diff --git a/.github/workflows/copilot-response.yml b/.github/workflows/copilot-response.yml
index 3f6cd61..16df316 100644
--- a/.github/workflows/copilot-response.yml
+++ b/.github/workflows/copilot-response.yml
@@ -1,6 +1,9 @@
 name: Copilot review response
-# Fires when Copilot submits a review on a Dependabot PR the AI upgrade
-# pipeline has already finished (label ai-complete). One Claude round
+# Responds to a Copilot review on a Dependabot PR the AI upgrade pipeline
+# has already finished (label ai-complete): started by the org dispatcher
+# (v6 callers, pr-number/review-id inputs) or, for v5 callers, by the
+# held-then-released pull_request_review run. Each review is answered at
+# most once (review= in the ai-meta marker). One Claude round
 # verifies each Copilot finding as an UNTRUSTED CLAIM — fixing what is real,
 # rejecting what is not, with evidence — then a deterministic job pushes via
 # the validated-push action, replies to every thread with its verdict,
@@ -27,6 +30,19 @@ on:
         description: "Bot actors allowed to trigger the Claude step"
         type: string
         default: talieisin-ai-upgrade-trigger
+      # Dispatch mode (bootstrap v6 callers): the org dispatcher starts the
+      # round via workflow_dispatch instead of the caller subscribing to
+      # pull_request_review, which GitHub holds at action_required on every
+      # PR Copilot reviews. Identifiers only — the gate re-reads the rest.
+      # Both empty = review-event mode (v5 callers).
+      pr-number:
+        description: "PR to respond on (dispatch mode); empty = take it from the pull_request_review event"
+        type: string
+        default: ""
+      review-id:
+        description: "Copilot review to respond to (dispatch mode); empty = take it from the pull_request_review event"
+        type: string
+        default: ""
     secrets:
       CLAUDE_CODE_OAUTH_TOKEN:
         required: true
@@ -36,11 +52,14 @@ on:
         required: true
 
 concurrency:
-  group: copilot-response-${{ github.repository }}-${{ github.event.pull_request.number }}
+  group: copilot-response-${{ github.repository }}-${{ inputs.pr-number || github.event.pull_request.number }}
   cancel-in-progress: false
 
 env:
   STICKY_MARKER: "<!-- talieisin-ai-upgrade -->"
+  # The automation app's login: dispatch-mode actor, and the only author
+  # whose ai-meta review= markers count as "answered".
+  AUTOMATION_BOT: "talieisin-org-automation[bot]"
 
