ci(evals): attempt-scoped artifacts, verdict-v2 PR comment, weekly pass-rate gate and one INFRA re-dispatch

- Slice, census and marathon artifacts carry -a<run_attempt>; reports
  download them per artifact (no merge), so records never overwrite and a
  re-run never replaces the first attempt's verdict.
- Planners pass --max-parallel for the capacity preflight (24/16 unchanged:
  the refreshed periodic plan needs 24 slices, the gate census 12).
- PR comment: jq-only job reads collector-outcomes v2 (headline, sanitized
  failure block); the group_by(.name)|last recomputation is gone.
- Reports stamp series identities, upload trial-outcomes-* for history, and
  shard logs upload always (a failed trial no longer reds its runner).
- Weekly report: headline + failure block of both lanes in the issue body,
  the eval:pass-rates --gate step (fails closed without history), close the
  issue on a green run, and UC-E1: when every red is machine-classified
  INFRA/INCOMPLETE, one re-dispatch as a new run in its own concurrency
  group (redispatch_of), both runs reported.

.github/workflows/evals.yml

@@ -144,7 +144,7 @@ jobs:
         if: github.event_name != 'workflow_dispatch' || inputs.validation_phase == 'all'
         env:
           EVALS_ALL: ${{ (github.event_name == 'workflow_dispatch' && inputs.evals_all) && '1' || '' }}
-        run: EVALS_TIER=gate bun --no-install run scripts/test-paid-shards.ts --tier gate --emit-plan /tmp/paid-plan/manifest.json --slice-budget 540 --jobs 2
+        run: EVALS_TIER=gate bun --no-install run scripts/test-paid-shards.ts --tier gate --emit-plan /tmp/paid-plan/manifest.json --slice-budget 540 --jobs 2 --max-parallel 16
 
       - name: Emit validation-phase manifest
         if: github.event_name == 'workflow_dispatch' && inputs.validation_phase != 'all'
@@ -299,11 +299,13 @@ jobs:
           path: /tmp/gstack-eval-input-cache
           key: eval-input-v1-${{ github.repository_id }}-pr-${{ github.event.pull_request.number }}-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.slice }}
 
+      # Attempt-scoped: a re-run attempt's trials are reported under that
+      # attempt and never replace (or collide with) the first attempt's.
       - name: Upload slice results
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
-          name: paid-slice-${{ matrix.slice }}
+          name: paid-slice-${{ matrix.slice }}-a${{ github.run_attempt }}
           path: /tmp/paid-slice-results
           retention-days: 90
 
@@ -323,11 +325,13 @@ jobs:
 
       # The spooled per-shard full logs — a red weekly/PR lane three weeks
       # later needs more than a summary line.
-      - name: Upload shard logs on failure
-        if: failure()
+      # always(), not failure(): a failed behavior trial is a verdict and no
+      # longer reds its runner, but its full log is the diagnosis evidence.
+      - name: Upload shard logs
+        if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
-          name: paid-slice-${{ matrix.slice }}-logs
+          name: paid-logs-slice-${{ matrix.slice }}-a${{ github.run_attempt }}
           include-hidden-files: true
           # The Fix-bun-temp step points TMPDIR at /home/runner/.cache, so the
           # runner's spool lands THERE, not /tmp — the original /tmp glob
@@ -372,11 +376,13 @@ jobs:
           name: paid-plan
           path: /tmp/paid-report
 
+      # One directory per attempt-scoped slice artifact (no merge): shard
+      # records can never overwrite each other, and the report keeps the
+      # first attempt's verdict.
       - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
         with:
-          pattern: paid-slice-[0-9]*
+          pattern: paid-slice-*
           path: /tmp/paid-report
-          merge-multiple: true
 
       - name: Reconcile slices against the manifest (fail-closed)
         id: reconcile
@@ -389,17 +395,31 @@ jobs:
           # (caught by the ship review army; the wiring test now pins this).
           echo "exit=${PIPESTATUS[0]}" >> "$GITHUB_OUTPUT"
 
