@@ -246,7 +246,20 @@ and read its `framework:`, `runner:`, and `image:` fields.
   the matching entry named.
 - N/A if the PR changes no master-config entries (state that in one line).
 
-## Check 7 — No benchmark hacks that change the model architecture
+## Check 7 — No submissions for deprecated models or scenarios
+Read the current `MODELS.md` in the checked-out default branch. It is the source of
+truth for active and deprecated models, scenarios, and model-scenario combinations.
+For every benchmark configuration or recipe that the PR adds, changes, or re-enables,
+identify its model prefix and scenario, including fixed-sequence, agentic, single-node,
+and multi-node entries.
+- Use `date -u +%F` to establish the review date. Honor an effective date in
+  `MODELS.md`, so a scheduled future deprecation is allowed until its stated date.
+- FAIL if the PR submits a model that is retired on the review date, a deprecated
+  scenario, or a deprecated model-scenario combination. Name the model prefix,
+  scenario, and the `MODELS.md` row or notice that prohibits it.
+- N/A if the PR adds, changes, or re-enables no benchmark configurations or recipes.
+
+## Check 8 — No benchmark hacks that change the model architecture
 Verify from the PR diff (server args in `benchmarks/**` and master-config changes)
 that nothing alters the model architecture or reduces its FLOPs — e.g. `--hf-overrides`
 that skip the indexer every N layers on a model that doesn't natively support it,
@@ -261,7 +274,7 @@ production by accuracy-sensitive customers.
 - FAIL with the exact flag/value if architecture FLOPs are reduced without native
   model support. PASS in one line otherwise; N/A if the PR touches no server args.
 
-## Check 8 — Speculative-decoding configs benchmark through chat templates
+## Check 9 — Speculative-decoding configs benchmark through chat templates
 If this PR adds or changes a speculative-decoding config (MTP / EAGLE / draft-model
 flags such as `--speculative-config`, `--speculative-algorithm`, `spec-decode`, config
 names ending in `-mtp`), verify the benchmark client exercises the model through its
@@ -271,7 +284,7 @@ completions) so the acceptance-length distribution matches real-world traffic.
   name the config and script line.
 - N/A if the PR has no speculative-decoding changes.
 
-## Check 9 — No engine patches without a waiver
+## Check 10 — No engine patches without a waiver
 The pinned upstream image must run AS SHIPPED — the community must be able to
 reproduce the number from the released image. From the PR diff (scripts under
 `benchmarks/**`, master configs, workflow changes), scan for anything that modifies
@@ -300,7 +313,7 @@ Installing the benchmark harness and client-side deps (aiperf, eval tooling) is
   waiver not linked / waiver does not cover this patch).
 - N/A if the PR touches no benchmark scripts, images, or configs.
 
-## Check 10 — Agentic spec-decode configs use the golden simulated acceptance length
+## Check 11 — Agentic spec-decode configs use the golden simulated acceptance length
 APPLICABILITY: this check covers AGENTIC-workload benchmark changes that enable
 speculative decoding. From the PR diff, identify configs that are BOTH:
 - agentic — scripts under `benchmarks/single_node/agentic/**`, multi-node recipes
@@ -342,12 +355,12 @@ Verify BOTH:
   additional detail section must state the source of the AL value (e.g. a pending
   golden-curve collection run); FAIL if it does not.
 - Also FAIL (as a benchmark hack) if simulated/synthetic-acceptance knobs appear on a
-  NON-agentic spec-decode config, where Check 8's real-traffic AL standard applies —
+  NON-agentic spec-decode config, where Check 9's real-traffic AL standard applies —
   unless the sign-off documents a sanctioned exception.
 - N/A if the PR has no agentic speculative-decoding changes (state that in one line).
 
 ## Verdict and output
-Decide PASS only if Checks 0-10 ALL pass (a check reported as `N/A` counts as a pass —
+Decide PASS only if Checks 0-11 ALL pass (a check reported as `N/A` counts as a pass —
 keep the `N/A — <reason>` row so the reviewer sees it was considered). Post EXACTLY ONE summary comment on
 PR #${PR_NUMBER} using `gh pr comment`. Start the comment with
 the hidden marker so reruns are identifiable:
@@ -373,8 +386,7 @@ single terse line either. Rules:
   restating the checklist, no hedging ("if X then maybe Y" — make the call). Link the
   run/recipe instead of describing it.
 