 jobs:
   gate:
@@ -49,27 +68,50 @@ jobs:
     permissions:
       contents: read
       pull-requests: read
+      issues: read # answered-marker lookup on the PR's comments
     outputs:
       verdict: ${{ steps.checks.outputs.verdict }}
+      pr-number: ${{ steps.checks.outputs.pr-number }}
+      review-id: ${{ steps.checks.outputs.review-id }}
       head-sha: ${{ steps.checks.outputs.head-sha }}
-      head-ref: ${{ github.event.pull_request.head.ref }}
+      head-ref: ${{ steps.checks.outputs.head-ref }}
+      base-ref: ${{ steps.checks.outputs.base-ref }}
       findings: ${{ steps.checks.outputs.findings }}
     steps:
       - name: Provenance checks and findings extraction
         id: checks
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          REVIEWER: ${{ github.event.review.user.login }}
-          REVIEW_ID: ${{ github.event.review.id }}
-          PR_AUTHOR: ${{ github.event.pull_request.user.login }}
-          PR_STATE: ${{ github.event.pull_request.state }}
-          HEAD_REPO: ${{ github.event.pull_request.head.repo.full_name }}
           BASE_REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ github.event.pull_request.number }}
-          LABELS: ${{ join(github.event.pull_request.labels.*.name, ',') }}
+          IN_PR: ${{ inputs.pr-number }}
+          IN_REVIEW: ${{ inputs.review-id }}
+          EVENT_PR: ${{ github.event.pull_request.number }}
+          EVENT_REVIEW: ${{ github.event.review.id }}
         run: |
           set -euo pipefail
           skip() { echo "verdict=skip" >> "$GITHUB_OUTPUT"; echo "findings=0" >> "$GITHUB_OUTPUT"; echo "head-sha=" >> "$GITHUB_OUTPUT"; echo "note: $1"; exit 0; }
+          # Dispatch mode (inputs set) or review-event mode (v5 callers).
+          # Either way the inputs are identifiers only: everything the
+          # checks below judge is read back from the API, never trusted
+          # from the dispatch or the event payload.
+          if [ -n "$IN_PR$IN_REVIEW" ]; then
+            PR_NUMBER=$IN_PR REVIEW_ID=$IN_REVIEW
+          else
+            PR_NUMBER=$EVENT_PR REVIEW_ID=$EVENT_REVIEW
+          fi
+          [[ "$PR_NUMBER" =~ ^[1-9][0-9]*$ && "$REVIEW_ID" =~ ^[1-9][0-9]*$ ]] \
+            || skip "no valid PR number and review id (pr='$PR_NUMBER' review='$REVIEW_ID')"
+          echo "pr-number=$PR_NUMBER" >> "$GITHUB_OUTPUT"
+          echo "review-id=$REVIEW_ID" >> "$GITHUB_OUTPUT"
+          # Scoped to the PR, so a review from another PR is a 404 (fails
+          # the job — the dispatcher's attempt cap bounds any retry).
+          gh api "repos/$BASE_REPO/pulls/$PR_NUMBER" > pr.json
+          gh api "repos/$BASE_REPO/pulls/$PR_NUMBER/reviews/$REVIEW_ID" > review.json
+          REVIEWER=$(jq -r '.user.login' review.json)
+          PR_AUTHOR=$(jq -r '.user.login' pr.json)
+          PR_STATE=$(jq -r '.state' pr.json)
+          HEAD_REPO=$(jq -r '.head.repo.full_name // ""' pr.json)
+          LABELS=$(jq -r '[.labels[].name] | join(",")' pr.json)
           case "$REVIEWER" in
             copilot-pull-request-reviewer*|github-copilot*|Copilot) : ;;
             *) skip "review author '$REVIEWER' is not Copilot" ;;
@@ -81,6 +123,26 @@ jobs:
             *,ai-complete,*) : ;;
             *) skip "PR is not in ai-complete state (labels: $LABELS) — the pipeline responds only on finished PRs" ;;
           esac
+          # Findings are line-anchored to the reviewed commit; a round that
+          # starts hours later (dispatcher cadence, stale-green refreshes)
+          # must not apply them to a head that has since moved. The
+          # dispatcher selects on the same predicate, so this never loops.
+          REVIEW_SHA=$(jq -r '.commit_id' review.json)
+          HEAD_SHA=$(jq -r '.head.sha' pr.json)
+          [ "$REVIEW_SHA" = "$HEAD_SHA" ] \
+            || skip "review $REVIEW_ID is on $REVIEW_SHA but the head is now $HEAD_SHA — request a new Copilot review"
+          # Answered at most once. push-response stamps review=<id> into
+          # the ai-meta marker of every comment it writes; only the
+          # automation app's own comments count, so nobody can suppress a
+          # round by pasting a marker. Retry = request a new Copilot review.
+          # Read to a file first: a failed read must fail the job, not
+          # pass as "unanswered" (and grep -q in a pipefail pipeline would
+          # SIGPIPE gh and turn a match into a miss).
+          gh api "repos/$BASE_REPO/issues/$PR_NUMBER/comments?per_page=100" --paginate \
+            --jq ".[] | select(.user.login == \"$AUTOMATION_BOT\") | .body" > own-comments.txt
+          if grep -Eq "^<!-- ai-meta v=1 [^>]* review=$REVIEW_ID " own-comments.txt; then
+            skip "review $REVIEW_ID has already been answered"
+          fi
           # Deterministic findings from the review's inline comments (a
           # quota-failure or comment-free review yields zero and we skip).
           gh api "repos/$BASE_REPO/pulls/$PR_NUMBER/reviews/$REVIEW_ID/comments?per_page=100" \
@@ -88,15 +150,26 @@ jobs:
             > findings.json
           n=$(jq length findings.json)
           [ "$n" -gt 0 ] || skip "review has no inline comments — nothing to respond to"
+          # Also answered if the automation app already replied to any of
+          # this review's inline comments: push-response posts those replies
+          # BEFORE the marked summary comment, whose write is best-effort.
+          gh api "repos/$BASE_REPO/pulls/$PR_NUMBER/comments?per_page=100" --paginate \
+            --jq ".[] | select(.user.login == \"$AUTOMATION_BOT\") | .in_reply_to_id // empty" > own-replies.txt
+          jq -r '.[].comment_id' findings.json > finding-ids.txt
+          if grep -Fxqf finding-ids.txt own-replies.txt; then
+            skip "review $REVIEW_ID already has replies from $AUTOMATION_BOT (its summary comment is missing)"
+          fi
           mkdir -p out && cp findings.json out/findings.json
           echo "verdict=proceed" >> "$GITHUB_OUTPUT"
           echo "findings=$n" >> "$GITHUB_OUTPUT"
-          echo "head-sha=$(gh api "repos/$BASE_REPO/pulls/$PR_NUMBER" --jq .head.sha)" >> "$GITHUB_OUTPUT"
+          echo "head-sha=$HEAD_SHA" >> "$GITHUB_OUTPUT"
+          echo "head-ref=$(jq -r .head.ref pr.json)" >> "$GITHUB_OUTPUT"
+          echo "base-ref=$(jq -r .base.ref pr.json)" >> "$GITHUB_OUTPUT"
       - name: Upload findings
         if: steps.checks.outputs.verdict == 'proceed'
         uses: actions/upload-artifact@v7
         with:
-          name: copilot-findings-${{ github.event.pull_request.number }}-${{ github.event.review.id }}
+          name: copilot-findings-${{ steps.checks.outputs.pr-number }}-${{ steps.checks.outputs.review-id }}
           path: out/
           retention-days: 14
 
