fix(build): hold back packages the cran mirror has not picked up yet (#188)

## Motivation

A resolute build aborted on a single package:

```
[23/361] AsyPeer_0.0.1 (r_minor_sensitive=TRUE)
Error: GitHub API error (404): Not Found
x URL not found: <https://api.github.com/repos/cran/AsyPeer/commits>
Retrying in 2 seconds. ... Retrying in 60 seconds.
Error in `rate_sleep()`: ! Request failed after 10 attempts.
Execution halted
```

This is not rate limiting — it is a **404**. `check_for_binary()` reads the published version from the `cran` GitHub mirror, and that mirror lags CRAN. `AsyPeer 0.0.1` was published today at 13:50 UTC and has no repository there yet.

The 404 is permanent, but the call is wrapped in `purrr::insistently` with `max_times = 10` and `pause_cap = 60`, so it retries on a 1/2/4/8/16/32/60/60/60/60 second backoff — about five minutes — and then aborts the whole shard.

## Change

Hold back release versions published within `CRAN_MIRROR_LAG_DAYS` (default 3).

Deferring them costs nothing: the daily update pipeline builds new and updated packages anyway, and they arrive here on the next run once the mirror has caught up.

## Measured against the live CRAN index

| lag | held back |
|---|---|
| 1 day | 29 of 24831 (0.12%) |
| **3 days** | **135 (0.54%)** |
| 7 days | 412 (1.66%) |

`AsyPeer` is among the 135 at three days.

## Worth doing separately

Retrying a 404 at all is wrong — it can never succeed, and any other permanent 404 (a package pulled from the mirror, say) will abort a shard the same way. `check_for_binary()` should distinguish a permanent 404 from a transient failure and, when the mirror simply lacks the package, treat the version as unknown rather than fatal. That is a bincraft change and I have not made it here.

Reviewed-on: #188

.crow/build-all-versions.yaml

@@ -35,6 +35,14 @@ variables:
       - 'resolute'
     default: '3.24'
   R_VERSION:
+    # The slot's *primary* R minor: what `local/build-all.R` builds into the
+    # generic slot. The loop below already runs `--sensitive-only` for every
+    # other installed minor, so filling a non-primary minor's gap needs this
+    # left alone, not changed.
+    #
+    # 4.6.0 was briefly offered here (#183) and removed: selecting it for a
+    # slot whose generic binaries are 4.5-built would publish 4.6 binaries
+    # into the generic slot and break every 4.5 client.
     description: 'Primary R version under /opt/R.'
     options:
       - 4.5.3

