly and communicate this requirement in README or contributor docs;
ensure package manager respects engines (e.g., npm/yarn) if you want
enforcement.