diff --git a/.planning/phases/13-launch-verification/evidence/06-hotfix-alpha24.md b/.planning/phases/13-launch-verification/evidence/06-hotfix-alpha24.md
new file mode 100644
index 0000000..2aa3a4b
--- /dev/null
+++ b/.planning/phases/13-launch-verification/evidence/06-hotfix-alpha24.md
@@ -0,0 +1,46 @@
+# Hotfix Release v1.0.0-alpha.24
+
+**Date:** 2026-09-29
+**Trigger:** Plan 13-04's LAUNCH-01 verification, retried after v1.0.0-alpha.23's PDF-extraction
+fix — the resubmitted job still failed, this time with a generic "An unexpected error occurred"
+message and, critically, **zero worker log lines** for the job on the production container.
+
+## Root cause
+
+Cross-checked BullMQ's own job history directly against Redis (Upstash) from inside the app
+container:
+
+```
+completed job 4 analysisId= ed338c9f-... ts= 2026-09-29T18:19:10.219Z
+completed job 3 analysisId= 8f1bea54-... ts= 2026-09-29T18:07:36.614Z
+```
+
+Both of this session's test jobs show as **completed** at the BullMQ level, with timestamps
+matching exactly when they were submitted — proving a worker process did pick them up and run
+them to completion (no exception escaped to mark them "failed" at the queue level). But
+`clarify-worker-prod`'s own logs (checked via `docker logs` and the raw
+`/var/lib/docker/containers/.../*-json.log` file directly) contain **zero** lines about either
+job — not even the unconditional `[Worker] Started processing...` line that is always the very
+first thing the processing function logs.
+
+Found the actual consumer: a local dev Docker container (`clarify-app`, `nuxt dev --port 3001`,
+running continuously for 3 days) shares the exact same Upstash Redis instance as production — the
+server's `.env` was intentionally copied from Carlos's local dev `.env` during 13-02, including
+`REDIS_HOST`/`REDIS_PORT`/`REDIS_TOKEN`. Local dev runs a single combined process (no
+`DISABLE_WORKER`), so it also runs the BullMQ worker in-process, competing with production's
+dedicated worker container for the same unprefixed `analysis-queue`. Whichever consumer's `BRPOPLPUSH`
+wins claims the job; this time it was local dev, which presumably failed it for reasons specific to
+that environment (not diagnosed further — out of scope once the collision itself was identified).
+
+## Fix
+
+PR #57 (`fix/bullmq-queue-prefix-per-environment` → `develop`): BullMQ's built-in `prefix` option
+namespaces all of its Redis keys. Added a shared `queuePrefix` constant (`"prod"` in production,
+`"dev"` everywhere else, keyed off `NODE_ENV`) used identically by both the `Queue`
+(`server/utils/queue.ts`) and the `Worker` (`server/plugins/worker.ts`) — they still see each
+other's jobs within the same environment, but no longer collide across environments.
+
+## This release
+
+Fourth same-day hotfix, following the same GitFlow process verified in plan 13-03 and repeated for
+v1.0.0-alpha.21, v1.0.0-alpha.22, and v1.0.0-alpha.23.
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 0d4b9a9..76231cb 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,11 @@
 
 All notable changes to this project will be documented in this file.
 
+## [1.0.0-alpha.24] - 2026-09-29
+
+### Fixed
+- Local dev and production shared the same unprefixed BullMQ "analysis-queue" on the same Upstash Redis instance, so a long-running local `nuxt dev` process raced production's worker for the same jobs and sometimes claimed and failed them before production ever saw them — discovered live during LAUNCH-01 verification via BullMQ's own job history showing jobs "completed" that the production worker's logs never mentioned. Fixed by namespacing BullMQ's Redis keys with an environment-derived `prefix` (`prod` vs `dev`).
+
 ## [1.0.0-alpha.23] - 2026-09-29
 
 ### Fixed
diff --git a/package.json b/package.json
index b255f3b..ffcb430 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "clarify",
-  "version": "1.0.0-alpha.23",
+  "version": "1.0.0-alpha.24",
   "private": true,
   "type": "module",
   "scripts": {
diff --git a/server/plugins/worker.ts b/server/plugins/worker.ts
index 98c6237..2a580ba 100644
--- a/server/plugins/worker.ts
+++ b/server/plugins/worker.ts
@@ -1,5 +1,5 @@
 import { Worker } from "bullmq";
-import { getRedisConnection } from "../utils/queue";
+import { getRedisConnection, queuePrefix } from "../utils/queue";
 import { extractTextFromPDF } from "../utils/pdf-parser";
 import { analyzeContract } from "../utils/openai-client";
 import { getWorkerSupabaseClient } from "../utils/worker-supabase";
@@ -166,6 +166,7 @@ export default defineNitroPlugin((_nitroApp) => {
     },
     {
       connection: getRedisConnection() as any, // Cast to any to resolve ioredis version mismatch
+      prefix: queuePrefix, // must match getAnalysisQueue()'s prefix in ../utils/queue
       concurrency: parseInt(process.env.BULLMQ_CONCURRENCY || '2', 10),
       defaultJobOptions: {
         attempts: 3,
diff --git a/server/utils/queue.ts b/server/utils/queue.ts
index c8c22d7..9e36e3e 100644
--- a/server/utils/queue.ts
+++ b/server/utils/queue.ts
@@ -4,6 +4,16 @@ import { Redis } from "ioredis";
 let analysisQueue: Queue | null = null;
 let redisConnection: Redis | null = null;
 
+// [LAUNCH-01 FIX] Namespaces BullMQ's Redis keys per environment. Local dev
+// and production point at the same Upstash instance (same REDIS_HOST/TOKEN
+// copied from .env for convenience), so without this both a locally running
+// `nuxt dev` and the production worker race to claim jobs from the same
+// unprefixed "analysis-queue" — discovered live when a production job was
+// silently stolen and failed by a 3-day-old local dev process. Must match
+// the prefix used when constructing the Worker in server/plugins/worker.ts.
+export const queuePrefix =
+  process.env.NODE_ENV === "production" ? "prod" : "dev";
+
 export const getRedisConnection = () => {
   if (!redisConnection) {
     const config = useRuntimeConfig();
@@ -31,6 +41,7 @@ export const getAnalysisQueue = () => {
   if (!analysisQueue) {
     analysisQueue = new Queue("analysis-queue", {
       connection: getRedisConnection() as any, // Cast to any to resolve ioredis version mismatch
+      prefix: queuePrefix,
       defaultJobOptions: {
         attempts: 3,
         backoff: {