FMVISION / COMPACTIONJAY + MERRITT

The case for removal

100 candidates from conduct. Each row argues against giving that entry a dedicated byte-table slot, ranked from strongest removal case to weakest.

Judge the argument against each candidate: duplicated meaning, fixed configuration, excessive specialization, or an overlapping variant. The evidence column shows code excerpts from the pinned source revision; comparisons support overlap arguments, while the removal recommendation remains a judgment. Nothing has been removed.

Download CSV

100 of 100 candidates

CandidateReasonEvidence
001trinity-passproduction protocol requirementsDuplicates trinity-pass in the analysis-protocol registry. Giving this requirements record a second byte creates two codes for the same method.
Candidate definitioninternal/production/protocols.go:22
22: "trinity-pass": {
23: ID:             "trinity-pass",
24: Name:           "Trinity Pass",
25: Category:       "Multi-Pass Analysis",
26: RequiredModels: []string{"claude-3-opus", "claude-3-sonnet"},
27: MinGPUMemoryGB: 40,
Comparison definitioncmd/coverage_protocol_defs.go:11
11: ID:       "trinity-pass",
12: Name:     "Trinity Pass",
13: Category: "Multi-Pass Analysis",
14: Description: "Three-perspective analysis simulating creative executive, producer, and director viewpoints for comprehensive coverage.",
60: {Order: 1, Name: "Extraction", Description: "Extract and parse screenplay into structured format", Models: []string{"local"}, Parallel: false},
61: {Order: 2, Name: "Creative Executive Pass", Description: "Analyze from studio executive perspective - commercial viability, market positioning, audience appeal", Models: []string{"claude-opus"}, Parallel: true},
62: {Order: 3, Name: "Producer Pass", Description: "Analyze from producer perspective - budget implications, casting needs, production feasibility", Models: []string{"claude-opus"}, Parallel: true},
63: {Order: 4, Name: "Director Pass", Description: "Analyze from director perspective - visual storytelling, performance opportunities, cinematic potential", Models: []string{"claude-opus"}, Parallel: true},
64: {Order: 5, Name: "Synthesis", Description: "Reconcile three perspectives into unified assessment with prioritized recommendations", Models: []string{"claude-opus"}, Parallel: false},
67: Outputs: []string{"trinity_coverage.md", "executive_notes.md", "producer_notes.md", "director_notes.md", "synthesis.json"},
002deep-diveproduction protocol requirementsDuplicates deep-dive in the analysis-protocol registry. Giving this requirements record a second byte creates two codes for the same method.
Candidate definitioninternal/production/protocols.go:31
31: "deep-dive": {
32: ID:             "deep-dive",
33: Name:           "Deep Dive",
34: Category:       "Multi-Pass Analysis",
35: RequiredModels: []string{"claude-3-opus"},
36: MinGPUMemoryGB: 80,
Comparison definitioncmd/coverage_protocol_defs.go:78
78: ID:       "deep-dive",
79: Name:     "Deep Dive",
80: Category: "Multi-Pass Analysis",
81: Description: "Four-pass progressive analysis that goes from surface-level overview to granular scene-by-scene examination.",
124: {Order: 1, Name: "Aerial Pass", Description: "High-level overview - logline, genre, tone, major structure", Models: []string{"claude-sonnet"}, Parallel: false},
125: {Order: 2, Name: "Act Analysis", Description: "Act-by-act breakdown with turning points and progression", Models: []string{"claude-opus"}, Parallel: false},
126: {Order: 3, Name: "Sequence Analysis", Description: "Eight-sequence breakdown with beat mapping", Models: []string{"claude-opus"}, Parallel: true},
127: {Order: 4, Name: "Scene Microscope", Description: "Detailed scene-by-scene analysis of flagged areas", Models: []string{"claude-sonnet"}, Parallel: true},
128: {Order: 5, Name: "Integration", Description: "Compile multi-resolution findings into coherent report", Models: []string{"claude-opus"}, Parallel: false},
131: Outputs: []string{"deep_dive_report.md", "aerial_view.md", "act_analysis.md", "sequence_breakdown.json", "scene_notes.json"},
003lightning-passproduction protocol requirementsDuplicates lightning-pass in the analysis-protocol registry. Giving this requirements record a second byte creates two codes for the same method.
Candidate definitioninternal/production/protocols.go:40
40: "lightning-pass": {
41: ID:             "lightning-pass",
42: Name:           "Lightning Pass",
43: Category:       "Speed Optimized",
44: RequiredModels: []string{"claude-3-haiku", "llama-70b"},
45: MinGPUMemoryGB: 20,
Comparison definitioncmd/coverage_protocol_defs.go:364
364: ID:       "lightning-pass",
365: Name:     "Lightning Pass",
366: Category: "Speed Optimized",
367: Description: "Ultra-fast coverage using parallel execution and smaller models for rapid decision-making.",
419: {Order: 1, Name: "Parallel Rubric", Description: "All 7 core rubric points analyzed simultaneously", Models: []string{"claude-haiku"}, Parallel: true},
420: {Order: 2, Name: "Quick Aggregate", Description: "Fast aggregation of parallel results", Models: []string{"claude-haiku"}, Parallel: false},
421: {Order: 3, Name: "One-Page Summary", Description: "Generate actionable one-page coverage", Models: []string{"claude-sonnet"}, Parallel: false},
424: Outputs: []string{"lightning_coverage.md", "scores.json"},
004council-of-readersproduction protocol requirementsDuplicates council-of-readers in the analysis-protocol registry. Giving this requirements record a second byte creates two codes for the same method.
Candidate definitioninternal/production/protocols.go:49
49: "council-of-readers": {
50: ID:             "council-of-readers",
51: Name:           "Council of Readers",
52: Category:       "Consensus Protocols",
53: RequiredModels: []string{"claude-3-opus", "gpt-4", "llama-70b"},
54: MinGPUMemoryGB: 140,
Comparison definitioncmd/coverage_protocol_defs.go:215
215: ID:       "council-of-readers",
216: Name:     "Council of Readers",
217: Category: "Consensus Protocols",
218: Description: "Multi-model ensemble that aggregates opinions from diverse AI models to produce consensus coverage with disagreement analysis.",
267: {Order: 1, Name: "Distribution", Description: "Send screenplay to multiple models simultaneously", Models: []string{"coordinator"}, Parallel: false},
268: {Order: 2, Name: "Claude Analysis", Description: "Full rubric analysis with Claude Opus", Models: []string{"claude-opus"}, Parallel: true},
269: {Order: 3, Name: "Llama Analysis", Description: "Full rubric analysis with Llama 3.3 70B", Models: []string{"llama-3.3-70b"}, Parallel: true},
270: {Order: 4, Name: "Command-R Analysis", Description: "Full rubric analysis with Command-R+", Models: []string{"command-r-plus"}, Parallel: true},
271: {Order: 5, Name: "Aggregation", Description: "Aggregate results using weighted consensus algorithm", Models: []string{"claude-sonnet"}, Parallel: false},
272: {Order: 6, Name: "Disagreement Analysis", Description: "Analyze and report on areas of model disagreement", Models: []string{"claude-opus"}, Parallel: false},
005studio-gradeproduction protocol requirementsDuplicates studio-grade in the analysis-protocol registry. Giving this requirements record a second byte creates two codes for the same method.