@@ -119,7 +192,7 @@ jobs:
       - name: Download findings
         uses: actions/download-artifact@v8
         with:
-          name: copilot-findings-${{ github.event.pull_request.number }}-${{ github.event.review.id }}
+          name: copilot-findings-${{ needs.gate.outputs.pr-number }}-${{ needs.gate.outputs.review-id }}
           path: ${{ runner.temp }}/copilot
       - name: Set up Node (npm)
         if: inputs.setup == 'npm'
@@ -191,12 +264,13 @@ jobs:
         with:
           claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
           github_token: ${{ github.token }}
-          # The triggering actor of a pull_request_review event IS Copilot.
-          allowed_bots: "${{ inputs.allowed-bots }},copilot-pull-request-reviewer,Copilot,github-copilot"
+          # The actor of a pull_request_review event IS Copilot; a dispatched
+          # run's actor is the automation app (the org dispatcher).
+          allowed_bots: "${{ inputs.allowed-bots }},copilot-pull-request-reviewer,Copilot,github-copilot${{ inputs.review-id != '' && ',talieisin-org-automation' || '' }}"
           prompt: |
             This working tree is a Dependabot dependency-upgrade PR branch
-            (base ${{ github.event.pull_request.base.ref }} in
-            ${{ github.repository }}, PR #${{ github.event.pull_request.number }})
+            (base ${{ needs.gate.outputs.base-ref }} in
+            ${{ github.repository }}, PR #${{ needs.gate.outputs.pr-number }})
             that an automated pipeline has already upgraded, reviewed, and
             taken to green CI. The commit messages on this branch record the
             earlier rounds' intent — read them (git log) before judging.
@@ -274,7 +348,7 @@ jobs:
       - name: Upload response artifact
         uses: actions/upload-artifact@v7
         with:
-          name: copilot-response-${{ github.event.pull_request.number }}-${{ github.event.review.id }}
+          name: copilot-response-${{ needs.gate.outputs.pr-number }}-${{ needs.gate.outputs.review-id }}
           path: ${{ runner.temp }}/response/
           retention-days: 14
 
@@ -302,7 +376,7 @@ jobs:
         continue-on-error: true
         uses: actions/download-artifact@v8
         with:
-          pattern: copilot-*-${{ github.event.pull_request.number }}-${{ github.event.review.id }}
+          pattern: copilot-*-${{ needs.gate.outputs.pr-number }}-${{ needs.gate.outputs.review-id }}
           path: ${{ runner.temp }}/in
           merge-multiple: true
       # hashFiles() only sees the workspace, so a runner.temp path always
@@ -325,7 +399,7 @@ jobs:
           token: ${{ steps.app-token.outputs.token }}
           app-slug: ${{ steps.app-token.outputs.app-slug }}
           repo: ${{ github.repository }}
-          pr-number: ${{ github.event.pull_request.number }}
+          pr-number: ${{ needs.gate.outputs.pr-number }}
           head-ref: ${{ needs.gate.outputs.head-ref }}
           expected-sha: ${{ needs.gate.outputs.head-sha }}
           patch-path: ${{ runner.temp }}/in/response.patch
@@ -336,7 +410,8 @@ jobs:
         env:
           GH_TOKEN: ${{ steps.app-token.outputs.token }}
           REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ github.event.pull_request.number }}
+          PR_NUMBER: ${{ needs.gate.outputs.pr-number }}
+          REVIEW_ID: ${{ needs.gate.outputs.review-id }}
           VP_OUTCOME: ${{ steps.vpush.outputs.outcome }}
           VP_DETAIL: ${{ steps.vpush.outputs.detail }}
           VP_SHA: ${{ steps.vpush.outputs.new-sha }}
@@ -349,7 +424,9 @@ jobs:
           # ai-meta / ai-outcome markers for the model-tier report (see
           # dependabot-upgrade.yml); a pushed round's outcome is recorded
           # by ci-watch once CI has judged the response commit.
-          META="<!-- ai-meta v=1 run=$GITHUB_RUN_ID round=copilot mode=copilot model=$MODEL -->"
+          # review=<id> is also the answered marker the gate checks: every
+          # comment written below carries it, whatever the outcome.
+          META="<!-- ai-meta v=1 run=$GITHUB_RUN_ID round=copilot mode=copilot model=$MODEL review=$REVIEW_ID -->"
           marks() { # $1 = outcome, or empty for the model line only
             printf '%s\n' "$META"
             [ -z "$1" ] || printf '<!-- ai-outcome v=1 run=%s round=copilot outcome=%s -->\n' "$GITHUB_RUN_ID" "$1"
@@ -466,7 +543,7 @@ jobs:
         env:
           GH_TOKEN: ${{ steps.app-token.outputs.token }}
           REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ github.event.pull_request.number }}
+          PR_NUMBER: ${{ needs.gate.outputs.pr-number }}
           OUTCOME: ${{ steps.watch.outputs.outcome }}
           SHA: ${{ needs.push-response.outputs.new-sha }}
         run: |