[Feature] System One: shared tasks, persistent frontend and replica pools

.dockerignore

@@ -52,6 +52,4 @@ paper/
 perf/
 src/training/model_classifier/verified_single_judge_*/
 dashboard/frontend/node_modules/
-dashboard/wizmap/node_modules/
 dashboard/frontend/dist/
-dashboard/wizmap/dist/

.github/dependabot.yml

@@ -14,19 +14,6 @@ updates:
         applies-to: security-updates
         patterns: ["*"]
 
-  - package-ecosystem: npm
-    directory: /dashboard/wizmap
-    schedule:
-      interval: weekly
-      day: monday
-      time: "04:00"
-      timezone: UTC
-    open-pull-requests-limit: 0
-    groups:
-      dashboard-wizmap-security:
-        applies-to: security-updates
-        patterns: ["*"]
-
   - package-ecosystem: npm
     directory: /website
     schedule:

.github/workflows/check-generated.yml

@@ -25,16 +25,16 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.12"
-      - name: Check Schema, OpenAPI, and CRD References
-        run: make config-schema-check api-docs-check docs-crd-check model-runtime-client-check
+      - name: Check Schema, OpenAPI, CRD, and Runtime References
+        run: make config-schema-check api-docs-check docs-crd-check model-runtime-client-check decision-runtime-catalog-check
       - name: Record Generated Contract Evidence
         env:
           VERIFICATIONS: ${{ inputs.verifications }}
         run: |
           python3 -m pip install PyYAML==6.0.3
           python3 tools/ci/workflow_evidence.py --verifications "$VERIFICATIONS" --id generated-contracts \
             --check config-schema-check --check api-docs-check --check docs-crd-check \
-            --check model-runtime-client-check \
+            --check model-runtime-client-check --check decision-runtime-catalog-check \
             --output .agent-harness/generated-contracts-result.json
       - uses: actions/upload-artifact@v4
         with:

.github/workflows/recipe-conformance.yml

@@ -106,6 +106,13 @@ jobs:
         if: inputs.prebuilt-images
         with:
           images: '["vllm-sr"]'
+      - name: Allocate CPUs for serial long-context probes
+        if: matrix.shard == 'built-in-latest-mom-v1'
+        run: |
+          # This shard runs one probe at a time, including full-context embeddings.
+          cpu_threads="$(nproc)"
+          printf 'VLLM_SRUN_CPU_THREADS=%s\n' "$cpu_threads" >> "$GITHUB_ENV"
+          printf 'Serial long-context worker threads: %s\n' "$cpu_threads"
       - name: Run maintained recipe probes
         env:
           RECIPE_CONFORMANCE_RECIPES: ${{ matrix.recipes }}

.pre-commit-config.yaml

@@ -57,6 +57,13 @@ repos:
         pass_filenames: false
         files: ^(config/catalog/|config/recipes/built-in/|src/vllm-sr/cli/model_bundle\.py$|src/semantic-router/pkg/catalog/zz_generated_catalog\.go$|src/vllm-sr/cli/model_assets/latest/catalog\.yaml$|dashboard/frontend/src/generated/modelCatalog\.json$|website/static/model-catalog/catalog\.json$|tools/catalog/|tools/make/model-catalog\.mk$)
 
+      - id: decision-runtime-catalog-generated
+        name: decision runtime catalog generated projection
+        entry: make decision-runtime-catalog-check
+        language: system
+        pass_filenames: false
+        files: ^(src/model-runtime/vllm_srun/(registry/tables/\w+|systemone|families/decision1/(family|questions)|families/decision2/family|families/vela2/request)\.py$|src/model-runtime/vllm_srun/registry/releases\.generated\.json$|dashboard/frontend/scripts/generate-decision-runtime-catalog\.py$|dashboard/frontend/src/pages/decisionRuntimeCatalog\.generated\.json$|src/semantic-router/pkg/modelservice/decision_catalog\.generated\.json$|tools/make/dashboard\.mk$|\.pre-commit-config\.yaml$)
+
       - id: cli-reference-generated
         name: CLI command reference
         entry: make docs-cli-check
@@ -111,7 +118,7 @@ repos:
         language: system
         require_serial: true
         files: \.md$
-        exclude: ^(node_modules/|website/node_modules/|dashboard/frontend/node_modules/|dashboard/wizmap/node_modules/|bench/|e2e/config/models/|models/|vsr/|website/docs/api/crd-reference\.md$|CLAUDE\.md$)
+        exclude: ^(node_modules/|website/node_modules/|dashboard/frontend/node_modules/|bench/|e2e/config/models/|models/|vsr/|website/docs/api/crd-reference\.md$|CLAUDE\.md$)
 