Candidate definitioninternal/production/protocols.go:58
58: "studio-grade": {
59: ID:             "studio-grade",
60: Name:           "Studio Grade",
61: Category:       "Comprehensive",
62: RequiredModels: []string{"claude-3-opus"},
63: MinGPUMemoryGB: 80,
Comparison definitioncmd/coverage_protocol_defs.go:891
891: ID:       "studio-grade",
892: Name:     "Studio Grade",
893: Category: "Comprehensive",
894: Description: "Full professional coverage matching major studio standards with all documentation and database integration.",
944: {Order: 1, Name: "Extraction", Description: "Parse screenplay and create database records", Models: []string{"local"}, Parallel: false},
945: {Order: 2, Name: "Multi-Model Primary", Description: "Parallel analysis with multiple models", Models: []string{"claude-opus", "llama-3.3-70b"}, Parallel: true},
946: {Order: 3, Name: "Full Rubric", Description: "Complete 16-point rubric analysis", Models: []string{"claude-opus"}, Parallel: false},
947: {Order: 4, Name: "Market Analysis", Description: "Commercial viability and positioning", Models: []string{"claude-opus"}, Parallel: true},
948: {Order: 5, Name: "Synthesis", Description: "Aggregate and reconcile all analyses", Models: []string{"claude-opus"}, Parallel: false},
949: {Order: 6, Name: "Audit", Description: "Verify claims and check for bias", Models: []string{"claude-opus"}, Parallel: false},
006audited-coverageproduction protocol requirementsDuplicates audited-coverage in the analysis-protocol registry. Giving this requirements record a second byte creates two codes for the same method.
Candidate definitioninternal/production/protocols.go:67
67: "audited-coverage": {
68: ID:             "audited-coverage",
69: Name:           "Audited Coverage",
70: Category:       "Quality Assurance",
71: RequiredModels: []string{"claude-3-opus", "claude-3-sonnet"},
72: MinGPUMemoryGB: 60,
88: RequiredModels: requirements.RequiredModels,
Comparison definitioncmd/coverage_protocol_defs.go:742
742: ID:       "audited-coverage",
743: Name:     "Audited Coverage",
744: Category: "Quality Assurance",
745: Description: "Coverage with built-in audit layer that verifies claims, checks for bias, and produces confidence scores.",
799: {Order: 1, Name: "Primary Analysis", Description: "Generate full coverage report", Models: []string{"claude-opus"}, Parallel: false},
800: {Order: 2, Name: "Claim Extraction", Description: "Extract all factual claims from coverage", Models: []string{"claude-sonnet"}, Parallel: false},
801: {Order: 3, Name: "Verification", Description: "Check each claim against screenplay text", Models: []string{"claude-opus"}, Parallel: true},
802: {Order: 4, Name: "Bias Detection", Description: "Scan for potential biases in analysis", Models: []string{"claude-opus"}, Parallel: true},
803: {Order: 5, Name: "Consistency Audit", Description: "Check internal consistency of coverage", Models: []string{"claude-sonnet"}, Parallel: false},
804: {Order: 6, Name: "Certification", Description: "Generate certified coverage with audit trail", Models: []string{"claude-opus"}, Parallel: false},
007gpu-setupsuggested workflowsThe guided workflow registry already contains gpu-setup. A separate byte would duplicate an existing command intent.
Candidate definitioncmd/suggest.go:65
65: Name:        "gpu-setup",
66: Description: "Check GPU availability and system specs before deployment",
71: Category: "GPU & Model Setup",
Comparison definitioninternal/workflows/registry.go:76
76: Name:          "gpu-setup",
78: Description:   "Set up NVIDIA GPU drivers, CUDA toolkit, and verify GPU access for ML workloads",
79: Category:      "Installation & Setup",
83: {Description: "Check for existing NVIDIA driver", Command: "nvidia-smi", ExpectedOutcome: "GPU information displayed with driver version", Troubleshooting: "If command not found, drivers need to be installed"},
84: {Description: "Install NVIDIA drivers (if needed)", Command: "sudo apt install -y nvidia-driver-535", ExpectedOutcome: "NVIDIA driver 535 installed", Troubleshooting: "Reboot required after driver installation"},
85: {Description: "Install CUDA toolkit", Command: "sudo apt install -y nvidia-cuda-toolkit", ExpectedOutcome: "CUDA toolkit installed", Troubleshooting: "Check version with 'nvcc --version'"},
86: {Description: "Verify CUDA installation", Command: "nvcc --version && nvidia-smi", ExpectedOutcome: "CUDA compiler version and GPU status shown", Troubleshooting: "Ensure CUDA_HOME is set in environment"},
87: {Description: "Test GPU compute capability", Command: "python3 -c \"import torch; print(f'CUDA available: {torch.cuda.is_available()}'); print(f'GPU: {torch.cuda.get_device_name(0)}')\"", ExpectedOutcome: "PyTorch detects CUDA and GPU", Troubleshooting: "Install pytorch with CUDA: pip install torch --index-url https://download.pytorch.org/whl/cu121"},
88: {Description: "Check GPU memory", Command: "nvidia-smi --query-gpu=memory.total,memory.free --format=csv", ExpectedOutcome: "Total and free GPU memory displayed", Troubleshooting: "Free up memory by stopping other GPU processes"},
93: Name:          "docker-setup",
008model-deploysuggested workflowsModel deployment is already represented by model-setup and deployment workflows. The extra entry encodes a target or setting rather than a distinct workflow.