-- If everything is to standard: post the verdict header + the eleven one-line rows
-  (with the green run URL). Do NOT @-mention anyone on a pass.
+- If everything is to standard: post the verdict header + the twelve one-line rows
 - If anything is NOT to standard: the verdict header must be immediately followed by a
   line that @-mentions the sign-off author as `@${SIGNOFF_AUTHOR}`
   with the blocking summary. Then the per-check lines, each failing one led by its root
 
@@ -166,6 +166,27 @@ and eval jobs retain their canonical settings. Adding or removing the modifier
 restarts the active sweep. Fast-mode runs are not eligible for artifact reuse
 after merge.
 
+## Trusted External-Fork Sweep Dispatch (PoC)
+
+Public-fork `pull_request` workflows receive no repository secrets. For an
+external PR, the ordinary `run-sweep.yml` run therefore validates the
+changelog but does not fan out onto GPU runners. A maintainer with `write`,
+`maintain`, or `admin` permission can add any modifier labels first, then apply
+one primary sweep label to approve the PR's exact current head SHA.
+`trusted-external-sweep.yml` then dispatches `e2e-tests.yml` from `main`, pins
+both the approved head and GitHub's merge SHA, and runs the generated matrix
+with the trusted workflow's secrets.
+
+The approval is revision-specific. A later push is not trusted automatically;
+remove and re-add the primary sweep label to approve the new SHA. The trusted
+dispatcher never checks out or executes PR code itself.
+
+This proof of concept produces benchmark and evaluation artifacts through the
+End-to-End Tests workflow. Those runs are not yet eligible for
+`/reuse-sweep-run`, which currently accepts only `run-sweep.yml` runs. The PoC
+also fans out the selected matrix immediately; it does not reproduce
+`run-sweep.yml`'s canary-first sequencing.
+
 ## Reusing an Approved PR Full Sweep
 
 `[skip-sweep]` skips PR benchmark setup only; changelog and reuse checks still
 
@@ -138,6 +138,16 @@ on:
         type: string
         required: false
         default: ""
+      require-power:
+        description: "Fail fixed-sequence result processing when GPU power is invalid"
+        type: boolean
+        required: false
+        default: false
+      power-producer-sha:
+        description: "Expected srt-slurm producer git commit for power validation"
+        type: string
+        required: false
+        default: ''
       # Agentic-coding inputs. Fixed-seq-len jobs leave these empty.
       scenario-type:
         description: "Scenario type (fixed-seq-len or agentic-coding)"
@@ -213,6 +223,8 @@ env:
   RUN_EVAL: ${{ inputs.run-eval }}
   EVAL_ONLY: ${{ inputs.eval-only }}
   EVAL_CONC: ${{ inputs.eval-conc }}
+  REQUIRE_POWER: ${{ inputs.require-power && '1' || '0' }}
+  POWER_PRODUCER_SHA: ${{ inputs.power-producer-sha }}
   # Agentic-coding env. Fixed-seq-len jobs leave these empty.
   SCENARIO_TYPE: ${{ inputs.scenario-type }}
   SCENARIO_SUBDIR: ${{ inputs.scenario-type == 'agentic-coding' && 'agentic/' || 'fixed_seq_len/' }}
@@ -358,6 +370,12 @@ jobs:
         env:
           RUNNER_TYPE: ${{ inputs.runner }}
         run: |
+          # Launcher stamp supplies the producer SHA unless the
+          # power-producer-sha input is set as a manual override
+          if [ -z "$POWER_PRODUCER_SHA" ] && [ -f power-producer-sha.txt ]; then
+            export POWER_PRODUCER_SHA="$(cat power-producer-sha.txt)"
+            echo "POWER_PRODUCER_SHA derived from launcher stamp: $POWER_PRODUCER_SHA"
+          fi
           # Process each result file
           for result_file in ${RESULT_FILENAME}_*.json; do
             if [ -f "$result_file" ]; then
@@ -386,6 +404,19 @@ jobs:
           name: bmk_${{ env.RESULT_FILENAME }}
           path: agg_${{ env.RESULT_FILENAME }}_*.json
 
+      - name: Upload power audit bundle
+        if: ${{ always() && !inputs.eval-only && inputs.scenario-type != 'agentic-coding' }}
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: power_audit_${{ env.RESULT_FILENAME }}
+          path: |
+            ${{ env.RESULT_FILENAME }}_*.json
+            agg_${{ env.RESULT_FILENAME }}_*.json
+            power_validation_${{ env.RESULT_FILENAME }}_*.json
+            LOGS/power/**
+            LOGS/*/results_*.json
+          if-no-files-found: ignore
+
       - name: Upload server logs
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
 