.crow/reindex.yaml

@@ -0,0 +1,193 @@
+# Re-index every slot without rebuilding anything.
+#
+# `weekly-rebuild-reindex` exists to run after `weekly-rebuild-missing`, so it
+# depends on that workflow and shares its gate: triggering it manually also
+# starts hours of package rebuilds. That is the wrong tool when only the index
+# needs regenerating - after a bincraft release that changes how the index is
+# written, for instance, where the objects in the bucket are already correct
+# and only `PACKAGES*` is stale.
+#
+# This workflow does the index half on its own. It installs the latest bincraft
+# release, republishes the generic and per-R-minor indexes for each slot, and
+# purges the edge. No package is built.
+#
+# Trigger with the `reindex` variable set to `all` or to a single
+# `<os>-<arch>`, e.g.
+#
+#   crow pipeline create devxy/build-cran-binaries --var reindex=all
+
+variables:
+  # A manual pipeline creation instantiates every file in .crow/, so the
+  # default must match no matrix row.
+  reindex:
+    description: "Re-index target: a specific <os>-<arch>, 'all' for every slot, or 'none'."
+    options:
+      - none
+      - all
+      - alpine-322-amd64
+      - alpine-322-arm64
+      - alpine-323-amd64
+      - alpine-323-arm64
+      - alpine-324-amd64
+      - alpine-324-arm64
+      - redhat-8-amd64
+      - redhat-8-arm64
+      - redhat-9-amd64
+      - redhat-9-arm64
+      - redhat-10-amd64
+      - redhat-10-arm64
+      - ubuntu-2204-amd64
+      - ubuntu-2204-arm64
+      - ubuntu-2404-amd64
+      - ubuntu-2404-arm64
+      - ubuntu-2604-amd64
+      - ubuntu-2604-arm64
+    default: none
+
+when:
+  - event: manual
+    evaluate: 'reindex == "all" || reindex == "${OS}-${ARCH}"'
+
+skip_clone: true
+
+labels:
+  group: rpkgs-${ARCH}
+
+matrix:
+  include:
+    - OS: alpine-322
+      ARCH: amd64
+      R_VERSION: 4.5.3
+      IMG: alpine:3.22
+    - OS: alpine-322
+      ARCH: arm64
+      R_VERSION: 4.5.3
+      IMG: alpine:3.22
+    - OS: alpine-323
+      ARCH: amd64
+      R_VERSION: 4.5.3
+      IMG: alpine:3.23
+    - OS: alpine-323
+      ARCH: arm64
+      R_VERSION: 4.5.3
+      IMG: alpine:3.23
+    - OS: alpine-324
+      ARCH: amd64
+      R_VERSION: 4.5.3
+      IMG: alpine:3.24
+    - OS: alpine-324
+      ARCH: arm64
+      R_VERSION: 4.5.3
+      IMG: alpine:3.24
+    - OS: redhat-8
+      ARCH: amd64
+      R_VERSION: 4.4.3
+      IMG: redhat:8
+    - OS: redhat-8
+      ARCH: arm64
+      R_VERSION: 4.4.3
+      IMG: redhat:8
+    - OS: redhat-9
+      ARCH: amd64
+      R_VERSION: 4.4.3
+      IMG: redhat:9
+    - OS: redhat-9
+      ARCH: arm64
+      R_VERSION: 4.4.3
+      IMG: redhat:9
+    - OS: redhat-10
+      ARCH: amd64
+      R_VERSION: 4.5.3
+      IMG: redhat:10
+    - OS: redhat-10
+      ARCH: arm64
+      R_VERSION: 4.5.3
+      IMG: redhat:10
+    - OS: ubuntu-2204
+      ARCH: amd64
+      R_VERSION: 4.4.3
+      IMG: ubuntu:jammy
+    - OS: ubuntu-2204
+      ARCH: arm64
+      R_VERSION: 4.4.3
+      IMG: ubuntu:jammy
+    - OS: ubuntu-2404
+      ARCH: amd64
+      R_VERSION: 4.4.3
+      IMG: ubuntu:noble
+    - OS: ubuntu-2404
+      ARCH: arm64
+      R_VERSION: 4.4.3
+      IMG: ubuntu:noble
+    - OS: ubuntu-2604
+      ARCH: amd64
+      R_VERSION: 4.4.3
+      IMG: ubuntu:resolute
+    - OS: ubuntu-2604
+      ARCH: arm64
+      R_VERSION: 4.4.3
+      IMG: ubuntu:resolute
+
+steps:
+  - name: 'Re-index the slot'
+    image: reg.devxy.io/rpkgs/build-env-${IMG}
+    pull: true
+    environment:
+      OTEL_R_TRACES_EXPORTER: none
+      OTEL_R_LOGS_EXPORTER: none
+      OTEL_R_METRICS_EXPORTER: none
+      RED_HAT_DEV_PW:
+        from_secret: RED_HAT_DEV_PW
+      B2_S3_ACCESS_KEY:
+        from_secret: B2_S3_ACCESS_KEY
+      B2_S3_SECRET_KEY:
+        from_secret: B2_S3_SECRET_KEY
+      REPO_RO_TOKEN:
+        from_secret: REPO_RO_TOKEN
+      GIT_USER: pat-s
+      R_LIBS_USER: /mnt/cache/R-pkgs
+      R_VERSION: ${R_VERSION}
+      PLATFORM: ${OS}
+      ARCH: ${ARCH}
+    commands:
+      - git clone -q https://pat-s:$$REPO_RO_TOKEN@git.devxy.io/devxy/build-cran-binaries.git .
+      - mkdir -p /mnt/cache/R-pkgs
+      - rm -rf /mnt/cache/R-pkgs/00LOCK-*
+      - /opt/R/$R_VERSION/bin/Rscript local/install-bincraft.R
+      # The codename is detected from the image's /etc/os-release.
+      - /opt/R/$R_VERSION/bin/R -q -e 'library(bincraft); upload_package_index(s3_endpoint = "https://s3.eu-central-003.backblazeb2.com", s3_region = "eu-central-003", s3_bucket = "devxy-rpkgs-binaries", s3_access_key_id = Sys.getenv("B2_S3_ACCESS_KEY"), s3_secret_access_key = Sys.getenv("B2_S3_SECRET_KEY"))'
+      - |
+        for RBIN in /opt/R/[0-9]*/bin/R; do
+          RMINOR=$(basename "$(dirname "$(dirname "$RBIN")")" | cut -d. -f1-2)
+          /opt/R/$R_VERSION/bin/R -q -e "library(bincraft); upload_package_index(r_minor = '$RMINOR', s3_endpoint = 'https://s3.eu-central-003.backblazeb2.com', s3_region = 'eu-central-003', s3_bucket = 'devxy-rpkgs-binaries', s3_access_key_id = Sys.getenv('B2_S3_ACCESS_KEY'), s3_secret_access_key = Sys.getenv('B2_S3_SECRET_KEY'))" || true
+        done
+
+  - name: Purge CDN cache
+    image: reg.devxy.io/docker.io/library/alpine:3.24
+    environment:
+      OTEL_R_TRACES_EXPORTER: none
+      OTEL_R_LOGS_EXPORTER: none
+      OTEL_R_METRICS_EXPORTER: none
+      BUNNYNET_API_KEY:
+        from_secret: BUNNYNET_API_KEY
+      # cran.rpkgs.com and cran.allianceswisspass.devxy.io are on separate
+      # Bunny pull zones, so both must be purged after the shared origin changes.
+      # The staging zone is listed too. It shares the B2 origin, so an index
+      # it still holds is a stale copy of the same object, and its
+      # cache_expiration_time is the same ~370 days: without a purge here it
+      # serves pre-reindex indexes indefinitely and any verification run
+      # against it measures the past.
+      BUNNY_PULLZONES: '3857050 cran.allianceswisspass.devxy.io cran-rpkgs-test.b-cdn.net'
+    commands:
+      - apk add --no-cache -q bash curl jq
+      # Crow carries the checkout from the re-index step into this step.
+      - bash scripts/purge_cdn_zone.sh "$BUNNYNET_API_KEY" $BUNNY_PULLZONES
+    # Runs on every row rather than on one designated slot: a cron fires only
+    # its own slot's row, so gating on a named slot would leave every other
+    # slot unpurged. A manual "all" run therefore purges the zone 18 times,
+    # which is a cheap API call and rare.
+    #
+    # Run it even when the re-index above failed: the objects were still
+    # replaced, and a stale edge is exactly what keeps them hidden.
+    when:
+      - status: [success, failure]

.crow/weekly-rebuild-reindex.yaml

@@ -175,7 +175,12 @@ steps:
         from_secret: BUNNYNET_API_KEY
       # cran.rpkgs.com and cran.allianceswisspass.devxy.io are on separate
       # Bunny pull zones, so both must be purged after the shared origin changes.
-      BUNNY_PULLZONES: '3857050 cran.allianceswisspass.devxy.io'
+      # The staging zone is listed too. It shares the B2 origin, so an index
+      # it still holds is a stale copy of the same object, and its
+      # cache_expiration_time is the same ~370 days: without a purge here it
+      # serves pre-reindex indexes indefinitely and any verification run
+      # against it measures the past.
+      BUNNY_PULLZONES: '3857050 cran.allianceswisspass.devxy.io cran-rpkgs-test.b-cdn.net'
     commands:
       - apk add --no-cache -q bash curl jq
       # Crow carries the checkout from the re-index step into this step.

cdn.tf

@@ -52,6 +52,26 @@
 
 ### cran.rpkgs.com
 
+locals {
+  rpkgs_slots = [
+    for pair in setproduct(
+      ["amd64", "arm64"],
+      ["resolute", "noble", "jammy", "rhel8", "rhel9", "rhel10", "alpine323", "alpine324"]
+    ) : "${pair[0]}/${pair[1]}"
+  ]
+
+  # The supported R minors: the current one plus the two previous, which is
+  # exactly what build-env-images installs as R_VERSION_LATEST / PREV1 / PREV2.
+  # These must stay in step. A minor listed here without a published index
+  # sends those clients to a 404; a published minor missing from this list
+  # sends them to CRAN for sources instead of serving the binaries we built.
+  rpkgs_supported_minors = ["4.4", "4.5", "4.6"]
+
+  # bunny.net serves every pull zone on <name>.b-cdn.net, so staging needs no
+  # DNS record and is never advertised.
+  rpkgs_test_hostname = "cran-rpkgs-test.b-cdn.net"
+}
+
 # The edge middleware that resolves the bare cran.rpkgs.com form to an
 # <arch>/<os> slot and routes PACKAGES* to the per-R-minor slot. The source of
 # truth is edge/rpkgs-router.ts; `tofu apply` publishes a new release.
@@ -69,9 +89,26 @@ resource "bunnynet_compute_script" "rpkgs_router" {
 # slot that is not listed here would hide every package the per-minor index does
 # not carry, so this stays empty until a slot has been backfilled.
 resource "bunnynet_compute_script_variable" "rpkgs_router_union_slots" {
+  script = bunnynet_compute_script.rpkgs_router.id
+  name   = "UNION_SLOTS"
+  # Enabled. Until this was set, every client resolved against the generic
+  # index and never reached a per-minor binary: an R 4.6.1 client on resolute
+  # downloaded the 4.5-built rlang (2079570 bytes) while the correct 4.6 build
+  # (2075106 bytes) sat unused one directory away, and died at load with
+  # `undefined symbol: SETLENGTH`.
+  #
+  # Verified before enabling, against the staging zone with the same script and
+  # the same origin: all 16 slots report zero regressions against the generic
+  # slot, an excluded R minor is sent to CRAN, a client without an R minor
+  # still gets the generic index, and tarball requests are never rewritten.
+  default_value = join(",", local.rpkgs_slots)
+  required      = false
+}
+
+resource "bunnynet_compute_script_variable" "rpkgs_router_known_minors" {
   script        = bunnynet_compute_script.rpkgs_router.id
-  name          = "UNION_SLOTS"
-  default_value = ""
+  name          = "KNOWN_MINORS"
+  default_value = join(",", local.rpkgs_supported_minors)
   required      = false
 }
 