Candidate definitioncmd/suggest.go:74
74: Name:        "model-deploy",
75: Description: "Download and deploy a model with vLLM on your GPU cluster",
81: Category: "GPU & Model Setup",
Comparison definitioninternal/workflows/registry.go:110
110: Name:          "model-setup",
112: Description:   "Download and configure your first model for inference with conduct",
113: Category:      "Installation & Setup",
117: {Description: "List available models", Command: "conduct models list", ExpectedOutcome: "List of supported models displayed", Troubleshooting: "Check internet connection if list is empty"},
118: {Description: "Set HuggingFace token (for gated models)", Command: "export HF_TOKEN=your_token_here", ExpectedOutcome: "HF_TOKEN environment variable set", Troubleshooting: "Get token from huggingface.co/settings/tokens"},
119: {Description: "Download a starter model", Command: "conduct models download microsoft/phi-2", ExpectedOutcome: "Model downloaded to ~/.cache/huggingface", Troubleshooting: "Try a smaller model if disk space is limited"},
120: {Description: "Verify model download", Command: "ls -la ~/.cache/huggingface/hub/models--microsoft--phi-2", ExpectedOutcome: "Model files present in cache directory", Troubleshooting: "Re-run download if files are missing"},
121: {Description: "Load model into vLLM", Command: "conduct load microsoft/phi-2", ExpectedOutcome: "Model loaded and ready for inference", Troubleshooting: "Ensure sufficient GPU memory for model size"},
126: Name:          "vllm-setup",
128: Description:   "Install and configure vLLM server for high-performance model inference",
009quick-llamasuggested workflowsA quick-start for one model is a parameter choice, not a new general intent. The extra entry encodes a target or setting rather than a distinct workflow.
Candidate definitioncmd/suggest.go:84
84: Name:        "quick-llama",
85: Description: "Fastest path to running Llama 3.3 70B with optimal settings",
90: Category: "GPU & Model Setup",
Comparison definitioninternal/workflows/registry.go:164
164: Name:          "load-model",
166: Description:   "Load a downloaded model into vLLM for inference",
167: Category:      "Model Management",
171: {Description: "List downloaded models", Command: "conduct models list --local", ExpectedOutcome: "Local models displayed with sizes", Troubleshooting: "Download a model first if list is empty"},
172: {Description: "Check GPU memory availability", Command: "nvidia-smi --query-gpu=memory.free --format=csv,noheader", ExpectedOutcome: "Free GPU memory shown in MiB", Troubleshooting: "Unload other models if memory is low"},
173: {Description: "Load the model", Command: "conduct load <model-name>", ExpectedOutcome: "Model loaded, server ready for inference", Troubleshooting: "Use --gpu-memory-utilization 0.8 for memory constraints"},
174: {Description: "Verify model is loaded", Command: "conduct models status", ExpectedOutcome: "Model shown as active/loaded", Troubleshooting: "Check server logs for loading errors"},
175: {Description: "Test model inference", Command: "conduct infer \"Hello, how are you?\"", ExpectedOutcome: "Model generates response", Troubleshooting: "Ensure model is fully loaded before inference"},
180: Name:          "switch-model",
182: Description:   "Unload current model and load a different one",
010multi-gpusuggested workflowsGPU count and parallelism belong in the deployment plan. The extra entry encodes a target or setting rather than a distinct workflow.
Candidate definitioncmd/suggest.go:93
93: Name:        "multi-gpu",
94: Description: "Configure tensor parallelism for large models across GPUs",
100: Category: "GPU & Model Setup",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
011server-setupsuggested workflowsServer preparation overlaps fresh-install and GPU/model setup. It adds another name for a sequence already expressible as fresh-install → gpu-setup → model-setup.
Candidate definitioncmd/suggest.go:105
105: Name:        "server-setup",
106: Description: "Register and configure a remote GPU server for deployments",
113: Category: "Server Management",
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
012remote-monitorsuggested workflowsRemote location changes the target, not the monitoring workflow. The extra entry encodes a target or setting rather than a distinct workflow.
Candidate definitioncmd/suggest.go:116
116: Name:        "remote-monitor",
117: Description: "Monitor GPU usage on a remote server in real-time",
122: Category: "Server Management",
Comparison definitioninternal/workflows/registry.go:426
426: Name:          "start-monitoring",
428: Description:   "Set up comprehensive monitoring for conduct services",
429: Category:      "Monitoring & Observability",
433: {Description: "Start GPU monitoring", Command: "conduct monitor gpu --continuous", ExpectedOutcome: "GPU metrics collection started", Troubleshooting: "Requires nvidia-smi to be available"},
434: {Description: "Start service monitoring", Command: "conduct monitor services", ExpectedOutcome: "Service health monitoring active", Troubleshooting: "Monitors vLLM, API, and inference services"},
435: {Description: "Enable metrics export", Command: "conduct monitor export --prometheus", ExpectedOutcome: "Prometheus metrics endpoint available", Troubleshooting: "Default endpoint: :9090/metrics"},
436: {Description: "Configure dashboards", Command: "conduct monitor dashboard --setup", ExpectedOutcome: "Monitoring dashboard accessible", Troubleshooting: "Access at http://localhost:3000"},
437: {Description: "Verify monitoring", Command: "conduct monitor status", ExpectedOutcome: "All monitors active", Troubleshooting: "Check individual monitor logs for issues"},
442: Name:          "gpu-dashboard",
444: Description:   "Launch and configure GPU monitoring dashboard",
013fleet-statussuggested workflowsA fleet is a target set for status collection. The extra entry encodes a target or setting rather than a distinct workflow.
Candidate definitioncmd/suggest.go:125
125: Name:        "fleet-status",
126: Description: "Check status across all registered servers",
132: Category: "Server Management",
014remote-devsuggested workflowsA developer convenience sequence can reference the existing server and environment setup. It adds another name for a sequence already expressible as prepare environment → verify access.
Candidate definitioncmd/suggest.go:135
135: Name:        "remote-dev",
136: Description: "Deploy binary and source code for remote GPU development",
144: Category: "Server Management",
015full-coveragesuggested workflowsThis guide wraps existing screenplay analysis capabilities. It adds another name for a sequence already expressible as selected analysis protocol with a source reference.
