Move the distributed bandit out of this repository

.github/workflows/ci.yml

@@ -17,7 +17,7 @@ jobs:
       matrix:
         # Every module has its own go.mod and is linted independently. Keep
         # this in step with MODULES in the Makefile.
-        module: [".", "lfu", "policies", "policies/arc", "policies/fifo", "policies/tinylfu", "metrics", "bandit", "bandit/redis", "benchclient", "bench", "examples/basic", "examples/migration"]
+        module: [".", "lfu", "policies", "policies/arc", "policies/fifo", "policies/tinylfu", "metrics", "bandit", "benchclient", "bench", "examples/basic", "examples/migration"]
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-go@v5
@@ -37,7 +37,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        module: [".", "lfu", "policies", "policies/arc", "policies/fifo", "policies/tinylfu", "metrics", "bandit", "bandit/redis", "benchclient", "bench", "examples/basic", "examples/migration"]
+        module: [".", "lfu", "policies", "policies/arc", "policies/fifo", "policies/tinylfu", "metrics", "bandit", "benchclient", "bench", "examples/basic", "examples/migration"]
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-go@v5
@@ -84,52 +84,3 @@ jobs:
           cache-dependency-path: go.sum
       # Reads each go.mod through `go mod edit -json`, so no network is needed.
       - run: ./scripts/release-check.sh
-
-  # The test job above runs bandit/redis against miniredis, which is a fake.
-  # This one runs the same suite against real servers, because the adapter
-  # leans on three things a fake can be too permissive about: TIME called
-  # inside a Lua script, SET with NX and PX, and HINCRBY on a key the script
-  # names itself rather than declaring in KEYS.
-  #
-  # Both engines run, because both are documented as supported. Testing one
-  # while documenting two is how that claim quietly stops being true. This is
-  # the CI equivalent of `make redis-test`; docker-compose.yml is the local one.
-  redis:
-    name: test bandit/redis (${{ matrix.engine.name }})
-    runs-on: ubuntu-latest
-    strategy:
-      fail-fast: false
-      matrix:
-        engine:
-          # Deriving the bucket from the server's clock inside a script needs
-          # effects replication, so the floor is Redis 7 / Valkey 7.2.
-          - name: valkey 8
-            image: valkey/valkey:8-alpine
-            ping: valkey-cli ping
-          - name: redis 7
-            image: redis:7-alpine
-            ping: redis-cli ping
-    services:
-      server:
-        image: ${{ matrix.engine.image }}
-        ports:
-          - 6379:6379
-        options: >-
-          --health-cmd "${{ matrix.engine.ping }}"
-          --health-interval 1s
-          --health-timeout 3s
-          --health-retries 20
-    steps:
-      - uses: actions/checkout@v4
-      - uses: actions/setup-go@v5
-        with:
-          go-version: "1.25"
-          cache-dependency-path: bandit/redis/go.sum
-      - name: go test
-        working-directory: bandit/redis
-        # Set, so the suite uses the service container rather than miniredis.
-        # The suite fails rather than skips if this address is unreachable, so
-        # a broken service container cannot pass as a green fake run.
-        env:
-          AS_CACHE_REDIS_ADDR: 127.0.0.1:6379
-        run: go test -race -count=1 ./...

Makefile

@@ -1,6 +1,6 @@
 # Each of these directories is a separate Go module (own go.mod), so tooling is
 # run once per module. The root .golangci.yml is shared by all of them.
-MODULES := . lfu policies policies/arc policies/fifo policies/tinylfu metrics bandit bandit/redis benchclient bench examples/basic examples/migration
+MODULES := . lfu policies policies/arc policies/fifo policies/tinylfu metrics bandit benchclient bench examples/basic examples/migration
 
 GOLANGCI_LINT_VERSION := v2.8.0
 
@@ -50,33 +50,6 @@ release-check: ## Check the repository could actually be released today
 evidence: ## Replay the workload suite and print the policy comparison tables
 	( cd bench && go test -count=1 -timeout 20m -v ./... )
 
-# The bandit/redis tests run against miniredis by default, which is a fake.
-# The Lua the adapter depends on - TIME inside a script, SET NX PX, HINCRBY on
-# a key the script names itself - is exactly what a fake can be too permissive
-# about, so the same suite runs against real servers here. See
-# docker-compose.yml for why both engines are covered rather than one.
-COMPOSE ?= docker compose
-
-.PHONY: redis-up
-redis-up: ## Start the local Valkey and Redis containers and wait for them
-	$(COMPOSE) up -d --wait
-
-.PHONY: redis-down
-redis-down: ## Stop the local Valkey and Redis containers
-	$(COMPOSE) down -v
-
-.PHONY: redis-test
-redis-test: ## Run the bandit/redis suite against real Valkey and Redis
-	@$(MAKE) redis-up
-	@status=0; \
-		for target in "valkey 127.0.0.1:63799" "redis 127.0.0.1:63798"; do \
-			set -- $$target; \
-			echo "==> bandit/redis against $$1 ($$2)"; \
-			( cd bandit/redis && AS_CACHE_REDIS_ADDR=$$2 go test -race -count=1 ./... ) || status=1; \
-		done; \
-		$(MAKE) redis-down; \
-		exit $$status
-
 .PHONY: tidy
 tidy: ## Run go mod tidy across all modules
 	@set -e; for m in $(MODULES); do \

README.md

@@ -40,9 +40,8 @@ been exercised under the race detector and adversarially reviewed. It is all in
 | [Configuration](docs/configuration.md) | Every `Settings` field, migration strategies, sampling, stability gates, tuning |
 | [Policies](docs/policies.md) | The nine ready-made arms, their caveats, adapting your own cache |
 | [Advisor mode](docs/advisor-mode.md) | `ObserveOnly`, `Advice()`, and the `metrics` module |
-| [Evidence](docs/evidence.md) | Every measured claim: policy tables, competing libraries, real traces, sampling fidelity, fleets |
+| [Evidence](docs/evidence.md) | Every measured claim: policy tables, competing libraries, real traces, sampling fidelity |
 | [Benchmarking](docs/benchmarking.md) | Reproducible replays, `benchclient`, `make evidence` |
-| [Running a fleet](docs/fleet.md) | Pooling evidence across replicas through Valkey or Redis |
 | [Project site](https://sshaplygin.github.io/as-cache/) | Landing page, plus an interactive explorer of the bandit's decisions on a phase-shift run |
 
 Past releases are recorded in the [changelog](CHANGELOG.md) and on the