@@ -147,6 +184,92 @@ resource "bunnynet_pullzone_hostname" "cran_rpkgs_com" {
   tls_enabled = true
 }
 
+### Staging zone for edge-router changes
+
+# Every published <arch>/<os> slot. The staging zone enables per-minor routing
+# for all of them at once; production adopts the same list only after
+# `scripts/verify-r-minor-routing.sh --live` passes against staging.
+
+# A second copy of the same router, bound to the same B2 origin, so UNION_SLOTS
+# can be exercised end to end before production is touched.
+resource "bunnynet_compute_script" "rpkgs_router_test" {
+  type    = "middleware"
+  name    = "rpkgs-router-test"
+  content = file("${path.module}/edge/rpkgs-router.ts")
+}
+
+resource "bunnynet_compute_script_variable" "rpkgs_router_test_union_slots" {
+  script        = bunnynet_compute_script.rpkgs_router_test.id
+  name          = "UNION_SLOTS"
+  default_value = join(",", local.rpkgs_slots)
+  required      = false
+}
+
+# Without this the staging zone rewrites to PUBLIC_CDN_ORIGIN, so its redirects
+# land on production and the test silently measures the wrong system.
+resource "bunnynet_compute_script_variable" "rpkgs_router_test_extra_hosts" {
+  script        = bunnynet_compute_script.rpkgs_router_test.id
+  name          = "EXTRA_PUBLIC_HOSTS"
+  default_value = local.rpkgs_test_hostname
+  required      = false
+}
+
+resource "bunnynet_compute_script_variable" "rpkgs_router_test_known_minors" {
+  script        = bunnynet_compute_script.rpkgs_router_test.id
+  name          = "KNOWN_MINORS"
+  default_value = join(",", local.rpkgs_supported_minors)
+  required      = false
+}
+
+resource "bunnynet_pullzone" "cran_rpkgs_test" {
+  name = "cran-rpkgs-test"
+
+  cache_errors = false
+
+  cache_expiration_time = 31919000
+  websockets_enabled    = false
+  errorpage_whitelabel  = true
+
+  origin {
+    type              = "OriginUrl"
+    url               = "https://devxy-rpkgs-binaries.s3.eu-central-003.backblazeb2.com"
+    middleware_script = bunnynet_compute_script.rpkgs_router_test.id
+  }
+
+  routing {
+    filters = [
+      "scripting",
+    ]
+  }
+
+  s3_auth_enabled = true
+  s3_auth_key     = var.B2_S3_ACCESS_KEY
+  s3_auth_secret  = var.B2_S3_SECRET_KEY
+  s3_auth_region  = "eu-central-003"
+
+  cache_enabled              = true
+  request_coalescing_enabled = true
+  block_post_requests        = true
+
+  cache_vary_headers = ["User-Agent"]
+
+  # Staging carries only synthetic verification traffic, so the production
+  # ceilings would be pure headroom.
+  limit_requests    = 500
+  limit_connections = 100
+
+  safehop_enabled      = true
+  add_canonical_header = true
+  cache_stale          = ["offline", "updating"]
+  block_ips            = var.cdn_block_ips
+
+  # 1 TB
+  limit_bandwidth = 1000000000000
+
+  block_root_path = true
+}
+
+
 # Alliance SwissPass historically used a separate, manually configured pull
 # zone. Adopt it so both public repositories use the same B2 origin, middleware
 # release and cache behavior.

edge/rpkgs-router.test.ts

@@ -17,6 +17,8 @@ const UNION_SLOTS = 'amd64/alpine324';
 
 const UA_R45_MUSL = 'R (4.5.3 x86_64-pc-linux-musl x86_64 linux-musl)';
 const UA_R46_MUSL = 'R (4.6.0 x86_64-pc-linux-musl x86_64 linux-musl)';
+const UA_R43_MUSL = 'R (4.3.3 x86_64-pc-linux-musl x86_64 linux-musl)';
+const UA_R47_MUSL = 'R (4.7.0 x86_64-pc-linux-musl x86_64 linux-musl)';
 const UA_R45_ALPINE = 'R/4.5.3 R (4.5.3 x86_64-pc-linux-musl x86_64 linux-musl) Alpine Linux 3.24';
 const UA_R45_RESOLUTE = 'R/4.5.3 (Ubuntu 26.04) (aarch64-unknown-linux-gnu aarch64 linux-gnu)';
 const UA_R45_FUTURE_UBUNTU =
@@ -96,6 +98,33 @@ Deno.test('rpkgs-router', async (t) => {
       assertEquals(res.location, `https://cran.rpkgs.com${SLOT}/4.6/PACKAGES.gz`);
     });
 