+      - name: Stamp trial history series
+        if: always() && hashFiles('/tmp/paid-report/trial-outcomes.jsonl') != ''
+        run: bun --no-install run scripts/eval-trial-series.ts /tmp/paid-report/trial-outcomes.jsonl
+
       - name: Upload reconciliation output for the comment job
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
-          name: report-verdict
+          name: report-verdict-a${{ github.run_attempt }}
           path: |
             /tmp/report.txt
             /tmp/paid-report/collector-outcomes.json
+            /tmp/paid-report/report-summary.md
           if-no-files-found: ignore
           retention-days: 30
 
+      - name: Upload trial outcomes for pass-rate history
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: trial-outcomes-pr-a${{ github.run_attempt }}
+          path: /tmp/paid-report/trial-outcomes.jsonl
+          if-no-files-found: ignore
+          retention-days: 90
+
       - name: Fail the workflow when reconciliation failed
         if: steps.reconcile.outputs.exit != '0'
         run: exit 1
@@ -426,18 +446,13 @@ jobs:
 
       - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
         with:
-          pattern: paid-slice-[0-9]*
-          path: /tmp/paid-report
-          merge-multiple: true
-
-      - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
-        with:
-          name: report-verdict
+          name: report-verdict-a${{ github.run_attempt }}
           path: /tmp/verdict
         continue-on-error: true
 