Candidate definitioncmd/suggest.go:149
149: Name:        "full-coverage",
150: Description: "Run complete screenplay analysis with all 16 rubric points",
155: Category: "Screenplay Coverage",
Comparison definitioncmd/coverage_protocol_defs.go:891
891: ID:       "studio-grade",
892: Name:     "Studio Grade",
893: Category: "Comprehensive",
894: Description: "Full professional coverage matching major studio standards with all documentation and database integration.",
944: {Order: 1, Name: "Extraction", Description: "Parse screenplay and create database records", Models: []string{"local"}, Parallel: false},
945: {Order: 2, Name: "Multi-Model Primary", Description: "Parallel analysis with multiple models", Models: []string{"claude-opus", "llama-3.3-70b"}, Parallel: true},
946: {Order: 3, Name: "Full Rubric", Description: "Complete 16-point rubric analysis", Models: []string{"claude-opus"}, Parallel: false},
947: {Order: 4, Name: "Market Analysis", Description: "Commercial viability and positioning", Models: []string{"claude-opus"}, Parallel: true},
948: {Order: 5, Name: "Synthesis", Description: "Aggregate and reconcile all analyses", Models: []string{"claude-opus"}, Parallel: false},
949: {Order: 6, Name: "Audit", Description: "Verify claims and check for bias", Models: []string{"claude-opus"}, Parallel: false},
016quick-analysissuggested workflowsQuickness can be an analysis depth or budget setting. It adds another name for a sequence already expressible as lightning-pass or a bounded analysis method.
Candidate definitioncmd/suggest.go:158
158: Name:        "quick-analysis",
159: Description: "Fast screenplay assessment focusing on key elements",
165: Category: "Screenplay Coverage",
Comparison definitioncmd/coverage_protocol_defs.go:364
364: ID:       "lightning-pass",
365: Name:     "Lightning Pass",
366: Category: "Speed Optimized",
367: Description: "Ultra-fast coverage using parallel execution and smaller models for rapid decision-making.",
419: {Order: 1, Name: "Parallel Rubric", Description: "All 7 core rubric points analyzed simultaneously", Models: []string{"claude-haiku"}, Parallel: true},
420: {Order: 2, Name: "Quick Aggregate", Description: "Fast aggregation of parallel results", Models: []string{"claude-haiku"}, Parallel: false},
421: {Order: 3, Name: "One-Page Summary", Description: "Generate actionable one-page coverage", Models: []string{"claude-sonnet"}, Parallel: false},
424: Outputs: []string{"lightning_coverage.md", "scores.json"},
017market-evalsuggested workflowsMarket evaluation overlaps the richer market-positioning analysis protocol. It adds another name for a sequence already expressible as market-positioning.
Candidate definitioncmd/suggest.go:168
168: Name:        "market-eval",
169: Description: "Evaluate screenplay's commercial potential and market fit",
175: Category: "Screenplay Coverage",
Comparison definitioncmd/coverage_protocol_defs.go:662
662: ID:       "market-positioning",
663: Name:     "Market Positioning",
664: Category: "Specialized Analysis",
665: Description: "Comprehensive market analysis with comparable films, audience targeting, and commercial viability assessment.",
719: {Order: 1, Name: "Genre Classification", Description: "Classify genre, tone, and style", Models: []string{"claude-sonnet"}, Parallel: false},
720: {Order: 2, Name: "Comparable Search", Description: "Identify comparable films with performance data", Models: []string{"claude-opus"}, Parallel: true},
721: {Order: 3, Name: "Audience Analysis", Description: "Define target demographics and platforms", Models: []string{"claude-opus"}, Parallel: true},
722: {Order: 4, Name: "Trend Analysis", Description: "Assess market timing and trend alignment", Models: []string{"claude-opus"}, Parallel: true},
723: {Order: 5, Name: "Viability Assessment", Description: "Calculate commercial viability score", Models: []string{"claude-opus"}, Parallel: false},
724: {Order: 6, Name: "Report Generation", Description: "Generate market positioning report and one-sheet", Models: []string{"claude-opus"}, Parallel: false},
018experiment-runsuggested workflowsThis points the operator into the existing experiment lifecycle. It adds another name for a sequence already expressible as prepare experiment → run → compare.
Candidate definitioncmd/suggest.go:180
180: Name:        "experiment-run",
181: Description: "Set up and run a multi-model comparison experiment",
187: Category: "Development & Testing",
019client-testsuggested workflowsA client check fits ordinary health or integration verification. It adds another name for a sequence already expressible as health-check / integration-test.
Candidate definitioncmd/suggest.go:190
190: Name:        "client-test",
191: Description: "Test model inference with the AI client",
197: Category: "Development & Testing",
Comparison definitioninternal/workflows/registry.go:540
540: Name:          "integration-test",
542: Description:   "Execute integration tests to verify component interactions",
543: Category:      "Testing & Validation",
547: {Description: "Run end-to-end test", Command: "conduct test end-to-end run -v", ExpectedOutcome: "E2E workflow completes", Troubleshooting: "Check each step for failures"},
548: {Description: "Run API integration test", Command: "conduct test api-integration run", ExpectedOutcome: "API endpoints working", Troubleshooting: "Verify API server is running"},
549: {Description: "Run sequence integration", Command: "conduct test sequence-integration run", ExpectedOutcome: "Sequences execute correctly", Troubleshooting: "Check sequence definitions"},
550: {Description: "Test concurrent access", Command: "conduct test concurrent-requests run", ExpectedOutcome: "Handles concurrent load", Troubleshooting: "Adjust concurrency level if failing"},
561: Name:          "diagnose-gpu",
563: Description:   "Troubleshoot GPU-related problems",
564: Category:      "Troubleshooting",
020initial-setupsuggested workflowsAn initial setup guide overlaps the fresh-install workflow. A separate byte would duplicate an existing command intent.
Candidate definitioncmd/suggest.go:202
202: Name:        "initial-setup",
203: Description: "First-time setup: detect tokens and initialize database",
209: Category: "Configuration & Utilities",
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
021dashboard-launchsuggested workflowsOpening a dashboard is one UI action, not an orchestration protocol. It does not justify a workflow-table slot.
Candidate definitioncmd/suggest.go:212
212: Name:        "dashboard-launch",
213: Description: "Start the web interface for visual management",
218: Category: "Configuration & Utilities",
022backup-configsuggested workflowsA configuration backup is a particular source for the general backup workflow. It adds another name for a sequence already expressible as snapshot → copy → verify recovery.
Candidate definitioncmd/suggest.go:221
221: Name:        "backup-config",
222: Description: "Export configuration and registry for backup",
228: Category: "Configuration & Utilities",
023general llama-70b-tp4deployment presetsNames a specific model/parallelism profile (llama-70b-tp4). A new model size or allocation would demand another code despite an unchanged deployment workflow.
Candidate definitioncmd/lambda_workflows_data.go:42
42: "general llama-70b-tp4": {
43: Name:        "general llama-70b-tp4",
44: Description: "Single Llama 3.3 70B with TP4 - balanced performance, 4 GPUs free",
45: GPUType:     "B200",
46: GPUCount:    8,
47: GPUMemoryGB: 180,
50: Name:         "Llama-70B",
51: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
52: TP:           4,
63: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
024general dual-llama-70b-tp4deployment presetsNames a specific model/parallelism profile (dual-llama-70b-tp4). A new model size or allocation would demand another code despite an unchanged deployment workflow.
