SemiAnalysisAI · functionstackx · Sep 19, 2026 · Sep 19, 2026
@@ -0,0 +1,79 @@
+name: Seal Phase 1 measurement receipt
+
+# Each reviewed approval produces an independent immutable record; later requests must not cancel pending issuances.
+on: # zizmor: ignore[concurrency-limits]
+  workflow_dispatch:
+    inputs:
+      kind:
+        description: Seal source measurement or its later publication reference
+        required: true
+        default: measurement
+        type: choice
+        options: [measurement, publication]
+      approval-path:
+        description: Reviewed default-branch qualification/phase1/*.json expectation
+        required: true
+        type: string
+
+permissions:
+  contents: read
+  actions: read # Read source workflow attempts and download the exact artifacts being sealed.
+
+jobs:
+  seal:
+    name: Seal reviewed measurement or publication
+    if: github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          ref: ${{ github.workflow_sha }}
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+      - name: Resolve reviewed expectation
+        env:
+          APPROVAL_PATH: ${{ inputs.approval-path }}
+        shell: python
+        run: |
+          import os
+          from pathlib import Path
+          path = Path(os.environ['APPROVAL_PATH']).resolve(strict=True)
+          root = Path('qualification/phase1').resolve()
+          if not path.is_relative_to(root) or path.suffix != '.json' or any(c in str(path) for c in '\r\n'):
+              raise ValueError('approval must be a reviewed qualification/phase1 JSON file')
+          with Path(os.environ['GITHUB_ENV']).open('a') as stream:
+              stream.write(f'REVIEWED_APPROVAL={path}\n')
+      - name: Seal independently approved complete source sweep
+        if: inputs.kind == 'measurement'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          TRUSTED_WORKFLOW_SHA: ${{ github.workflow_sha }}
+        shell: python
+        run: |
+          import os, subprocess
+          subprocess.run(['uv', 'run', '--locked', 'python', '-m', 'infx.workflows.phase1_publication',
+                          '--approval', os.environ['REVIEWED_APPROVAL'], '--archives', 'verified-archives',
+                          '--output', 'receipt.json'], check=True)
+      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        if: inputs.kind == 'measurement'
+        with:
+          name: measurement-receipt
+          path: receipt.json
+          if-no-files-found: error
+      - name: Bind accepted source to reviewed publication
+        if: inputs.kind == 'publication'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          TRUSTED_ISSUER_SHAS: ${{ vars.INFX_RECEIPT_ISSUER_SHAS }}
+          DEPLOYED_READER_SHA: ${{ vars.INFX_PHASE1_READER_REVISION }}
+        shell: python
+        run: |
+          import os, subprocess
+          subprocess.run(['uv', 'run', '--locked', 'python', '-m', 'infx.workflows.phase1_record',
+                          '--approval', os.environ['REVIEWED_APPROVAL'], '--output', 'publication.json'], check=True)
+      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        if: inputs.kind == 'publication'
+        with:
+          name: publication-record
+          path: publication.json
+          if-no-files-found: error
@@ -21,13 +21,34 @@ concurrency:
 jobs:
     trigger-agentic-ingest:
         name: trigger-agentic-ingest
+        if: github.ref == 'refs/heads/main'
         runs-on: ubuntu-latest
+        permissions:
+            contents: read
+            actions: read # Verify issuer workflow attempts and download accepted receipt artifacts.
         steps:
+            - name: Checkout trusted receipt resolver
+              uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+              with:
+                ref: ${{ github.workflow_sha }}
+                persist-credentials: false
+            - name: Set up uv
+              uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+            - name: Resolve accepted immutable publication
+              id: receipt
+              env:
+                GH_TOKEN: ${{ github.token }}
+                SOURCE_RUN_ID: ${{ inputs.source-run-id }}
+                MERGE_RUN_ID: ${{ inputs.merge-run-id }}
+                INFX_RECEIPT_ISSUER_SHAS: ${{ vars.INFX_RECEIPT_ISSUER_SHAS }}
+                INFX_RECEIPT_ISSUER_WORKFLOW: ${{ vars.INFX_RECEIPT_ISSUER_WORKFLOW }}
+              run: uv run --locked python -m infx.workflows.receipt_transport --publication-required
             - name: Trigger agentic database ingest
               uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
               env:
                 SOURCE_RUN_ID: ${{ inputs.source-run-id }}
                 MERGE_RUN_ID: ${{ inputs.merge-run-id }}
+                RECEIPT_TRANSPORT: ${{ steps.receipt.outputs.payload }}
               with:
                 # Cross-repository dispatch credential for maintainer-triggered ingest recovery.
                 github-token: ${{ secrets.FRONTEND_PAT }} # zizmor: ignore[secrets-outside-env]
@@ -37,6 +58,7 @@ jobs:
                     repo: "InferenceX-app",
                     event_type: "ingest-agentic-results",
                     client_payload: {
+                      ...JSON.parse(process.env.RECEIPT_TRANSPORT),
                       "source-run-id": process.env.SOURCE_RUN_ID,
                       "merge-run-id": process.env.MERGE_RUN_ID,
                       "database-target": "production"

