diff --git a/CHANGELOG.md b/CHANGELOG.md
index 880b569..6dab17b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,18 @@ new version heading in the same commit.
 
 ## [Unreleased]
 
+## [0.281.5] — 2026-07-31
+### Fixed
+- **The approval-friction signal no longer divides by a denominator that isn't there.** v0.280.0 started
+  recording the `approved` count so friction could be a rate; per-pass entries written before that carry
+  rejections with **no denominator**, and `recentTally` summed them anyway — so a window of mostly-legacy
+  entries read as ~100% rejection. Live instapods was nagging every agent that "recent actions were
+  rejected at human approval" while its true 7-day rate was **5.4%** (35 approved / 2 rejected). Entries
+  lacking the denominator now contribute to neither side of the ratio (their budget/error signals still
+  count), so the signal stays quiet until real evidence exists rather than inventing it. Verified against
+  both live tenants' actual window shapes: the false instapods nag stops; genuine friction with a real
+  sample still fires.
+
 ## [0.281.4] — 2026-07-31
 ### Fixed
 - **Bump `TOPICS_VERSION` to 3.** v0.281.3 tightened the topic extractor (opaque hex ids) without bumping
diff --git a/package-lock.json b/package-lock.json
index 6753a20..44fe036 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,12 +1,12 @@
 {
   "name": "agent-os",
-  "version": "0.281.4",
+  "version": "0.281.5",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "agent-os",
-      "version": "0.281.4",
+      "version": "0.281.5",
       "license": "MIT",
       "bin": {
         "agent-os": "bin/agent-os"
diff --git a/package.json b/package.json
index a0c1b83..c6c5f0a 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "agent-os",
-  "version": "0.281.4",
+  "version": "0.281.5",
   "description": "A generic, governed operating system for running autonomous agents safely across brands. Ships with a local web console.",
   "license": "MIT",
   "type": "commonjs",
diff --git a/src/edge/dreaming.ts b/src/edge/dreaming.ts
index 1090a43..8a88807 100644
--- a/src/edge/dreaming.ts
+++ b/src/edge/dreaming.ts
@@ -464,8 +464,14 @@ function recentTally(s: DreamState): { sessions: number; success: number; approv
   for (const e of (s.recent ?? []).slice(0, RECENT_PASSES)) {
     r.sessions += e.sessions || 0;
     r.success += e.success || 0;
-    r.approved += e.approved || 0;
-    r.rejected += e.rejected || 0;
+    // Approval friction is a RATIO, so an entry that predates the `approved` denominator must contribute
+    // NEITHER side — counting its rejections against a missing denominator reads as 100% rejection. Live
+    // instapods computed 22% (and nagged every agent) while its true rate was 5.4%, because six of seven
+    // window entries were legacy. Skipping them means the signal stays quiet until real evidence exists.
+    if (e.approved !== undefined) {
+      r.approved += e.approved;
+      r.rejected += e.rejected || 0;
+    }
     r.budgetStops += e.budgetStops || 0;
     r.errors += e.errors || 0;
   }