-      # Verified counts come from the read-only report job, not repo code in
-      # this write-token job. Keeps the
+      # Every count, verdict and failure line comes from the read-only report
+      # job's collector-outcomes v2 (panelVerdict() ran there); this job runs
+      # no repo code and never recomputes a verdict. Keeps the
       # "## E2E Evals" marker so the upsert keeps updating the same comment.
       # Runs even when reconciliation failed — a red lane on the PR is the point.
       - name: Post PR comment
@@ -446,13 +461,14 @@ jobs:
           RECONCILE_EXIT: ${{ needs.slices-report.outputs.reconcile-exit }}
         run: |
           # shellcheck disable=SC2086,SC2059
-          RESULTS=$(find /tmp/paid-report -name '*.json' ! -name 'manifest.json' ! -name 'slice-*.json' ! -name '_partial*' 2>/dev/null | sort)
-          TOTAL=0; PASSED=0; FAILED=0; MANUAL=0; FLAKY=0; EXECUTED=0; REUSED=0; COST="0"
+          TOTAL=0; PASSED=0; FAILED=0; MANUAL=0; EXECUTED=0; REUSED=0; COST="0"
           SUITE_LINES=""
           VERIFIED=/tmp/verdict/paid-report/collector-outcomes.json
           if ! jq -e '
             . as $summary |
-            .version == 1 and (.files | type == "array") and (.totals | type == "object") and
+            .version == 2 and (.files | type == "array") and (.totals | type == "object") and
+            (.headline | type == "array") and (.failures | type == "array") and (.panels | type == "array") and
+            (.verdict.verdict == "GREEN" or .verdict.verdict == "RED") and
             ([.files[] | .total == (.passed + .failed + .manual_accepted) and
               (.total == (.executed + .reused)) and
               ([.total,.passed,.failed,.manual_accepted,.executed,.reused,.attempts,.flaky] | all(. >= 0 and (floor == .))) ] | all) and
@@ -461,100 +477,68 @@ jobs:
               all(. as $key | ([$summary.files[] | .[$key]] | add // 0) == $summary.totals[$key]))
           ' "$VERIFIED" >/dev/null 2>&1; then
             VERIFIED=""
-            echo 'Verified collector summary unavailable; manual acceptance is unavailable/unverified.'
+            echo 'Verified report summary unavailable; no verdict, counts or manual acceptance can be shown.'
           fi
+          HEADLINE='(no verified report headline)'
+          FAILURES=""
           if [ -n "$VERIFIED" ]; then
             while IFS=$'\t' read -r f T P F M FL EX RE _ATTEMPTS C TIER SHARD; do
               [ "$T" -eq 0 ] && continue
               TOTAL=$((TOTAL + T)); PASSED=$((PASSED + P)); FAILED=$((FAILED + F))
-              MANUAL=$((MANUAL + M)); FLAKY=$((FLAKY + FL))
+              MANUAL=$((MANUAL + M))
               EXECUTED=$((EXECUTED + EX)); REUSED=$((REUSED + RE))
               COST=$(echo "$COST + $C" | bc)
               STATUS_ICON="✅"
               [ "$M" -gt 0 ] && STATUS_ICON="⚠ manual/unscored"
               [ "$F" -gt 0 ] && STATUS_ICON="❌"
-              [ "$F" -eq 0 ] && [ "$M" -eq 0 ] && [ "$FL" -gt 0 ] && STATUS_ICON="✅⚠"
               SUITE_LINES="${SUITE_LINES}| ${TIER}/${SHARD} | ${P}/${T} | ${M} | ${EX} | ${RE} | ${STATUS_ICON} | \$${C} |\n"
             done < <(jq -r '.files[] | [.file,.total,.passed,.failed,.manual_accepted,.flaky,.executed,.reused,.attempts,.cost,.tier,.shard] | @tsv' "$VERIFIED")
-          else
-          for f in $RESULTS; do
-            if ! jq -e '.total_tests' "$f" >/dev/null 2>&1; then
-              echo "Skipping malformed JSON: $f"
-              continue
-            fi
-            # FINAL-attempt accounting: eval-store keeps EVERY retry attempt
-            # as its own record (that's the flake telemetry), so counting raw
-            # records marks a pass-on-retry as a failure and inflates totals.
-            # Group by test name and judge the LAST record. Retry metadata
-            # includes both passing and failing final outcomes; show it separately.
-            # Guarded: a file with total_tests but a null/non-array `tests`
-            # passes the -e probe, the group_by then fails, and an empty $T
-            # would abort the whole step under bash -e ([ "" -eq 0 ] is an
-            # error) — killing the comment on exactly the corrupted-artifact
-            # runs where the red evidence matters (claude adversarial).
-            STATS=$(jq -r '[.tests | group_by(.name)[] | last] as $final | "\($final | length) \([$final[] | select(.passed)] | length) \([$final[] | select(.passed | not)] | length) \(.flaky_retries // [] | length) \([$final[] | select(.execution != "reused")] | length) \([$final[] | select(.execution == "reused")] | length)"' "$f" 2>/dev/null) || { echo "Skipping malformed tests[] in: $f"; continue; }
-            read -r T P F FL EX RE <<< "$STATS"
-            [ -z "$T" ] && { echo "Skipping malformed tests[] in: $f"; continue; }
-            C=$(jq -r '.total_cost_usd // 0' "$f")
-            TIER=$(jq -r '.tier // "unknown"' "$f")
-            SHARD=$(jq -r '.shard // "-"' "$f")
-            [ "$T" -eq 0 ] && continue
-            TOTAL=$((TOTAL + T))
-            PASSED=$((PASSED + P))
-            FAILED=$((FAILED + F))
-            FLAKY=$((FLAKY + FL))
-            EXECUTED=$((EXECUTED + EX))
-            REUSED=$((REUSED + RE))
-            COST=$(echo "$COST + $C" | bc)
-            STATUS_ICON="✅"
-            [ "$F" -gt 0 ] && STATUS_ICON="❌"
-            [ "$F" -eq 0 ] && [ "$FL" -gt 0 ] && STATUS_ICON="✅⚠"
-            SUITE_LINES="${SUITE_LINES}| ${TIER}/${SHARD} | ${P}/${T} | unverified | ${EX} | ${RE} | ${STATUS_ICON} | \$${C} |\n"
-          done
+            # Report-sanitized lines (no @-mentions, one capped line each), fenced here.
+            HEADLINE=$(jq -r '.headline[]' "$VERIFIED")
+            FAILURES=$(jq -r '.failures[]' "$VERIFIED")
           fi
 
           COVERAGE=$(jq -r '"Profile: \(.profile // "full") / \(.prCoverage.mode // "broad"); selected behaviors: \(.selection.e2e | if . == null then "all" else length end), judges: \(.selection.judges | if . == null then "all" else length end). Deferred to scheduled/release coverage: \(.prCoverage.deferred // [] | length) behaviors and \(.prCoverage.deferredPromptFiles // [] | length) changed prompt files. Deferred checks did not run and receive no PR-pass credit."' /tmp/paid-report/manifest.json) || COVERAGE='Coverage manifest unavailable; no coverage claim.'
 