+    // We publish binaries only for the supported window. An excluded minor has
+    // no slot we can serve safely, so it goes to CRAN for sources rather than
+    // to a 404 or to binaries built under another minor.
+    await t.step('sends an excluded R minor to CRAN for the index', async () => {
+      const res = await probe(`${SLOT}/PACKAGES.gz`, UA_R43_MUSL);
+      assertEquals(res.location, 'https://cran.r-project.org/src/contrib/PACKAGES.gz');
+    });
+
+    await t.step('sends a future R minor to CRAN too', async () => {
+      const res = await probe(`${SLOT}/PACKAGES.gz`, UA_R47_MUSL);
+      assertEquals(res.location, 'https://cran.r-project.org/src/contrib/PACKAGES.gz');
+    });
+
+    // The index and the tarballs R resolves against it have to come from the
+    // same place. Serving one from CRAN and the other from here would hand R a
+    // binary where it expects a source tarball.
+    await t.step('sends an excluded minor to CRAN for tarballs as well', async () => {
+      const res = await probe(`${SLOT}/foo_1.0.tar.gz`, UA_R43_MUSL);
+      assertEquals(res.location, 'https://cran.r-project.org/src/contrib/foo_1.0.tar.gz');
+    });
+
+    await t.step('leaves an excluded minor alone on a slot outside UNION_SLOTS', async () => {
+      const res = await probe(`${OTHER_SLOT}/PACKAGES.gz`, UA_R43_MUSL);
+      assertEquals(res.location, null);
+      assertEquals(res.status, 200);
+    });
+
     await t.step('routes PACKAGES and PACKAGES.rds too', async () => {
       for (const file of ['PACKAGES', 'PACKAGES.rds']) {
         const res = await probe(`${SLOT}/${file}`, UA_R45_MUSL);
@@ -146,18 +175,12 @@ Deno.test('rpkgs-router', async (t) => {
 
     await t.step('resolves Ubuntu 26.04 to the resolute slot', async () => {
       const res = await probe('/src/contrib/PACKAGES.gz', UA_R45_RESOLUTE);
-      assertEquals(
-        res.location,
-        'https://cran.rpkgs.com/arm64/resolute/latest/src/contrib/PACKAGES.gz',
-      );
+      assertEquals(res.location, 'https://cran.rpkgs.com/arm64/resolute/latest/src/contrib/PACKAGES.gz');
     });
 
     await t.step('resolves a future Ubuntu release from its codename', async () => {
       const res = await probe('/src/contrib/PACKAGES.gz', UA_R45_FUTURE_UBUNTU);
-      assertEquals(
-        res.location,
-        'https://cran.rpkgs.com/arm64/dynamic-dugong/latest/src/contrib/PACKAGES.gz',
-      );
+      assertEquals(res.location, 'https://cran.rpkgs.com/arm64/dynamic-dugong/latest/src/contrib/PACKAGES.gz');
     });
 
     await t.step('sends an unidentifiable distro to CRAN', async () => {

edge/rpkgs-router.ts

@@ -27,7 +27,17 @@ import * as BunnySDK from 'https://esm.sh/@bunny.net/edgescript-sdk@0.12';
 
 const PUBLIC_CDN_ORIGIN = 'https://cran.rpkgs.com';
 const CRAN_ORIGIN = 'https://cran.r-project.org';
-const PUBLIC_CDN_HOSTS = new Set(['cran.rpkgs.com', 'cran.allianceswisspass.devxy.io']);
+const PUBLIC_CDN_HOSTS = new Set([
+  'cran.rpkgs.com',
+  'cran.allianceswisspass.devxy.io',
+  // Staging hostnames, so the identical script can run on a test pull zone and
+  // redirect within itself. Without this a test zone rewrites to
+  // PUBLIC_CDN_ORIGIN, quietly exercising production instead of itself.
+  ...(Deno.env.get('EXTRA_PUBLIC_HOSTS') ?? '')
+    .split(',')
+    .map((host) => host.trim())
+    .filter((host) => host.length > 0),
+]);
 
 /** Slots ("<arch>/<os>", comma separated) whose per-minor index is a union. */
 const UNION_SLOTS = new Set(
@@ -37,6 +47,21 @@ const UNION_SLOTS = new Set(
     .filter((slot) => slot.length > 0),
 );
 
+/**
+ * R minors for which a per-minor index is actually published.
+ *
+ * contribPath() has no way to probe the origin, so a minor that is not
+ * published here must fall back to the flat index. Routing an unlisted minor
+ * would send that client to a 404 and it would see no packages at all - a
+ * silent, total failure rather than a degraded one.
+ */
+const KNOWN_MINORS = new Set(
+  (Deno.env.get('KNOWN_MINORS') ?? '4.4,4.5,4.6')
+    .split(',')
+    .map((minor) => minor.trim())
+    .filter((minor) => minor.length > 0),
+);
+
 /** `/<arch>/<os>/latest/src/contrib[/<rest>]` */
 const SLOT_PATH_REGEX = /^\/(amd64|arm64)\/([a-z0-9._-]+)\/latest\/src\/contrib\/?(.*)$/;
 
@@ -96,6 +121,18 @@ function publicCdnOrigin(url: URL): string {
   return PUBLIC_CDN_HOSTS.has(url.hostname) ? url.origin : PUBLIC_CDN_ORIGIN;
 }
 
+/**
+ * True when the client reports an R minor that we deliberately do not serve.
+ *
+ * A client that reports no minor at all is not "unsupported": non-R fetchers
+ * (mirror scripts, image builds) must keep getting the flat slot. Only a
+ * known-and-excluded minor falls through to CRAN.
+ */
+function isExcludedMinor(userAgent: string): boolean {
+  const rMinor = extractRMinor(userAgent);
+  return rMinor !== null && !KNOWN_MINORS.has(rMinor);
+}
+
 function extractRMinor(userAgent: string): string | null {
   for (const regex of R_MINOR_REGEXES) {
     const match = userAgent.match(regex);
@@ -177,8 +214,8 @@ function parseMacUserAgent(userAgent: string): { os: string; arch: string; rver:
  * The contrib path a request should be served from, relative to the slot.
  *
  * Returns the per-minor path for an index file when the slot is known to carry
- * a union index and the client's R minor is known; otherwise the flat path,
- * which is what every client sees today.
+ * a union index and the client's R minor is one we publish; otherwise the flat
+ * path, which is what every client sees today.
  */
 function contribPath(slot: string, rest: string, userAgent: string): string {
   const flat = rest ? `/${slot}/latest/src/contrib/${rest}` : `/${slot}/latest/src/contrib`;
@@ -188,7 +225,7 @@ function contribPath(slot: string, rest: string, userAgent: string): string {
   }
 
   const rMinor = extractRMinor(userAgent);
-  return rMinor ? `/${slot}/latest/src/contrib/${rMinor}/${rest}` : flat;
+  return rMinor && KNOWN_MINORS.has(rMinor) ? `/${slot}/latest/src/contrib/${rMinor}/${rest}` : flat;
 }
 
 BunnySDK.net.http
@@ -224,6 +261,16 @@ BunnySDK.net.http
         return Promise.resolve(ctx.request);
       }
 
+      // An R minor outside the supported window has no binaries we can safely
+      // serve, so the whole interaction goes to CRAN: the index and the
+      // tarballs R will resolve against it. Serving the index from CRAN but
+      // tarballs from here would hand R a binary where it expects a source
+      // tarball, which fails in a far more confusing way than not being
+      // served at all.
+      if (UNION_SLOTS.has(slot) && isExcludedMinor(userAgent)) {
+        return Promise.resolve(redirectTo(`${CRAN_ORIGIN}/src/contrib/${rest}`));
+      }
+
       const target = contribPath(slot, rest, userAgent);
       if (target === path) {
         return Promise.resolve(ctx.request);
@@ -238,6 +285,10 @@ BunnySDK.net.http
         return Promise.resolve(redirectTo(`${CRAN_ORIGIN}${path}`));
       }
 