@@ -1000,12 +1000,33 @@ jobs:
               needs.setup.outputs.reuse-enabled == 'true'
             )
         runs-on: ubuntu-latest
+        permissions:
+            contents: read
+            actions: read # Verify issuer workflow attempts and download accepted receipt artifacts.
         steps:
+            - name: Checkout trusted receipt resolver
+              uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+              with:
+                ref: ${{ needs.setup.outputs.tooling-ref }}
+                persist-credentials: false
+            - name: Set up uv
+              uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+            - name: Resolve accepted immutable publication
+              id: receipt
+              env:
+                GH_TOKEN: ${{ github.token }}
+                SOURCE_RUN_ID: ${{ needs.setup.outputs.reuse-enabled == 'true' && needs.setup.outputs.reuse-source-run-id || github.run_id }}
+                MERGE_RUN_ID: ${{ github.run_id }}
+                INFX_RECEIPT_ISSUER_SHAS: ${{ vars.INFX_RECEIPT_ISSUER_SHAS }}
+                INFX_RECEIPT_ISSUER_WORKFLOW: ${{ vars.INFX_RECEIPT_ISSUER_WORKFLOW }}
+              run: uv run --locked python -m infx.workflows.receipt_transport --publication-required --defer-unsealed
             - name: Trigger database ingest
+              if: steps.receipt.outputs.ready == 'true'
               uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
               env:
                 SOURCE_RUN_ID: ${{ needs.setup.outputs.reuse-enabled == 'true' && needs.setup.outputs.reuse-source-run-id || github.run_id }}
                 MERGE_RUN_ID: ${{ github.run_id }}
+                RECEIPT_TRANSPORT: ${{ steps.receipt.outputs.payload }}
               with:
                 # Repository integration credential for the scoped sweep/ingest job.
                 github-token: ${{ secrets.FRONTEND_PAT }} # zizmor: ignore[secrets-outside-env]
@@ -1015,6 +1036,7 @@ jobs:
                     repo: "InferenceX-app",
                     event_type: "ingest-results",
                     client_payload: {
+                      ...JSON.parse(process.env.RECEIPT_TRANSPORT),
                       "source-run-id": process.env.SOURCE_RUN_ID,
                       "merge-run-id": process.env.MERGE_RUN_ID
                     }
@@ -1065,12 +1087,33 @@ jobs:
               )
             )
         runs-on: ubuntu-latest
+        permissions:
+            contents: read
+            actions: read # Verify issuer workflow attempts and download accepted receipt artifacts.
         steps:
+            - name: Checkout trusted receipt resolver
+              uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+              with:
+                ref: ${{ needs.setup.outputs.tooling-ref }}
+                persist-credentials: false
+            - name: Set up uv
+              uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
+            - name: Resolve accepted immutable publication
+              id: receipt
+              env:
+                GH_TOKEN: ${{ github.token }}
+                SOURCE_RUN_ID: ${{ needs.setup.outputs.reuse-enabled == 'true' && needs.setup.outputs.reuse-source-run-id || github.run_id }}
+                MERGE_RUN_ID: ${{ github.run_id }}
+                INFX_RECEIPT_ISSUER_SHAS: ${{ vars.INFX_RECEIPT_ISSUER_SHAS }}
+                INFX_RECEIPT_ISSUER_WORKFLOW: ${{ vars.INFX_RECEIPT_ISSUER_WORKFLOW }}
+              run: uv run --locked python -m infx.workflows.receipt_transport --publication-required --defer-unsealed
             - name: Trigger agentic database ingest
+              if: steps.receipt.outputs.ready == 'true'
               uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
               env:
                 SOURCE_RUN_ID: ${{ needs.setup.outputs.reuse-enabled == 'true' && needs.setup.outputs.reuse-source-run-id || github.run_id }}
                 MERGE_RUN_ID: ${{ github.run_id }}
+                RECEIPT_TRANSPORT: ${{ steps.receipt.outputs.payload }}
               with:
                 # Repository integration credential for the scoped sweep/ingest job.
                 github-token: ${{ secrets.FRONTEND_PAT }} # zizmor: ignore[secrets-outside-env]
@@ -1080,6 +1123,7 @@ jobs:
                     repo: "InferenceX-app",
                     event_type: "ingest-agentic-results",
                     client_payload: {
+                      ...JSON.parse(process.env.RECEIPT_TRANSPORT),
                       "source-run-id": process.env.SOURCE_RUN_ID,
                       "merge-run-id": process.env.MERGE_RUN_ID,
                       "database-target": "production"