Candidate definitioncmd/lambda_workflows_data.go:83
83: "general dual-llama-70b-tp4": {
84: Name:        "general dual-llama-70b-tp4",
85: Description: "Dual Llama 3.3 70B instances with TP4 each - high throughput",
86: GPUType:     "B200",
87: GPUCount:    8,
88: GPUMemoryGB: 180,
91: Name:         "Llama-70B-A",
92: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
93: TP:           4,
100: Name:         "Llama-70B-B",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
025general jamba-1.7-tp8deployment presetsNames a specific model/parallelism profile (jamba-1.7-tp8). A new model size or allocation would demand another code despite an unchanged deployment workflow.
Candidate definitioncmd/lambda_workflows_data.go:144
144: "general jamba-1.7-tp8": {
145: Name:        "general jamba-1.7-tp8",
146: Description: "Full Jamba 1.7 398B - maximum capability, uses all 8 GPUs",
147: GPUType:     "B200",
148: GPUCount:    8,
149: GPUMemoryGB: 180,
152: Name:         "Jamba-1.7-Large",
153: ModelID:      "ai21labs/AI21-Jamba-Large-1.7-FP8",
154: TP:           8,
165: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
026general jamba-1.5-large-tp4deployment presetsNames a specific model/parallelism profile (jamba-1.5-large-tp4). A new model size or allocation would demand another code despite an unchanged deployment workflow.
Candidate definitioncmd/lambda_workflows_data.go:194
194: "general jamba-1.5-large-tp4": {
195: Name:        "general jamba-1.5-large-tp4",
196: Description: "Jamba 1.5 Large 94B with TP4 - hybrid SSM with 256K context",
197: GPUType:     "B200",
198: GPUCount:    8,
199: GPUMemoryGB: 180,
202: Name:         "Jamba-1.5-Large",
203: ModelID:      "ai21labs/AI21-Jamba-1.5-Large",
204: TP:           4,
215: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
027enterprise command-r-plus-tp4deployment presetsThe command-r-plus-tp4 model selection changes configuration, not execution order. It duplicates the general deployment workflow.
Candidate definitioncmd/lambda_workflows_data.go:236
236: "enterprise command-r-plus-tp4": {
237: Name:        "enterprise command-r-plus-tp4",
238: Description: "Command-R+ 104B - enterprise RAG and tool use, 4 GPUs free",
239: GPUType:     "B200",
240: GPUCount:    8,
241: GPUMemoryGB: 180,
244: Name:         "Command-R-Plus",
245: ModelID:      "CohereForAI/c4ai-command-r-plus",
246: TP:           4,
257: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
028enterprise command-r-plus-llamadeployment presetsThe command-r-plus-llama model selection changes configuration, not execution order. It duplicates the general deployment workflow.
Candidate definitioncmd/lambda_workflows_data.go:274
274: "enterprise command-r-plus-llama": {
275: Name:        "enterprise command-r-plus-llama",
276: Description: "Command-R+ for RAG + Llama 70B for general tasks",
277: GPUType:     "B200",
278: GPUCount:    8,
279: GPUMemoryGB: 180,
282: Name:         "Command-R-Plus",
283: ModelID:      "CohereForAI/c4ai-command-r-plus",
284: TP:           4,
291: Name:         "Llama-70B",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
029code focuseddeployment presetsCoding/development specialization is a model-profile choice. This entry spends a workflow code on that profile without adding a distinct orchestration pattern.
Candidate definitioncmd/lambda_workflows_data.go:337
337: "code focused": {
338: Name:        "code focused",
339: Description: "Mamba-Codestral for code + Llama 70B for reasoning",
340: GPUType:     "B200",
341: GPUCount:    8,
342: GPUMemoryGB: 180,
345: Name:         "Mamba-Codestral",
346: ModelID:      "mistralai/mamba-codestral-7B-v0.1",
347: TP:           1,
354: Name:         "Llama-70B",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
030code multideployment presetsCoding/development specialization is a model-profile choice. This entry spends a workflow code on that profile without adding a distinct orchestration pattern.
Candidate definitioncmd/lambda_workflows_data.go:396
396: "code multi": {
397: Name:        "code multi",
398: Description: "Multiple code models: Codestral + Phi-4 + Mistral-Small",
399: GPUType:     "B200",
400: GPUCount:    8,
401: GPUMemoryGB: 180,
404: Name:         "Mamba-Codestral",
405: ModelID:      "mistralai/mamba-codestral-7B-v0.1",
406: TP:           1,
413: Name:         "Phi-4",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
031dev multi-smalldeployment presetsCoding/development specialization is a model-profile choice. This entry spends a workflow code on that profile without adding a distinct orchestration pattern.
Candidate definitioncmd/lambda_workflows_data.go:481
481: "dev multi-small": {
482: Name:        "dev multi-small",
483: Description: "Multiple small models for development and A/B testing",
484: GPUType:     "B200",
485: GPUCount:    8,
486: GPUMemoryGB: 180,
489: Name:         "Phi-4",
490: ModelID:      "microsoft/phi-4",
491: TP:           1,
498: Name:         "Mistral-Small",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
032coverage paralleldeployment presetsThe parallel coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:610
610: "coverage parallel": {
611: Name:        "coverage parallel",
612: Description: "4x Llama-70B TP2 for parallel analysis",
613: GPUType:     "B200",
614: GPUCount:    8,
615: GPUMemoryGB: 180,
618: Name:         "Llama-70B-0",
619: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
620: TP:           2,
627: Name:         "Llama-70B-1",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
033coverage fulldeployment presetsThe full coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:691
691: "coverage full": {
692: Name:        "coverage full",
693: Description: "Jamba-1.5-Large TP4 + Llama-70B TP2 + Mistral-Small TP2",
694: GPUType:     "B200",
695: GPUCount:    8,
696: GPUMemoryGB: 180,
699: Name:         "Jamba-1.5-Large",
700: ModelID:      "ai21labs/AI21-Jamba-1.5-Large",
701: TP:           4,
708: Name:         "Llama-70B",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
034coverage characterdeployment presetsThe character coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:761
761: "coverage character": {
762: Name:        "coverage character",
763: Description: "4x Llama-70B TP2 for parallel character analysis",
764: GPUType:     "B200",
765: GPUCount:    8,
766: GPUMemoryGB: 180,
769: Name:         "Llama-70B-0",
770: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
771: TP:           2,
778: Name:         "Llama-70B-1",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
035coverage structuredeployment presetsThe structure coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:842
842: "coverage structure": {
843: Name:        "coverage structure",
844: Description: "Jamba-1.5-Large TP4 + 2x Llama-70B TP2 for structure analysis",
845: GPUType:     "B200",
846: GPUCount:    8,
847: GPUMemoryGB: 180,
850: Name:         "Jamba-1.5-Large",
851: ModelID:      "ai21labs/AI21-Jamba-1.5-Large",
852: TP:           4,
859: Name:         "Llama-70B-0",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
036coverage comparativedeployment presetsThe comparative coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:911
911: "coverage comparative": {
912: Name:        "coverage comparative",
913: Description: "Multi-model comparison: analyze same screenplay with different models",
914: GPUType:     "B200",
915: GPUCount:    8,
916: GPUMemoryGB: 180,
919: Name:         "Llama-70B",
920: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
921: TP:           2,
928: Name:         "Jamba-1.5-Large",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
037coverage scenedeployment presetsThe scene coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:992
992: "coverage scene": {
993: Name:        "coverage scene",
994: Description: "8x Mistral-Small-24B for massive parallel processing",
995: GPUType:     "B200",
996: GPUCount:    8,
997: GPUMemoryGB: 180,
1000: Name:         "Mistral-Small-0",
1001: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
1002: TP:           1,
1009: Name:         "Mistral-Small-1",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
038coverage dialoguedeployment presetsThe dialogue coverage slice or layout is a profile of an existing analysis/deployment process. A dedicated code duplicates that process.