@@ -129,6 +129,11 @@ on:
         required: false
         type: string
         default: '3600'
+      require-power:
+        description: "Fail fixed-sequence result processing when GPU power is invalid"
+        required: false
+        type: boolean
+        default: false
       agentx-fast:
         description: "Use one warmup request per AgentX lane and a 20-minute profile"
         required: false
@@ -179,6 +184,7 @@ env:
   KV_P2P_TRANSFER: ${{ inputs.kv-p2p-transfer }}
   TOTAL_CPU_DRAM_GB: ${{ inputs.total-cpu-dram-gb }}
   DURATION: ${{ inputs.duration }}
+  REQUIRE_POWER: ${{ inputs.require-power && '1' || '0' }}
   AIPERF_EXPERIMENTAL_FAST: ${{ inputs.agentx-fast && '1' || '0' }}
   EVAL_LIMIT: ${{ inputs.eval-limit }}
   SWEBENCH_GEN_MODE: ${{ inputs.swebench-gen-mode }}
@@ -356,7 +362,26 @@ jobs:
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: ${{ inputs.eval-only && 'eval_gpu_metrics_' || 'gpu_metrics_' }}${{ env.RESULT_FILENAME }}
-          path: gpu_metrics.csv
+          path: |
+            gpu_metrics.csv
+            gpu_metrics_energy_start.csv
+            gpu_metrics_energy_end.csv
+            gpu_metrics_identity.json
+          if-no-files-found: ignore
+
+      - name: Upload power audit bundle
+        if: ${{ always() && !inputs.eval-only && inputs.scenario-type != 'agentic-coding' }}
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: power_audit_${{ env.RESULT_FILENAME }}
+          path: |
+            ${{ env.RESULT_FILENAME }}.json
+            agg_${{ env.RESULT_FILENAME }}.json
+            gpu_metrics.csv
+            gpu_metrics_energy_start.csv
+            gpu_metrics_energy_end.csv
+            gpu_metrics_identity.json
+            power_validation_${{ env.RESULT_FILENAME }}.json
           if-no-files-found: ignore
 
       - name: Upload eval results (if any)
 
@@ -25,12 +25,14 @@ name: CODEOWNER Sign-off Verify
 #   6. vLLM/SGLang configs run upstream images (hub.docker.com/u/vllm,
 #      hub.docker.com/u/lmsysorg) on established hardware, and vLLM/SGLang
 #      submissions land before additional frameworks (TRT-LLM, ATOM, ...).
-#   7. No benchmark hacks that change the model architecture / cut FLOPs.
-#   8. Speculative-decoding configs benchmark through chat templates.
-#   9. No patches to the inference engine / serving stack (the pinned image
+#   7. No submissions for models, scenarios, or model-scenario combinations
+#      deprecated in MODELS.md.
+#   8. No benchmark hacks that change the model architecture / cut FLOPs.
+#   9. Speculative-decoding configs benchmark through chat templates.
+#  10. No patches to the inference engine / serving stack (the pinned image
 #      runs as shipped) — the only exception is a patch covered by a
 #      filled-out docs/waiver/<PR_NUMBER>.md waiver.
-#  10. Agentic spec-decode configs simulate acceptance at the committed golden
+#  11. Agentic spec-decode configs simulate acceptance at the committed golden
 #      AL from golden_al_distribution/ (SGLang SGLANG_SIMULATE_ACC_* env vars,
 #      vLLM synthetic rejection sampling in --speculative-config).
 # If any of those are not to standard, Claude posts a single comment that
 
@@ -10,7 +10,7 @@ on:
         description: "EP library to sweep — 'all' runs every EP backend in one matrix"
         type: choice
         default: all
-        options: [all, deepep-v2, mori, uccl-ep, nccl-ep]
+        options: [all, deepep-v2, mori, uccl-ep, nccl-ep, flashinfer-ep]
       only_sku:
         description: Restrict to one GHA runner pool; blank = all
         type: string
@@ -224,6 +224,11 @@ jobs:
             || { echo "CollectiveX allocation cleanup failed; results may still be changing" >&2; exit 1; }
           python3 experimental/CollectiveX/summarize.py \
             --results-dir experimental/CollectiveX/results >> "$GITHUB_STEP_SUMMARY" || true
+          # Second view over the same documents: latency answers how long a point took,
+          # this answers how fast it moved its payload. Both are renderers over the staged
+          # JSON and neither gates the leg, so both are `|| true`.
+          python3 experimental/CollectiveX/bandwidth.py \
+            --results-dir experimental/CollectiveX/results >> "$GITHUB_STEP_SUMMARY" || true
           shopt -s nullglob
           results=(experimental/CollectiveX/results/*.json)
           if [ "${#results[@]}" -eq 0 ]; then