           STATUS="✅ PASS"
-          if [ "${RECONCILE_EXIT:-1}" != "0" ] || [ "$FAILED" -gt 0 ]; then STATUS="❌ FAIL"; fi
+          if [ "${RECONCILE_EXIT:-1}" != "0" ]; then STATUS="❌ FAIL"; fi
           if [ "$STATUS" = '✅ PASS' ] && [ "$MANUAL" -gt 0 ]; then STATUS='⚠ MANUAL ACCEPTED (unscored)'; fi
-          if [ -z "$VERIFIED" ]; then STATUS='❌ FAIL (manual acceptance unavailable/unverified)'; fi
+          if [ -z "$VERIFIED" ]; then STATUS='❌ FAIL (verified report unavailable)'; fi
 
           BODY="## E2E Evals: ${STATUS}
 
-          **${PASSED} automated passed / ${TOTAL} final results** | **${FAILED} failed, ${MANUAL} manual accepted (unscored; no score-cache credit)** | **${EXECUTED} executed, ${REUSED} reused** | **\$${COST}** total cost | reconcile exit: ${RECONCILE_EXIT:-missing}$([ "$FLAKY" -gt 0 ] && printf ' | ⚠ %s cases with multiple attempts' "$FLAKY")
+          \`\`\`
+          ${HEADLINE}
+          \`\`\`
+
+          **${EXECUTED} executed, ${REUSED} reused** rule/judge records | **${MANUAL} manual accepted (unscored; no score-cache credit)** | **\$${COST}** rule/judge cost | reconcile exit: ${RECONCILE_EXIT:-missing}
 
           ${COVERAGE}
 
+          <details><summary>Rule and judge shards</summary>
+
           | Shard | Automated result | Manual/unscored | Executed | Reused | Status | Cost |
           |-------|------------------|-----------------|----------|--------|--------|------|
           $(echo -e "$SUITE_LINES")
+          </details>
 
           <details><summary>Fail-closed reconciliation</summary>
 
           \`\`\`
-          $(tail -c 4000 /tmp/verdict/report.txt 2>/dev/null || echo '(no reconciliation output)')
+          $(tail -c 4000 /tmp/verdict/report.txt 2>/dev/null | sed 's/@/@\xe2\x80\x8b/g' || echo '(no reconciliation output)')
           \`\`\`
           </details>
 
           ---
-          *Sliced lane: declared PR profile or broad fallback via scripts/test-paid-shards.ts (planner → duration-packed executors → fail-closed report). Reused results retain their original provenance and expiry.*"
+          *Sliced lane: planner → duration-packed executors → fail-closed report. Behavior cases run a pre-registered 3-trial panel (PASS at 2/3 with no contract violation); a PASS 2/3 is shown with its failed trial, never as a clean pass. Reused results retain their original provenance and expiry.*"
 
-          if [ "$FAILED" -gt 0 ]; then
-            FAILURES=""
-            for f in $RESULTS; do
-              if ! jq -e '.failed' "$f" >/dev/null 2>&1; then continue; fi
-              if [ -n "$VERIFIED" ]; then
-                FAILS=$(jq -r '[.tests | group_by(.name)[] | last | select(.passed == false and (has("manual_review") | not))][] | "- ❌ \(.name): \(.exit_reason // "unknown")"' "$f" 2>/dev/null || echo "- ⚠️ parse error")
-              else
-                FAILS=$(jq -r '[.tests | group_by(.name)[] | last | select(.passed == false)][] | "- ❌ \(.name): \(.exit_reason // "unknown")"' "$f" 2>/dev/null || echo "- ⚠️ parse error")
-              fi
-              FAILURES="${FAILURES}${FAILS}\n"
-            done
+          if [ -n "$FAILURES" ]; then
             BODY="${BODY}
 
-          ### Failures
-          $(echo -e "$FAILURES")"
+          ### Failures and split verdicts
+          \`\`\`
+          ${FAILURES}
+          \`\`\`"
           fi
 
           COMMENT_ID=$(gh api repos/${{ github.repository }}/issues/${{ github.event.pull_request.number }}/comments \