Candidate definitioncmd/lambda_workflows_data.go:1105
1105: "coverage dialogue": {
1106: Name:        "coverage dialogue",
1107: Description: "4x Llama-70B TP2 for parallel dialogue analysis",
1108: GPUType:     "B200",
1109: GPUCount:    8,
1110: GPUMemoryGB: 180,
1113: Name:         "Llama-70B-0",
1114: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
1115: TP:           2,
1122: Name:         "Llama-70B-1",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
039gh200 llama-70bdeployment presetsHardcodes GH200 hardware and the llama-70b profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:1940
1940: "gh200 llama-70b": {
1941: Name:        "gh200 llama-70b",
1942: Description: "Llama 3.3 70B on GH200 - full precision BF16, single GPU",
1943: GPUType:     "GH200",
1944: GPUCount:    1,
1945: GPUMemoryGB: 141,
1948: Name:         "Llama-70B",
1949: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
1950: TP:           1,
1961: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
040gh200 jamba-1.5-largedeployment presetsHardcodes GH200 hardware and the jamba-1.5-large profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:1981
1981: "gh200 jamba-1.5-large": {
1982: Name:        "gh200 jamba-1.5-large",
1983: Description: "Jamba 1.5 Large 94B on GH200 - hybrid SSM with 256K context",
1984: GPUType:     "GH200",
1985: GPUCount:    1,
1986: GPUMemoryGB: 141,
1989: Name:         "Jamba-1.5-Large",
1990: ModelID:      "ai21labs/AI21-Jamba-1.5-Large",
1991: TP:           1,
2002: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
041gh200 command-r-plusdeployment presetsHardcodes GH200 hardware and the command-r-plus profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2019
2019: "gh200 command-r-plus": {
2020: Name:        "gh200 command-r-plus",
2021: Description: "Command-R+ 104B on GH200 - enterprise RAG and tool use",
2022: GPUType:     "GH200",
2023: GPUCount:    1,
2024: GPUMemoryGB: 141,
2027: Name:         "Command-R-Plus",
2028: ModelID:      "CohereForAI/c4ai-command-r-plus",
2029: TP:           1,
2040: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
042gh200 qwen-72bdeployment presetsHardcodes GH200 hardware and the qwen-72b profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2057
2057: "gh200 qwen-72b": {
2058: Name:        "gh200 qwen-72b",
2059: Description: "Qwen 2.5 72B on GH200 - strong multilingual and code capabilities",
2060: GPUType:     "GH200",
2061: GPUCount:    1,
2062: GPUMemoryGB: 141,
2065: Name:         "Qwen-72B",
2066: ModelID:      "Qwen/Qwen2.5-72B-Instruct",
2067: TP:           1,
2078: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
043gh200 code deepseekdeployment presetsHardcodes GH200 hardware and the code deepseek profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2095
2095: "gh200 code deepseek": {
2096: Name:        "gh200 code deepseek",
2097: Description: "DeepSeek Coder V2 Lite on GH200 - optimized for code generation",
2098: GPUType:     "GH200",
2099: GPUCount:    1,
2100: GPUMemoryGB: 141,
2103: Name:         "DeepSeek-Coder",
2104: ModelID:      "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct",
2105: TP:           1,
2116: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
044gh200 code qwendeployment presetsHardcodes GH200 hardware and the code qwen profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2133
2133: "gh200 code qwen": {
2134: Name:        "gh200 code qwen",
2135: Description: "Qwen 2.5 Coder 32B on GH200 - specialized coding model",
2136: GPUType:     "GH200",
2137: GPUCount:    1,
2138: GPUMemoryGB: 141,
2141: Name:         "Qwen-Coder-32B",
2142: ModelID:      "Qwen/Qwen2.5-Coder-32B-Instruct",
2143: TP:           1,
2154: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
045gh200 mistral-smalldeployment presetsHardcodes GH200 hardware and the mistral-small profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2171
2171: "gh200 mistral-small": {
2172: Name:        "gh200 mistral-small",
2173: Description: "Mistral Small 24B on GH200 - fast, efficient general purpose",
2174: GPUType:     "GH200",
2175: GPUCount:    1,
2176: GPUMemoryGB: 141,
2179: Name:         "Mistral-Small",
2180: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
2181: TP:           1,
2192: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
046gh200 phi-4deployment presetsHardcodes GH200 hardware and the phi-4 profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2209
2209: "gh200 phi-4": {
2210: Name:        "gh200 phi-4",
2211: Description: "Microsoft Phi-4 14B on GH200 - efficient reasoning model",
2212: GPUType:     "GH200",
2213: GPUCount:    1,
2214: GPUMemoryGB: 141,
2217: Name:         "Phi-4",
2218: ModelID:      "microsoft/phi-4",
2219: TP:           1,
2230: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
047gh200 llama-70b-awqdeployment presetsHardcodes GH200 hardware and the llama-70b-awq profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2247
2247: "gh200 llama-70b-awq": {
2248: Name:        "gh200 llama-70b-awq",
2249: Description: "Llama 3.3 70B AWQ quantized on GH200 - more memory headroom for context",
2250: GPUType:     "GH200",
2251: GPUCount:    1,
2252: GPUMemoryGB: 141,
2255: Name:         "Llama-70B-AWQ",
2256: ModelID:      "hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4",
2257: TP:           1,
2268: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
048gh200 dual mistral-phideployment presetsHardcodes GH200 hardware and the dual mistral-phi profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2285
2285: "gh200 dual mistral-phi": {
2286: Name:        "gh200 dual mistral-phi",
2287: Description: "Mistral Small + Phi-4 on GH200 - dual model for coverage analysis",
2288: GPUType:     "GH200",
2289: GPUCount:    1,
2290: GPUMemoryGB: 141,
2293: Name:         "Mistral-Small",
2294: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
2295: TP:           1,
2302: Name:         "Phi-4",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
049gh200 dual qwen-coderdeployment presetsHardcodes GH200 hardware and the dual qwen-coder profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2334
2334: "gh200 dual qwen-coder": {
2335: Name:        "gh200 dual qwen-coder",
2336: Description: "Qwen 2.5 Coder 32B + DeepSeek Coder on GH200 - code-focused dual model",
2337: GPUType:     "GH200",
2338: GPUCount:    1,
2339: GPUMemoryGB: 141,
2342: Name:         "Qwen-Coder-32B",
2343: ModelID:      "Qwen/Qwen2.5-Coder-32B-Instruct",
2344: TP:           1,
2351: Name:         "DeepSeek-Coder",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
