diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 0b65d9a..6f480c1 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -6,6 +6,14 @@ on:
   pull_request:
     types: [opened, synchronize, reopened]
 
+concurrency:
+  # Supersede an in-flight run for the SAME PR. Push runs deliberately get a
+  # unique group (run_id): a shared group keeps only one pending run, so rapid
+  # pushes to main would evict an intermediate commit's queued run before it
+  # ever ran, leaving that commit unverified.
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.run_id }}
+  cancel-in-progress: true
+
 jobs:
   detect-changes:
     name: Detect Changed Paths
diff --git a/.github/workflows/issue-triage.yml b/.github/workflows/issue-triage.yml
index 24ebbca..55f3dbf 100644
--- a/.github/workflows/issue-triage.yml
+++ b/.github/workflows/issue-triage.yml
@@ -15,6 +15,10 @@ jobs:
         with:
           script: |
             const labelName = 'needs triage';
+            // Returns true when the label was ABSENT at the moment this run
+            // looked - whether this run created it or lost the race to a
+            // concurrent one. That is exactly the condition under which GitHub
+            // already dropped it from the opening issue's template frontmatter.
             const ensureLabel = async (name, color, description) => {
               try {
                 await github.rest.issues.getLabel({
@@ -22,6 +26,7 @@ jobs:
                   repo: context.repo.repo,
                   name,
                 });
+                return false;
               } catch {
                 try {
                   await github.rest.issues.createLabel({
@@ -35,7 +40,11 @@ jobs:
                   if (createError.status !== 422) {
                     throw createError;
                   }
+                  // 422 means a concurrent run created it first. The label was
+                  // still missing when THIS run looked, so this run's issue had
+                  // its template label dropped too and needs the same recovery.
                 }
+                return true;
               }
             };
             const applyLabelIfMissing = async (name) => {
@@ -49,8 +58,19 @@ jobs:
                 });
               }
             };
-            await ensureLabel(labelName, 'ededed', 'Not yet reviewed or categorized');
-            await applyLabelIfMissing(labelName);
+            // Every issue template already sets `needs triage` in its frontmatter,
+            // so applying it here unconditionally was pure duplication. ensureLabel
+            // must stay: GitHub silently drops a template label that does not exist
+            // in the repo, so this call is what creates it.
+            //
+            // The one case that still needs an explicit apply is the run that
+            // creates the label: the template's request was already dropped, and
+            // creating the label now does not retroactively label the issue that
+            // triggered this run. Every later issue gets it from its template.
+            const createdTriageLabel = await ensureLabel(labelName, 'ededed', 'Not yet reviewed or categorized');
+            if (createdTriageLabel) {
+              await applyLabelIfMissing(labelName);
+            }
             if (!context.payload.issue.assignee) {
               await github.rest.issues.addAssignees({
                 owner: context.repo.owner,
diff --git a/.github/workflows/pr-comment.yml b/.github/workflows/pr-comment.yml
index 64a237d..b8a8362 100644
--- a/.github/workflows/pr-comment.yml
+++ b/.github/workflows/pr-comment.yml
@@ -22,6 +22,20 @@ on:
     workflows: ["Build"]
     types: [completed]
 
+# Serialize report runs for the same PR branch. The sticky-comment update is a
+# lookup-then-create: two Build runs completing close together would both find
+# no marker and both post, producing exactly the duplicate comments the marker
+# exists to prevent. GitHub offers no atomic comment upsert, so serializing per
+# PR is what makes the pair safe. Keyed on the trusted workflow_run head fields
+# because the PR number is not resolved until a step below.
+#
+# cancel-in-progress is false so an in-flight post is never killed midway. A
+# queued run can still be superseded by a newer one, which is correct here: the
+# newest report is the one worth posting.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.workflow_run.head_repository.full_name }}-${{ github.event.workflow_run.head_branch }}
+  cancel-in-progress: false
+
 permissions:
   contents: read
   actions: read # download artifacts + read job conclusions from the run
@@ -185,13 +199,24 @@ jobs:
           # The valgrind job prints ">>> valgrind <bin>" per test binary and
           # "::error::Valgrind reported ..." on a failing one. Count those markers
           # (integers only) - no fork string reaches the rendered body.
+          # A log with no ">>> valgrind" markers means the job changed its output
+          # format: say so in the comment. Dropping the chart silently is how a
+          # broken report goes unnoticed. The note is a trusted literal - no
+          # fork-controlled text reaches the body through this path.
           emit_valgrind_pie() {
             local logf="$1" total fails clean
             total="$(grep -Ec '^>>> valgrind ' "$logf" || true)"
             fails="$(grep -Ec '^::error::Valgrind reported' "$logf" || true)"
-            { [[ "$total" =~ ^[0-9]+$ ]] && [[ "$fails" =~ ^[0-9]+$ ]] && [ "$total" -gt 0 ] && [ "$fails" -le "$total" ]; } || return 0
+            if ! { [[ "$total" =~ ^[0-9]+$ ]] && [[ "$fails" =~ ^[0-9]+$ ]] && [ "$total" -gt 0 ] && [ "$fails" -le "$total" ]; }; then
+              printf '_Chart unavailable: no ">>> valgrind" markers found in the log._\n\n'
+              return 0
+            fi
             clean=$(( total - fails ))
             printf '```mermaid\n'
+            # Mermaid's default pie palette is dark blue/purple, which reads as
+            # arbitrary. Force clean=green, leaking=red: pie1 colors the first
+            # slice below, pie2 the second.
+            printf '%s\n' "%%{init: {'themeVariables': {'pie1': '#2da44e', 'pie2': '#cf222e'}}}%%"
             printf 'pie showData\n'
             printf '    title Test binaries under Valgrind (%s total)\n' "$total"
             printf '    "Clean" : %s\n' "$clean"
@@ -199,6 +224,10 @@ jobs:
             printf '```\n\n'
           }
 
+          # Identifies our own report comment so each push updates it in place
+          # instead of appending one per push. Invisible in rendered markdown.
+          MARKER="<!-- mux-ci-report -->"
+
           : > body.md
           wrote=0
           for spec in "${specs[@]}"; do
@@ -248,4 +277,17 @@ jobs:
             exit 0
           fi
           printf '_Commit `%s` - [full run](%s)_\n' "$HEAD_SHA" "$RUN_URL" >> body.md
-          gh pr comment "$PR" --repo "$REPO" --body-file body.md
+          printf '\n%s\n' "$MARKER" >> body.md
+
+          # Update our own previous comment instead of stacking one per push. The
+          # author filter is load-bearing: anyone can post a comment containing
+          # the marker, and only a github-actions[bot] comment is ours to edit.
+          existing="$(gh api --paginate --slurp "repos/$REPO/issues/$PR/comments?per_page=100" \
+            | jq -r --arg m "$MARKER" \
+                '[.[][] | select(.user.login == "github-actions[bot]" and (.body | contains($m)))] | .[0].id // empty')"
+          if [ -n "$existing" ]; then
+            gh api -X PATCH "repos/$REPO/issues/comments/$existing" -F body=@body.md --silent
+            echo "Updated report comment $existing on PR #$PR."
+          else
+            gh pr comment "$PR" --repo "$REPO" --body-file body.md
+          fi