+      if (UNION_SLOTS.has(slot) && isExcludedMinor(userAgent)) {
+        return Promise.resolve(redirectTo(`${CRAN_ORIGIN}${path}`));
+      }
+
       const rest = srcContrib ? srcContrib[1] : '';
       return Promise.resolve(redirectTo(`${publicOrigin}${contribPath(slot, rest, userAgent)}`));
     }

local/build-all.R

@@ -110,10 +110,25 @@ con <- DBI::dbConnect(
   password = Sys.getenv("PGPASS"),
   sslmode = "require"
 )
+# Scope the skip to the R minor this pass is running under. `single_builds`
+# records `r_version` per attempt, but querying without it made a non-primary
+# pass skip everything the primary pass had already attempted under a different
+# minor - so `--sensitive-only` under 4.6 skipped packages that had only ever
+# been built for 4.5, and the per-minor slots never filled. That is why
+# amd64/resolute served 4000 fewer packages to a 4.6 client than to a 4.5 one.
+r_minor <- paste(
+  R.version$major,
+  strsplit(R.version$minor, ".", fixed = TRUE)[[1L]][1L],
+  sep = "."
+)
 built <- DBI::dbGetQuery(
   con,
-  "SELECT name, tag FROM single_builds WHERE platform = $1 AND arch = $2",
-  params = list(platform, arch)
+  paste(
+    "SELECT name, tag FROM single_builds",
+    "WHERE platform = $1 AND arch = $2",
+    "AND substring(r_version from '^[0-9]+[.][0-9]+') = $3"
+  ),
+  params = list(platform, arch, r_minor)
 )
 DBI::dbDisconnect(con)
 before <- nrow(chunk)
@@ -121,8 +136,9 @@ chunk <- chunk[
   !paste(chunk$Package, chunk$Version) %in% paste(built$name, built$tag),
 ]
 sprintf(
-  "Skipped %d already-attempted package versions; %d remaining for this job",
+  "Skipped %d package versions already attempted under R %s; %d remaining for this job",
   before - nrow(chunk),
+  r_minor,
   nrow(chunk)
 )
 

local/packages-to-build.R

@@ -68,9 +68,36 @@ archive_versions <- archive_versions[
 ]
 
 # Now get release versions (assuming cran_release has Package and Version columns)
+#
+# Packages published in the last few days are held back. `check_for_binary()`
+# reads the published version from the `cran` GitHub mirror
+# (`GET /repos/cran/<pkg>/commits`), and that mirror lags CRAN: a package that
+# has just appeared has no repository there yet. The call then 404s, which is
+# permanent, but it is wrapped in `purrr::insistently` and retried ten times
+# with a backoff capped at 60s - so one unmirrored package burns about five
+# minutes and then aborts the whole shard.
+#
+# Holding them back costs nothing: the daily update pipeline builds new and
+# updated packages anyway, and they arrive here on the next run once the mirror
+# has caught up.
+mirror_lag_days <- as.numeric(
+  Sys.getenv("CRAN_MIRROR_LAG_DAYS", unset = "3")
+)
+published <- as.POSIXct(cran_release$Published, tz = "UTC")
+too_recent <- !is.na(published) &
+  published > (Sys.time() - mirror_lag_days * 86400)
+if (any(too_recent)) {
+  message(sprintf(
+    "Holding back %d package(s) published in the last %g day(s); the cran GitHub mirror will not have them yet: %s",
+    sum(too_recent),
+    mirror_lag_days,
+    paste(utils::head(cran_release$Package[too_recent], 10L), collapse = ", ")
+  ))
+}
+
 release_versions <- data.table(
-  Package = cran_release$Package,
-  Version = as.character(cran_release$Version)
+  Package = cran_release$Package[!too_recent],
+  Version = as.character(cran_release$Version[!too_recent])
 )
 
 pkgs_to_build <- unique(rbind(archive_versions, release_versions, fill = TRUE))

scripts/purge_cdn_zone.sh

@@ -47,12 +47,31 @@ resolve_zone_id() {
   fi
 
   response_file=$(mktemp)
-  curl -sS -o "${response_file}" \
-    -H "AccessKey: ${api_key}" \
-    "https://api.bunny.net/pullzone"
+  local status
+  status=$(
+    curl -sS -o "${response_file}" -w '%{http_code}' \
+      -H "AccessKey: ${api_key}" \
+      "https://api.bunny.net/pullzone?perPage=1000"
+  )
+
+  if [[ "${status}" != "200" ]]; then
+    echo "Listing BunnyCDN pull zones failed with HTTP ${status}:" >&2
+    head -c 500 "${response_file}" >&2
+    echo >&2
+    rm -f "${response_file}"
+    exit 1
+  fi
+
+  # The endpoint answers with a bare array on some accounts and a paginated
+  # object on others. `.Items // .` looks like it covers both but does not:
+  # indexing an array with a string is an *error*, and `//` only substitutes
+  # for null, so the array case aborted with
+  # "Cannot index array with string" and the zone was never purged.
   zone_id=$(
     jq -r --arg hostname "${zone}" \
-      '(.Items // .)[] | select(any(.Hostnames[]?; .Value == $hostname)) | .Id' \
+      '(if type == "object" then (.Items // []) else . end)[]
+         | select(any(.Hostnames[]?; .Value == $hostname))
+         | .Id' \
       "${response_file}"
   )
   rm -f "${response_file}"
@@ -62,6 +81,12 @@ resolve_zone_id() {
     exit 1
   fi
 