050gh200 coveragedeployment presetsHardcodes GH200 hardware and the coverage profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2383
2383: "gh200 coverage": {
2384: Name:        "gh200 coverage",
2385: Description: "Llama 70B AWQ on GH200 - optimized for coverage analysis",
2386: GPUType:     "GH200",
2387: GPUCount:    1,
2388: GPUMemoryGB: 141,
2391: Name:         "Coverage-Analyzer",
2392: ModelID:      "hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4",
2393: TP:           1,
2404: Name:     "Model Download",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
051gh200 coverage dualdeployment presetsHardcodes GH200 hardware and the coverage dual profile. The code becomes tied to one deployment configuration rather than a reusable instruction.
Candidate definitioncmd/lambda_workflows_data.go:2421
2421: "gh200 coverage dual": {
2422: Name:        "gh200 coverage dual",
2423: Description: "Llama 70B AWQ + Qwen Coder on GH200 - dual model coverage analysis",
2424: GPUType:     "GH200",
2425: GPUCount:    1,
2426: GPUMemoryGB: 141,
2429: Name:         "Primary-Analyzer",
2430: ModelID:      "hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4",
2431: TP:           1,
2438: Name:         "Code-Analyzer",
Comparison definitioninternal/workflows/registry.go:388
388: Name:          "deploy-cluster",
390: Description:   "Deploy conduct across multiple machines for distributed inference",
391: Category:      "Deployment",
395: {Description: "Define cluster", Command: "conduct cluster create --name ml-cluster --nodes node1,node2,node3", ExpectedOutcome: "Cluster configuration created", Troubleshooting: "Ensure all nodes are in server registry"},
396: {Description: "Validate cluster config", Command: "conduct cluster validate ml-cluster", ExpectedOutcome: "All nodes reachable and compatible", Troubleshooting: "Fix connectivity issues before deploying"},
397: {Description: "Deploy to cluster", Command: "conduct cluster deploy ml-cluster", ExpectedOutcome: "Deployment replicated to all nodes", Troubleshooting: "Use --parallel for faster deployment"},
398: {Description: "Configure load balancing", Command: "conduct cluster lb-config ml-cluster --strategy round-robin", ExpectedOutcome: "Load balancer configured", Troubleshooting: "Options: round-robin, least-connections, gpu-aware"},
399: {Description: "Test cluster", Command: "conduct cluster test ml-cluster", ExpectedOutcome: "All nodes responding correctly", Troubleshooting: "Check individual node status for failures"},
404: Name:          "lambda-deploy",
406: Description:   "Deploy inference endpoint as a serverless Lambda function",
052vllminstallation sequencesOne runtime installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:711
711: r.Sequences["vllm"] = buildSequence(
712: "vllm",
713: "vLLM Full Stack",
714: "Complete vLLM installation with NVIDIA drivers, CUDA, PyTorch",
715: "~15GB",
716: []string{"core", "python", "nvidia", "pytorch", "vllm"},
717: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
053pytorchinstallation sequencesOne framework installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:719
719: r.Sequences["pytorch"] = buildSequence(
720: "pytorch",
721: "PyTorch GPU Stack",
722: "PyTorch with CUDA support for deep learning",
723: "~12GB",
724: []string{"core", "python", "nvidia", "pytorch"},
725: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
054llamainstallation sequencesOne model-family installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:731
731: r.Sequences["llama"] = buildSequence(
732: "llama",
733: "Llama 3.3 70B on vLLM",
734: "Complete stack with Llama 3.3 70B download and GPU-optimized loading",
735: "~155GB",
736: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "huggingface-auth", "hf-download-llama-70b", "vllm-load-llama-70b"},
737: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
055llama-8binstallation sequencesA size-specific variant of the Llama installation sequence. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:739
739: r.Sequences["llama-8b"] = buildSequence(
740: "llama-8b",
741: "Llama 3.1 8B Quick Start",
742: "Lightweight Llama 8B for testing - fits on single GPU",
743: "~31GB",
744: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "huggingface-auth", "hf-download-llama-8b", "vllm-serve-llama-8b"},
745: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
056mistralinstallation sequencesOne model-family installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:747
747: r.Sequences["mistral"] = buildSequence(
748: "mistral",
749: "Mistral Small 24B",
750: "Fast and capable Mistral Small model with vLLM",
751: "~63GB",
752: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "hf-download-mistral-small", "vllm-serve-mistral"},
753: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
057qweninstallation sequencesOne model-family installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:755
755: r.Sequences["qwen"] = buildSequence(
756: "qwen",
757: "Qwen 2.5 72B",
758: "Alibaba's powerful Qwen 2.5 72B Instruct model",
759: "~159GB",
760: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "hf-download-qwen-72b"},
761: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
058deepseekinstallation sequencesOne model-family installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:763
763: r.Sequences["deepseek"] = buildSequence(
764: "deepseek",
765: "DeepSeek Coder V2",
766: "DeepSeek's coding-optimized model for code generation",
767: "~215GB",
768: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "hf-download-deepseek-coder"},
769: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
059multi-modelinstallation sequencesA model-set parameter on a dependency-aware installation plan. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:771
771: r.Sequences["multi-model"] = buildSequence(
772: "multi-model",
773: "Multi-Model Setup",
774: "Multiple models for A/B testing: Llama 8B + Mistral Small",
775: "~80GB",
776: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "multi-model-setup", "serve-multi"},
777: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
060claude-codeinstallation sequencesA particular developer-tool installation target. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:783
783: r.Sequences["claude-code"] = buildSequence(
784: "claude-code",
785: "Claude Code CLI",
786: "Install Anthropic's official AI coding assistant",
787: "~350MB",
788: []string{"core", "nodejs", "claude-code"},
789: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
061devinstallation sequencesA development-environment package profile. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:791