@@ -40,6 +40,16 @@ jobs:
           GH_TOKEN: ${{ github.token }}
         run: uv run --locked python -m infx.workflows.stage_results
 
+      - name: Resolve accepted immutable source receipt
+        id: receipt
+        env:
+          GH_TOKEN: ${{ github.token }}
+          SOURCE_RUN_ID: ${{ steps.request.outputs.run-id }}
+          MERGE_RUN_ID: ${{ steps.request.outputs.run-id }}
+          INFX_RECEIPT_ISSUER_SHAS: ${{ vars.INFX_RECEIPT_ISSUER_SHAS }}
+          INFX_RECEIPT_ISSUER_WORKFLOW: ${{ vars.INFX_RECEIPT_ISSUER_WORKFLOW }}
+        run: uv run --locked python -m infx.workflows.receipt_transport
+
       - name: Acknowledge staging request
         id: acknowledge
         uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
@@ -67,6 +77,7 @@ jobs:
           RUN_DATE: ${{ steps.request.outputs.run-date }}
           REQUESTED_BY: ${{ steps.request.outputs.requested-by }}
           COMMENT_ID: ${{ steps.acknowledge.outputs.comment-id }}
+          RECEIPT_TRANSPORT: ${{ steps.receipt.outputs.payload }}
         with:
           # Cross-repository dispatch credential; trusted control code validates maintainer authorization.
           github-token: ${{ secrets.FRONTEND_PAT }} # zizmor: ignore[secrets-outside-env]
@@ -76,6 +87,7 @@ jobs:
               repo: 'InferenceX-app',
               event_type: 'stage-results',
               client_payload: {
+                ...JSON.parse(process.env.RECEIPT_TRANSPORT),
                 'source-repository': `${context.repo.owner}/${context.repo.repo}`,
                 'pr-number': String(context.issue.number),
                 'run-id': process.env.RUN_ID,

@@ -230,7 +230,7 @@ Treat fast results as bring-up evidence, never as a replacement for the canonica
 
 ## 8. Preserve trace and run provenance
 
-AgentX defaults to recorded assistant-response replay. Live server outputs are measured but discarded when constructing later turns. Set `AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES=1` only for an explicitly different live-assistant experiment. The selected trace corpus is model-family dependent unless `WEKA_LOADER_OVERRIDE` pins it. The resolver logs both loader and Hugging Face dataset ([trace resolution](../benchmarks/benchmark_lib.sh#L2023-L2102), [replay semantics](../benchmarks/benchmark_lib.sh#L2104-L2270)).
+At the pilot client revision `754356e9a39acc6cc6afb242d123bb57c3fb6f75`, AgentX always constructs later turns from recorded assistant-response deltas. Live server outputs are measured; `AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES` does not change this loader's behavior. A different replay methodology requires its own qualification. The legacy resolver chooses a model-family-dependent corpus unless `WEKA_LOADER_OVERRIDE` pins it. The prepared H100 client explicitly uses `semianalysis_cc_traces_weka_062126`, 393 entries, and no replay context filter ([Python client](../infx/benchmarks/agentx.py)).
 
 Capture orchestration provenance immediately:
 

@@ -228,7 +228,7 @@ Fast 结果只能作为 bring-up 证据，绝不能替代 canonical candidate。
 
 ## 8. 保留 trace 与运行 provenance
 
-AgentX 默认 replay 已记录的 assistant response。实时服务输出会被测量，但构造后续 turn 时会丢弃。只有在明确要进行不同的 live-assistant 实验时，才设置 `AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES=1`。除非用 `WEKA_LOADER_OVERRIDE` 固定，否则所选 trace corpus 依赖模型 family；resolver 会同时记录 loader 与 Hugging Face dataset（[trace 解析](../benchmarks/benchmark_lib.sh#L2023-L2102)、[replay 语义](../benchmarks/benchmark_lib.sh#L2104-L2270)）。
+在试点固定的客户端版本 `754356e9a39acc6cc6afb242d123bb57c3fb6f75` 中，AgentX 始终使用已记录的 assistant response 增量构造后续轮次。实时服务输出会被测量；`AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES` 不会改变此 loader 的行为。不同的回放方法需要单独验证。旧版 resolver 会根据模型系列选择语料，除非用 `WEKA_LOADER_OVERRIDE` 固定。预先准备的 H100 客户端明确使用 `semianalysis_cc_traces_weka_062126`、393 条记录，并且不按上下文长度过滤回放数据（[Python 客户端](../infx/benchmarks/agentx.py)）。
 
 立即记录 orchestration provenance：