+  # Two zones sharing a hostname would purge only whichever jq emitted first.
+  if [[ $(wc -l <<<"${zone_id}") -gt 1 ]]; then
+    echo "Hostname ${zone} matched multiple pull zones: ${zone_id//$'\n'/ }" >&2
+    exit 1
+  fi
+
   echo "${zone_id}"
 }
 

scripts/verify-r-minor-routing.sh

@@ -0,0 +1,434 @@
+#!/usr/bin/env bash
+#
+# Verify per-R-minor index routing for cran.rpkgs.com across every published
+# <arch>/<os> slot.
+#
+# The edge router (edge/rpkgs-router.ts) rewrites PACKAGES* requests to
+# `contrib/<x.y>/` when the slot is listed in UNION_SLOTS and the client's
+# User-Agent carries an R minor. Two properties have to hold before a slot may
+# be added to UNION_SLOTS:
+#
+#   1. the per-minor index is a UNION of the per-minor and flat slots, so
+#      routing to it hides nothing the flat index carries; and
+#   2. every R minor a client might report resolves to an index that exists,
+#      because contribPath() does not check existence and has no fallback.
+#
+# Modes:
+#   (default)  Resolve routing decisions without depending on UNION_SLOTS being
+#              set. Safe to run before enabling: it reads the per-minor indexes
+#              directly and reproduces the router's target path.
+#   --live     Additionally drive the real CDN with R User-Agents and assert the
+#              bytes served match the expected index. Only meaningful once the
+#              slot is in UNION_SLOTS.
+#
+# Usage:
+#   scripts/verify-r-minor-routing.sh
+#   scripts/verify-r-minor-routing.sh --live
+#   MINORS="4.4 4.5" SAMPLE=10 scripts/verify-r-minor-routing.sh
+#
+# Exits non-zero if any check fails.
+
+set -uo pipefail
+
+BASE=${BASE:-https://cran.rpkgs.com}
+ARCHES=${ARCHES:-"amd64 arm64"}
+DISTROS=${DISTROS:-"resolute noble jammy rhel8 rhel9 rhel10 alpine323 alpine324"}
+# The supported window: the current R minor plus the two previous, matching
+# build-env-images' R_VERSION_LATEST/PREV1/PREV2 and cdn.tf's
+# local.rpkgs_supported_minors. Each of these must have a published index.
+MINORS=${MINORS:-"4.4 4.5 4.6"}
+# Minors we deliberately do not serve. These must have NO published index and,
+# once routing is live, must be sent to CRAN for sources rather than 404ing or
+# being handed binaries built under another minor.
+EXCLUDED_MINORS=${EXCLUDED_MINORS:-"4.3"}
+# How many Path: targets to HEAD-check per slot/minor. 0 disables.
+SAMPLE=${SAMPLE:-5}
+# Package-count shortfall against the best minor on the same slot, above which
+# coverage is reported as uneven. Reported, not failed on: the packages a
+# non-primary minor lacks are ABI-risky ones built under the primary minor,
+# which a client on another minor cannot safely load anyway, so their absence
+# is correct. This gates the *claim* ("full coverage for ABI-sensitive
+# packages"), not whether routing is safe to enable - MAX_REGRESSIONS does
+# that.
+PARITY_TOLERANCE=${PARITY_TOLERANCE:-25}
+# Source fallbacks are reported, not failed on. Since bincraft learned to keep
+# a matching-minor generic binary out of a fallback's shadow, a remaining
+# fallback means the generic slot's binary was built under a *different* minor,
+# which is unsafe for this client anyway: serving source there is correct, just
+# slow. The gate below is what actually matters.
+#
+# A REGRESSION is a package this client would receive as source through
+# per-minor routing but as a binary built under its own minor from the generic
+# slot. That is strictly worse than not routing at all, and must be zero before
+# a slot is added to UNION_SLOTS.
+MAX_REGRESSIONS=${MAX_REGRESSIONS:-0}
+LIVE=0
+
+for arg in "$@"; do
+  case "$arg" in
+    --live) LIVE=1 ;;
+    -h | --help)
+      sed -n '2,32p' "$0"
+      exit 0
+      ;;
+    *)
+      echo "unknown argument: $arg" >&2
+      exit 2
+      ;;
+  esac
+done
+
+WORK=$(mktemp -d)
+trap 'rm -rf "$WORK"' EXIT
+
+PASS=0
+FAIL=0
+FAILURES=""
+
+ok() {
+  PASS=$((PASS + 1))
+  printf '  ok    %s\n' "$1"
+}
+
+bad() {
+  FAIL=$((FAIL + 1))
+  FAILURES="${FAILURES}\n  - $1"
+  printf '  FAIL  %s\n' "$1"
+}
+
+# Fetch a URL into a file, echoing the HTTP status. Cached per URL.
+fetch() {
+  local url=$1 dest=$2 ua=${3:-}
+  if [ -s "$dest" ]; then
+    cat "$dest.status"
+    return 0
+  fi
+  local status
+  # -L: the router answers an index request with a redirect, so the bytes a
+  # client ends up with are only visible by following it.
+  #
+  # no-cache: a purge is asynchronous, so a run started right after a reindex
+  # otherwise measures whatever the edge still holds.
+  if [ -n "$ua" ]; then
+    status=$(curl -sSL -A "$ua" -H 'Cache-Control: no-cache' -o "$dest" -w '%{http_code}' --max-time 120 "$url" 2>/dev/null)
+  else
+    status=$(curl -sSL -H 'Cache-Control: no-cache' -o "$dest" -w '%{http_code}' --max-time 120 "$url" 2>/dev/null)
+  fi
+  echo "$status" > "$dest.status"
+  echo "$status"
+}
+
+head_status() {
+  curl -sS -o /dev/null -w '%{http_code}' -I --max-time 60 "$1" 2>/dev/null
+}
+
+# Package names from a gzipped PACKAGES index, sorted.
+pkg_names() {
+  gunzip -c "$1" 2>/dev/null | awk '/^Package:/ {print $2}' | sort -u
+}
+
+# "<steered> <source-fallbacks>" for a per-minor index: how many entries carry a
+# Path: field, and how many of those lack a Built: field (i.e. are sources).
+fallback_counts() {
+  gunzip -c "$1" 2>/dev/null | awk '
+    /^Package:/ { pkg = $2; path = ""; built = "" }
+    /^Path:/    { path = $2 }
+    /^Built:/   { built = $2 }
+    /^$/        { if (pkg != "" && path != "") { n++; if (built == "") s++ } pkg = "" }
+    END         { if (pkg != "" && path != "") { n++; if (built == "") s++ }
+                  printf "%d %d\n", n, s }
+  '
+}
+
+# Packages this index serves from the generic slot with a binary built under a
+# different R minor, while some other per-minor slot carries a build of them -
+# which proves the ABI classifier called them risky. Serving those is the
+# load-time crash the per-minor slots exist to prevent. bincraft drops them at
+# index time, so a non-zero count means the slot has not been reindexed since
+# that guard shipped.
+abi_unsafe_count() {
