diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index ea325467..5cf29a20 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -79,9 +79,9 @@ jobs:
         # us from having to retrofit it later.
         run: go test -race -cover ./...
 
-      - name: Coverage floor (internal/cli, internal/submit must not rot)
+      - name: Coverage floor (cli, submit, push, cluster must not rot)
         # `go test -cover` above prints numbers but asserts nothing. This
-        # enforces a per-package floor on the two load-bearing, historically
+        # enforces a per-package floor on the load-bearing, historically
         # thin-tested packages (the money path + submit orchestration) so a
         # test deletion can't silently drop coverage. Floors ratchet UP only —
         # see scripts/coverage-floor.sh (backend#1009).
diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml
index de54c5c3..1a9eada2 100644
--- a/.github/workflows/e2e.yml
+++ b/.github/workflows/e2e.yml
@@ -7,11 +7,20 @@ name: Integration (e2e)
 # 0% in unit coverage — this is where every live bug actually lived).
 #
 # Heavy (kind boot + image pulls), so it does NOT run on every PR:
+#   - on every push to develop
 #   - nightly (schedule)
 #   - manual (workflow_dispatch)
 #   - on a PR only when it carries the `e2e` label
+#
+# The push trigger exists because scheduled workflows always execute the
+# DEFAULT branch's tree (main), which trails develop by dozens of commits —
+# discovery_e2e_test.go, delete_e2e_test.go, and the delete-teardown job
+# don't exist on main, so the nightly never actually ran them (#293).
+# Running on develop pushes exercises the tree the tests were written for.
 
 on:
+  push:
+    branches: [develop]
   schedule:
     - cron: "0 3 * * *"
   workflow_dispatch:
diff --git a/scripts/coverage-floor.sh b/scripts/coverage-floor.sh
index e9775f4c..723cfd34 100755
--- a/scripts/coverage-floor.sh
+++ b/scripts/coverage-floor.sh
@@ -10,7 +10,13 @@
 # The floors are a RATCHET: set just under the current numbers, and bumped UP
 # as coverage improves — never silently down. Lowering a floor must be a
 # deliberate, reviewed edit here (with a reason), not a side effect of deleting
-# tests. Current (develop): internal/cli ~70%, internal/submit ~75%.
+# tests. Current (develop, 2026-07-14, ubuntu CI runner): internal/cli 82.9%,
+# internal/submit 80.4%, internal/push 89.0%, internal/cluster 74.6%.
+#
+# NOTE: internal/cluster measures higher on a dev machine with a real
+# ~/.kube/config (78.5% on macOS) than on the bare CI runner (74.6%) — the
+# kubeconfig-resolution paths only execute where one exists. Floors must be
+# set against the CI number, since that's where the gate runs.
 #
 # Usage: scripts/coverage-floor.sh   (run from the repo root)
 #
@@ -20,8 +26,10 @@ set -euo pipefail
 # "package:floor" entries. Keep floors integers; coverage is compared as a
 # float against them.
 FLOORS="
-internal/cli:68
-internal/submit:72
+internal/cli:80
+internal/submit:78
+internal/push:87
+internal/cluster:73
 "
 
 status=0