diff --git a/.github/actions/setup-sccache/action.yml b/.github/actions/setup-sccache/action.yml
index 802029107..a17ae66a3 100644
--- a/.github/actions/setup-sccache/action.yml
+++ b/.github/actions/setup-sccache/action.yml
@@ -3,7 +3,8 @@ description: >
   Install sccache and select a backend. GCS on ARC when GCE metadata
   identity is present and an objects.list probe of evalops-prod-ci-sccache
   succeeds. Else local disk when the runner service exports
-  SCCACHE_LOCAL_DIR to an absolute writable path (WSL:
+  SCCACHE_LOCAL_DIR to an absolute writable path, or Borg exports its owned
+  BORG_CACHE_ROOT/sccache as SCCACHE_DIR (WSL:
   /var/cache/evalops-sccache, SCCACHE_CACHE_SIZE default 40G). GitHub
   Actions cache is an opt-out fallback for callers that still require it.
   Callers compile with RUSTC_WRAPPER=sccache.
@@ -18,7 +19,7 @@ inputs:
       sccache storage backend. `auto` (default) probes the GCE metadata
       server and uses `gcs` when the runner identity answers and an
       objects.list probe of the sccache prefix returns HTTP 200; else
-      `local` when SCCACHE_LOCAL_DIR is an absolute writable directory;
+      `local` when SCCACHE_LOCAL_DIR or Borg's owned SCCACHE_DIR is writable;
       else `gha`. The GCS probe is GET .../b/<bucket>/o?maxResults=1&prefix=...
       (storage.objects.list), not buckets.get, because the runner GSA
       is roles/storage.objectUser. A 404 or 403 warns and falls back to
diff --git a/.github/actions/setup-sccache/configure.sh b/.github/actions/setup-sccache/configure.sh
index 6bb42b0bb..60e3c2434 100755
--- a/.github/actions/setup-sccache/configure.sh
+++ b/.github/actions/setup-sccache/configure.sh
@@ -3,8 +3,8 @@
 # Precedence: GCS on ARC when GCE metadata identity is present AND an
 # objects.list probe against the bucket prefix succeeds; else local disk.
 # GCP runners never fall through to GitHub's Azure-hosted cache data plane.
-# Off-GCP runners use SCCACHE_LOCAL_DIR when it is an absolute writable
-# directory (WSL: /var/cache/evalops-sccache, 40G by default). GHA is only
+# Off-GCP runners use SCCACHE_LOCAL_DIR or Borg's owned SCCACHE_DIR when it
+# is absolute and writable (WSL: /var/cache/evalops-sccache, 40G default). GHA is only
 # selected when the caller explicitly permits that fallback.
 # Hard-fail only if GCS or local disk was selected and sccache then reports
 # a different Cache location.
@@ -77,6 +77,12 @@ gce_identity="${gce_identity//$'\n'/}"
 # runner service environment. Ephemeral ARC pods do not; they keep GCS/GHA.
 # Default size cap 40G; override with SCCACHE_CACHE_SIZE on the host.
 SCCACHE_LOCAL_DIR="${SCCACHE_LOCAL_DIR:-}"
+if [[ -z "${SCCACHE_LOCAL_DIR}" && -n "${BORG_CACHE_ROOT:-}" \
+  && "${SCCACHE_DIR:-}" == "${BORG_CACHE_ROOT}/sccache" ]]; then
+  # Borg already owns and probes this persistent cache. Keep explicit local
+  # configuration first, and retain the existing writable/backend checks.
+  SCCACHE_LOCAL_DIR="${SCCACHE_DIR}"
+fi
 SCCACHE_CACHE_SIZE="${SCCACHE_CACHE_SIZE:-40G}"
 
 local_disk_ready() {
diff --git a/.repository-projection.json b/.repository-projection.json
index 32af10b5e..a8322be40 100644
--- a/.repository-projection.json
+++ b/.repository-projection.json
@@ -3,11 +3,11 @@
   "projection": "deixic-code",
   "projectionSchemaVersion": 1,
   "sourceRepository": "dx-corp/mono",
-  "sourceSha": "b9c1342ea2f64a75617cc18edfc77c674f24e542",
+  "sourceSha": "2feea8e8f1e564f1929d91a082f9040f4b87c5b7",
   "destinationRepository": "dx-corp/code",
-  "priorProjectedBase": "97792e942d877f788c43de0c0d3d48fbbc11c502",
+  "priorProjectedBase": "65cad020159af5708a765ef30ca18480cd913292",
   "definitionDigest": "cb9d429542ebb0a2de9b42a7aad60d9d8696a648ceba47c30f05c0b285ca0db7",
   "toolDigest": "5b23a8f676ff72aaeedce0075b6014d88b01f010343c2822ea47fd9955a98726",
-  "contentDigest": "ba1011690d1266129cf0e4e26b8a4c8ca458ea0d21a9602fd6beb57e40513b67",
+  "contentDigest": "bc5029b7d1003d5bbc74a27ff68b7a47eaaaf43bbca829be08abd47e6b798733",
   "publicationEligible": true
 }
diff --git a/vendor/dex-loop/tests/scenarios.rs b/vendor/dex-loop/tests/scenarios.rs
index 59e6efeb7..6ddc72c6b 100644
--- a/vendor/dex-loop/tests/scenarios.rs
+++ b/vendor/dex-loop/tests/scenarios.rs
@@ -66,7 +66,7 @@ async fn plain_answer_streams_text_then_final() {
 
 // 2. Three reads run as one wave: they overlap, take about one tool delay, and
 // their results enter history in call order although they finish in reverse.
-#[tokio::test]
+#[tokio::test(start_paused = true)]
 async fn read_only_wave_overlaps_and_keeps_call_order() {
     let log = FakeLog::default();
     let model = FakeModel::new(vec![
@@ -85,7 +85,7 @@ async fn read_only_wave_overlaps_and_keeps_call_order() {
     let engine = engine(&log, &model, &tools, budget());
     let mut ctx = log.start_turn("t1", "look");
 
-    let started = Instant::now();
+    let started = tokio::time::Instant::now();
     let exit = engine.run(&mut ctx, &CancellationToken::new()).await;
     let elapsed = started.elapsed();
 
@@ -128,7 +128,7 @@ async fn read_only_wave_overlaps_and_keeps_call_order() {
 
 // 3. A mutation waits for the wave before it; a read after it starts a new
 // wave. The mutation is claimed in the ledger and its outcome recorded.
-#[tokio::test]
+#[tokio::test(start_paused = true)]
 async fn mutating_call_runs_serially_after_the_wave() {
     let log = FakeLog::default();
     let model = FakeModel::new(vec![
diff --git a/vendor/dex-loop/tests/scenarios/capacity_controls.rs b/vendor/dex-loop/tests/scenarios/capacity_controls.rs
index 1704c9d03..5d3cff6e1 100644
--- a/vendor/dex-loop/tests/scenarios/capacity_controls.rs
+++ b/vendor/dex-loop/tests/scenarios/capacity_controls.rs
@@ -278,7 +278,7 @@ async fn explicit_policy_controls_proactive_compaction_but_allows_hard_recovery(
     }
 }
 
-#[tokio::test]
+#[tokio::test(start_paused = true)]
 async fn legacy_and_disabled_policies_do_not_wait_for_unused_pressure_reads() {
     struct SlowPressure {
         inner: FakeModel,