+  local minor_file=$1 minor=$2 risky_file=$3
+  gunzip -c "$minor_file" 2>/dev/null | awk -v m="$minor" '
+    /^Package:/ { pkg = $2; path = ""; built = "" }
+    /^Path:/    { path = $2 }
+    /^Built:/   { built = $2 " " $3 }
+    /^$/        { if (pkg != "" && path == "" && built != "" && built !~ ("^R " m "\\.")) print pkg; pkg = "" }
+  ' | sort -u > "$minor_file.mismatched"
+  comm -12 "$minor_file.mismatched" "$risky_file" | wc -l
+}
+
+# How many packages a client of <minor> would receive as source through
+# per-minor routing while the generic slot holds a binary built under that very
+# minor. Zero is the bar for enabling a slot.
+regression_count() {
+  local minor_file=$1 flat_file=$2 minor=$3
+  gunzip -c "$flat_file" 2>/dev/null | awk -v m="$minor" '
+    /^Package:/ { pkg = $2; built = "" }
+    /^Built:/   { built = $2 " " $3 }
+    /^$/        { if (pkg != "" && built ~ ("^R " m "\\.")) print pkg; pkg = "" }
+  ' | sort -u > "$minor_file.flatbin"
+  gunzip -c "$minor_file" 2>/dev/null | awk '
+    /^Package:/ { pkg = $2; path = ""; built = "" }
+    /^Path:/    { path = $2 }
+    /^Built:/   { built = $2 }
+    /^$/        { if (pkg != "" && path != "" && built == "") print pkg; pkg = "" }
+  ' | sort -u > "$minor_file.src"
+  comm -12 "$minor_file.src" "$minor_file.flatbin" | wc -l
+}
+
+# "<Package> <Path>" pairs for entries that carry a Path: field.
+path_entries() {
+  gunzip -c "$1" 2>/dev/null | awk '
+    /^Package:/ { pkg = $2; ver = ""; path = "" }
+    /^Version:/ { ver = $2 }
+    /^Path:/    { path = $2 }
+    /^$/        { if (pkg != "" && path != "") print pkg, ver, path; pkg = "" }
+    END         { if (pkg != "" && path != "") print pkg, ver, path }
+  '
+}
+
+# An R User-Agent of the shape R actually sends.
+r_user_agent() {
+  printf 'R/%s.0 (Ubuntu 24.04; codename=noble) (x86_64-pc-linux-gnu x86_64 linux-gnu)' "$1"
+}
+
+echo "verify-r-minor-routing: $BASE"
+echo "  slots:  $(echo "$ARCHES" | wc -w) arch x $(echo "$DISTROS" | wc -w) os"
+echo "  minors: $MINORS  (excluded: $EXCLUDED_MINORS)"
+echo "  live:   $LIVE"
+echo
+
+for arch in $ARCHES; do
+  for distro in $DISTROS; do
+    slot="$arch/$distro"
+    echo "$slot"
+
+    flat_url="$BASE/$slot/latest/src/contrib/PACKAGES.gz"
+    flat_file="$WORK/${arch}-${distro}-flat.gz"
+    flat_status=$(fetch "$flat_url" "$flat_file")
+
+    if [ "$flat_status" != "200" ]; then
+      bad "$slot flat index unreachable (HTTP $flat_status)"
+      continue
+    fi
+
+    pkg_names "$flat_file" > "$flat_file.names"
+    flat_count=$(wc -l < "$flat_file.names")
+    if [ "$flat_count" -lt 1000 ]; then
+      bad "$slot flat index has only $flat_count packages"
+      continue
+    fi
+    ok "$slot flat index: $flat_count packages"
+
+    for minor in $MINORS; do
+      minor_url="$BASE/$slot/latest/src/contrib/$minor/PACKAGES.gz"
+      minor_file="$WORK/${arch}-${distro}-${minor}.gz"
+      minor_status=$(fetch "$minor_url" "$minor_file")
+
+      # A minor the router would route to must exist, or clients on that R
+      # version get a 404 and see no packages at all.
+      if [ "$minor_status" != "200" ]; then
+        bad "$slot R $minor index missing (HTTP $minor_status) - routing would 404 for R $minor clients"
+        continue
+      fi
+
+      pkg_names "$minor_file" > "$minor_file.names"
+      minor_count=$(wc -l < "$minor_file.names")
+
+      # Union property: nothing the flat index carries may be missing here.
+      # bincraft deliberately drops an ABI-risky package whose only binary was
+      # built under another R minor: serving it is the load-time crash the
+      # per-minor slots exist to prevent. Those absences are correct.
+      #
+      # What must never go missing is a generic package built under *this*
+      # minor, which is safe to serve and has no reason to disappear.
+      comm -23 "$flat_file.names" "$minor_file.names" > "$minor_file.absent"
+      absent_count=$(wc -l < "$minor_file.absent")
+      gunzip -c "$flat_file" 2>/dev/null | awk -v m="$minor" '
+        /^Package:/ { pkg = $2; built = "" }
+        /^Built:/   { built = $2 " " $3 }
+        /^$/        { if (pkg != "" && built ~ ("^R " m "\\.")) print pkg; pkg = "" }
+      ' | sort -u > "$minor_file.flatsame"
+      lost=$(comm -12 "$minor_file.absent" "$minor_file.flatsame" | wc -l)
+
+      if [ "${lost:-0}" -ne 0 ]; then
+        bad "$slot R $minor index dropped $lost generic package(s) built under R $minor, which were safe to serve"
+      else
+        ok "$slot R $minor index: $minor_count packages, union holds ($absent_count ABI-unsafe dropped)"
+      fi
+
+      # A per-minor entry that is a source fallback resolves fine but makes the
+      # client compile. Routing to a slot that is mostly fallbacks does not
+      # deliver the binaries we advertise.
+      read -r steered fallbacks <<< "$(fallback_counts "$minor_file")"
+      if [ "${steered:-0}" -gt 0 ]; then
+        pct=$((fallbacks * 100 / steered))
+        printf '        note: %s/%s per-minor entries are source fallbacks (%s%%)\n' \
+          "$fallbacks" "$steered" "$pct"
+      fi
+
+      # The gate: nothing may arrive as source here that the generic slot would
+      # have served as a binary built under this same minor.
+      regressions=$(regression_count "$minor_file" "$flat_file" "$minor")
+      if [ "${regressions:-0}" -gt "$MAX_REGRESSIONS" ]; then
+        bad "$slot R $minor: $regressions packages would be served as source but exist as an R $minor binary in the generic slot"
+      else
+        ok "$slot R $minor: no regression against the generic slot"
+      fi
+
+      # Path: entries steer to per-minor binaries; they must resolve.
+      if [ "$SAMPLE" -gt 0 ]; then
+        path_entries "$minor_file" > "$minor_file.paths"
+        total_paths=$(wc -l < "$minor_file.paths")
+        broken=0
+        checked=0
+        while read -r pkg ver path; do
+          [ -z "${pkg:-}" ] && continue
+          tarball="$BASE/$slot/latest/src/contrib/$path/${pkg}_${ver}.tar.gz"
+          status=$(head_status "$tarball")
+          checked=$((checked + 1))