   # Yaml specific hooks
   - repo: local

bench/agent_crew/configs/crew.yaml

@@ -119,7 +119,6 @@ routing:
 
 global:
   router:
-    auto_model_name: MoM
 
   stores:
     response_cache:
@@ -153,3 +152,7 @@ global:
         model_ref: ""
         model_id: ""
         use_mmbert_32k: false
+
+entrypoints:
+  - model_names: ["MoM"]
+    recipe: default

bench/grounded_fusion/fusioneval/grounding_test.go

@@ -34,18 +34,19 @@ func TestProduceArmKeepsRealAndPlaceboGroundingRequestLocal(t *testing.T) {
 	unblock := func() { releaseOnce.Do(func() { close(release) }) }
 	t.Cleanup(unblock)
 	var realCalls, forbiddenCalls atomic.Int32
-	forbidden := func(context.Context, string, string, string) ([]string, float32, error) {
+	forbidden := func(context.Context, string, string, string) (looper.GroundingEvidence, error) {
 		forbiddenCalls.Add(1)
-		return nil, 0, nil
+		return looper.GroundingEvidence{}, nil
 	}
-	real := func(ctx context.Context, _, _, _ string) ([]string, float32, error) {
+	real := func(ctx context.Context, _, _, _ string) (looper.GroundingEvidence, error) {
 		realCalls.Add(1)
 		startOnce.Do(func() { close(entered) })
 		select {
 		case <-release:
-			return []string{"unsupported"}, 0.25, nil
+			score := float32(0.25)
+			return looper.GroundingEvidence{Unsupported: true, Spans: []string{"unsupported"}, Probability: &score}, nil
 		case <-ctx.Done():
-			return nil, 0, ctx.Err()
+			return looper.GroundingEvidence{}, ctx.Err()
 		}
 	}
 	done := make(chan answerRecord, 1)
@@ -65,8 +66,9 @@ func TestProduceArmKeepsRealAndPlaceboGroundingRequestLocal(t *testing.T) {
 	require.True(t, placebo.GroundingPresent)
 	require.Len(t, placebo.Panel, 2)
 
-	other := produceArm(fusion, client, opt, it, entry, "C", &looper.GroundingBackends{Detect: func(context.Context, string, string, string) ([]string, float32, error) {
-		return []string{"unsupported"}, 0.75, nil
+	other := produceArm(fusion, client, opt, it, entry, "C", &looper.GroundingBackends{Detect: func(context.Context, string, string, string) (looper.GroundingEvidence, error) {
+		score := float32(0.75)
+		return looper.GroundingEvidence{Unsupported: true, Spans: []string{"unsupported"}, Probability: &score}, nil
 	}})
 	assertArmScores(t, other, 0.25)
 	unblock()

bench/grounded_fusion/fusioneval/main.go

@@ -245,7 +245,7 @@ func run(opt options) error {
 // given (seed, context, answer) but carrying no real signal. Mirrors the
 // in-package test placebo so arm D weights on noise, isolating the score's signal.
 func placeboDetector(seed uint64) looper.HallucinationDetectFunc {
-	return func(_ context.Context, contextText, _, answer string) ([]string, float32, error) {
+	return func(_ context.Context, contextText, _, answer string) (looper.GroundingEvidence, error) {
 		h := fnv.New64a()
 		var b [8]byte
 		binary.LittleEndian.PutUint64(b[:], seed)
@@ -255,7 +255,8 @@ func placeboDetector(seed uint64) looper.HallucinationDetectFunc {
 		_, _ = h.Write([]byte(answer))
 		// Deterministic seed for the placebo; the uint64->int64 wrap is harmless.
 		r := rand.New(rand.NewSource(int64(h.Sum64()))) //nolint:gosec // deterministic eval seed, overflow harmless
-		return []string{"placebo"}, float32(r.Float64()), nil
+		score := float32(float32(r.Float64()))
+		return looper.GroundingEvidence{Unsupported: true, Spans: []string{"placebo"}, Probability: &score}, nil
 	}
 }
 

bench/router_flow/configs/amd_auto_gpqa_omni.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1200s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -267,3 +264,7 @@ routing:
             {{end}}
 
             If this is a multiple-choice question, return exactly one final answer letter: A, B, C, or D. Do not include prose, labels, or punctuation.
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_hle_closed.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1800s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -326,3 +323,7 @@ routing:
             Explanation: {your concise reasoning}
             Exact Answer: {your succinct final answer}
             Confidence: {integer percentage between 0% and 100%}
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_hle_glm52.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1800s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -417,3 +414,7 @@ routing:
             Explanation: {your concise reasoning}
             Exact Answer: {your succinct final answer}
             Confidence: {integer percentage between 0% and 100%}
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_hle_hybrid.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1800s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -428,3 +425,7 @@ routing:
             Explanation: {your concise reasoning}
             Exact Answer: {your succinct final answer}
             Confidence: {integer percentage between 0% and 100%}
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_livecode_glm52_omni.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1200s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -516,3 +513,7 @@ routing:
 
             If code is requested, return exactly one complete ```python code
             block and no prose.
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_livecode_kimi_k27_code_omni.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1200s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -516,3 +513,7 @@ routing:
 
             If code is requested, return exactly one complete ```python code
             block and no prose.
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_livecode_omni.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1200s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -520,3 +517,7 @@ routing:
 
             If code is requested, return exactly one complete ```python code
             block and no prose.
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_scicode_omni.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1200s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -458,3 +455,7 @@ routing:
             {{end}}
 
             Return exactly one ```python code block and no prose.
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default

bench/router_flow/configs/amd_auto_swe_glm52.yaml

@@ -7,9 +7,6 @@ listeners:
     timeout: 1800s
 
 global:
-  router:
-    auto_model_names:
-      - vllm-sr/auto
   integrations:
     looper:
       endpoint: http://vllm-sr-envoy-container:8899/v1/chat/completions
@@ -479,3 +476,7 @@ routing:
             the checked-out source. When ready to submit, the single command
             must be exactly
             `echo COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT && cat patch.txt`.
+
+entrypoints:
+  - model_names: ["vllm-sr/auto"]
+    recipe: default