+          if [ "$status" != "200" ]; then
+            broken=$((broken + 1))
+            [ "$broken" -le 2 ] && printf '        broken target: %s (HTTP %s)\n' "$tarball" "$status"
+          fi
+        done < <(shuf -n "$SAMPLE" "$minor_file.paths" 2>/dev/null || head -n "$SAMPLE" "$minor_file.paths")
+
+        if [ "$broken" -ne 0 ]; then
+          bad "$slot R $minor: $broken/$checked sampled Path: targets do not resolve (of $total_paths total)"
+        elif [ "$checked" -gt 0 ]; then
+          ok "$slot R $minor: $checked/$checked sampled Path: targets resolve (of $total_paths total)"
+        fi
+      fi
+
+      # Live routing: what a real R client on this minor actually receives.
+      if [ "$LIVE" -eq 1 ]; then
+        ua=$(r_user_agent "$minor")
+        live_file="$WORK/${arch}-${distro}-${minor}-live.gz"
+        live_status=$(fetch "$flat_url" "$live_file" "$ua")
+        if [ "$live_status" != "200" ]; then
+          bad "$slot R $minor live request failed (HTTP $live_status)"
+        elif cmp -s "$live_file" "$minor_file"; then
+          ok "$slot R $minor live request served the per-minor index"
+        elif cmp -s "$live_file" "$flat_file"; then
+          bad "$slot R $minor live request served the FLAT index - slot not in UNION_SLOTS?"
+        else
+          bad "$slot R $minor live request served neither the per-minor nor the flat index"
+        fi
+      fi
+    done
+
+    # Coverage parity across minors. The union property only guarantees no
+    # client loses packages relative to the flat index; it says nothing about a
+    # 4.4 client seeing fewer packages than a 4.5 client on the same slot.
+    best=0
+    for minor in $MINORS; do
+      f="$WORK/${arch}-${distro}-${minor}.gz.names"
+      [ -s "$f" ] || continue
+      c=$(wc -l < "$f")
+      [ "$c" -gt "$best" ] && best=$c
+    done
+    if [ "$best" -gt 0 ]; then
+      uneven=""
+      for minor in $MINORS; do
+        f="$WORK/${arch}-${distro}-${minor}.gz.names"
+        [ -s "$f" ] || continue
+        c=$(wc -l < "$f")
+        gap=$((best - c))
+        [ "$gap" -gt "$PARITY_TOLERANCE" ] && uneven="$uneven R$minor:-$gap"
+      done
+      if [ -n "$uneven" ]; then
+        printf '        note: %s coverage uneven across minors (vs best %s):%s\n' \
+          "$slot" "$best" "$uneven"
+      else
+        ok "$slot coverage parity across minors (best $best, all within $PARITY_TOLERANCE)"
+      fi
+    fi
+
+    # Packages carrying a Path in any per-minor index are risky by construction.
+    : > "$WORK/${arch}-${distro}.risky"
+    for minor in $MINORS; do
+      f="$WORK/${arch}-${distro}-${minor}.gz"
+      [ -s "$f" ] || continue
+      gunzip -c "$f" 2>/dev/null | awk '
+        /^Package:/ { pkg = $2; path = "" }
+        /^Path:/    { path = $2 }
+        /^$/        { if (pkg != "" && path != "") print pkg; pkg = "" }
+      ' >> "$WORK/${arch}-${distro}.risky"
+    done
+    sort -u -o "$WORK/${arch}-${distro}.risky" "$WORK/${arch}-${distro}.risky"
+
+    for minor in $MINORS; do
+      f="$WORK/${arch}-${distro}-${minor}.gz"
+      [ -s "$f" ] || continue
+      unsafe=$(abi_unsafe_count "$f" "$minor" "$WORK/${arch}-${distro}.risky")
+      if [ "${unsafe:-0}" -gt 0 ]; then
+        bad "$slot R $minor serves $unsafe ABI-risky package(s) built under another R minor - reindex this slot"
+      else
+        ok "$slot R $minor serves no ABI-risky package from another minor"
+      fi
+    done
+
+    # Excluded minors: no published index, and under --live a redirect to CRAN.
+    for minor in $EXCLUDED_MINORS; do
+      ex_url="$BASE/$slot/latest/src/contrib/$minor/PACKAGES.gz"
+      ex_status=$(curl -sS -o /dev/null -w '%{http_code}' --max-time 60 "$ex_url" 2>/dev/null)
+      if [ "$ex_status" = "200" ]; then
+        bad "$slot R $minor is excluded but an index is published - the two lists disagree"
+      else
+        ok "$slot R $minor correctly has no published index"
+      fi
+
+      if [ "$LIVE" -eq 1 ]; then
+        loc=$(curl -sS -o /dev/null -w '%{redirect_url}' -A "$(r_user_agent "$minor")" \
+          --max-time 60 "$flat_url" 2>/dev/null)
+        case "$loc" in
+          https://cran.r-project.org/*)
+            ok "$slot R $minor is sent to CRAN ($loc)"
+            ;;
+          "")
+            bad "$slot R $minor was served directly instead of being sent to CRAN"
+            ;;
+          *)
+            bad "$slot R $minor redirected somewhere unexpected: $loc"
+            ;;
+        esac
+      fi
+    done
+
+    # A client whose User-Agent carries no R version must keep getting the flat
+    # index, never a per-minor one.
+    if [ "$LIVE" -eq 1 ]; then
+      plain_file="$WORK/${arch}-${distro}-plain.gz"
+      plain_status=$(fetch "$flat_url" "$plain_file" "curl/8.0.0")
+      if [ "$plain_status" != "200" ]; then
+        bad "$slot non-R User-Agent request failed (HTTP $plain_status)"
+      elif cmp -s "$plain_file" "$flat_file"; then
+        ok "$slot non-R User-Agent still served the flat index"
+      else
+        bad "$slot non-R User-Agent was routed away from the flat index"
+      fi
+
+      # Tarball requests must never be rewritten into a per-minor directory:
+      # flat-slot packages do not live there.
+      sample_pkg=$(gunzip -c "$flat_file" | awk '/^Package:/ {p=$2} /^Version:/ {print p, $2; exit}')
+      if [ -n "$sample_pkg" ]; then
+        # shellcheck disable=SC2086 # deliberate split into $1 (package) and $2 (version)
+        set -- $sample_pkg
+        tb="$BASE/$slot/latest/src/contrib/${1}_${2}.tar.gz"
+        tb_status=$(curl -sS -o /dev/null -w '%{http_code}' -A "$(r_user_agent 4.5)" --max-time 60 "$tb" 2>/dev/null)
+        if [ "$tb_status" = "200" ]; then
+          ok "$slot tarball request under an R User-Agent still resolves"
+        else
+          bad "$slot tarball ${1}_${2}.tar.gz broke under an R User-Agent (HTTP $tb_status)"
+        fi
+      fi
+    fi
+  done
+done
+
+echo
+echo "passed: $PASS   failed: $FAIL"
+if [ "$FAIL" -ne 0 ]; then
+  printf 'failures:%b\n' "$FAILURES"
+  exit 1
+fi