791: r.Sequences["dev"] = buildSequence(
792: "dev",
793: "Development Environment",
794: "Full dev setup: Go, Node.js, Python, Docker, Git",
795: "~2GB",
796: []string{"core", "python", "golang", "nodejs", "docker", "dev-environment"},
797: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
062ai-devinstallation sequencesAn AI-development package profile overlapping the ordinary development sequence. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:799
799: r.Sequences["ai-dev"] = buildSequence(
800: "ai-dev",
801: "AI Development Environment",
802: "Complete AI/ML development with GPU support and Claude Code",
803: "~20GB",
804: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "nodejs", "claude-code", "ai-dev-environment"},
805: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
063docker-gpuinstallation sequencesA container/GPU package profile already covered by setup workflows. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:811
811: r.Sequences["docker-gpu"] = buildSequence(
812: "docker-gpu",
813: "Docker with GPU Support",
814: "Docker Engine with NVIDIA Container Toolkit for GPU containers",
815: "~5GB",
816: []string{"core", "nvidia", "docker", "nvidia-docker"},
817: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
064monitoringinstallation sequencesA monitoring package profile separate from the act of monitoring. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:819
819: r.Sequences["monitoring"] = buildSequence(
820: "monitoring",
821: "Monitoring Stack",
822: "System and GPU monitoring tools: htop, nvitop, gpustat, prometheus",
823: "~200MB",
824: []string{"core", "python", "monitoring-tools", "prometheus"},
825: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
065inference-cycleinstallation sequencesA reusable idea already covered by prepare/load/infer/verify methods. It adds no separate workflow meaning beyond prepare → infer → verify → report; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:831
831: r.Sequences["inference-cycle"] = buildSequence(
832: "inference-cycle",
833: "Complete Inference Cycle",
834: "Download Llama 8B, start server, test inference, benchmark",
835: "~32GB",
836: []string{"core", "python", "nvidia", "pytorch", "vllm", "huggingface-cli", "huggingface-auth", "hf-download-llama-8b", "vllm-serve-llama-8b", "test-inference", "benchmark-inference"},
837: )
066testinstallation sequencesA target-specific test sequence covered by the verification workflow. It adds no separate workflow meaning beyond run required checks → record evidence; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:839
839: r.Sequences["test"] = buildSequence(
840: "test",
841: "Test Inference",
842: "Send test request to running vLLM server on port 8000",
843: "~0",
844: []string{"test-inference"},
845: )
Comparison definitioninternal/workflows/registry.go:495
495: Name:          "run-all-tests",
497: Description:   "Execute all conduct tests to validate system functionality",
498: Category:      "Testing & Validation",
502: {Description: "List available tests", Command: "conduct test list", ExpectedOutcome: "All tests displayed by category", Troubleshooting: "Review test descriptions before running"},
503: {Description: "Run all tests", Command: "conduct test all", ExpectedOutcome: "All tests execute", Troubleshooting: "Use -v for verbose output"},
504: {Description: "View results", Command: "conduct test report", ExpectedOutcome: "Test results summary displayed", Troubleshooting: "Check failed tests for details"},
505: {Description: "Export results", Command: "conduct test report --json > results.json", ExpectedOutcome: "Results exported to JSON", Troubleshooting: "Use for CI/CD integration"},
510: Name:          "health-check",
512: Description:   "Run a quick health check to verify system status",
513: Category:      "Testing & Validation",
067benchmarkinstallation sequencesA benchmark sequence covered by baseline/measure/compare. It adds no separate workflow meaning beyond baseline → measure → compare; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:847
847: r.Sequences["benchmark"] = buildSequence(
848: "benchmark",
849: "Benchmark Inference",
850: "Run throughput benchmark on running vLLM server",
851: "~0",
852: []string{"python", "benchmark-inference"},
853: )
Comparison definitioninternal/workflows/registry.go:525
525: Name:          "performance-test",
527: Description:   "Execute performance tests to measure system throughput and latency",
528: Category:      "Testing & Validation",
532: {Description: "Run latency test", Command: "conduct test inference-latency run -v", ExpectedOutcome: "Latency metrics collected", Troubleshooting: "Run multiple times for accuracy"},
533: {Description: "Run throughput test", Command: "conduct test inference-throughput run -v", ExpectedOutcome: "Throughput metrics collected", Troubleshooting: "Ensure no other workloads running"},
534: {Description: "Run stress test", Command: "conduct test stress run --duration 5m", ExpectedOutcome: "System stability verified", Troubleshooting: "Monitor GPU temperature during test"},
535: {Description: "Generate performance report", Command: "conduct benchmark report", ExpectedOutcome: "Performance report generated", Troubleshooting: "Compare with previous baselines"},
540: Name:          "integration-test",
542: Description:   "Execute integration tests to verify component interactions",
543: Category:      "Testing & Validation",
068full-cycleinstallation sequencesA broad wrapper whose component steps should be selected from the canonical execution plan. It adds no separate workflow meaning beyond read task → execute explicit plan → verify → report; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:855
855: r.Sequences["full-cycle"] = buildSequence(
856: "full-cycle",
857: "Full Inference Status",
858: "Check GPU, server health, run test inference",
859: "~0",
860: []string{"full-inference-cycle"},
861: )
069stopinstallation sequencesStopping is already an atomic phrase and a stop/preserve/record workflow. It adds no separate workflow meaning beyond stop → preserve → record; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:867
867: r.Sequences["stop"] = buildSequence(
868: "stop",
869: "Stop All Servers",
870: "Gracefully stop all running vLLM inference servers",
871: "~0",
872: []string{"stop-all-servers"},
873: )
070cleanupinstallation sequencesCleanup is a scoped maintenance operation; the exact cleanup sequence belongs in the task. It adds no separate workflow meaning beyond inventory → plan bounded cleanup → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:875
875: r.Sequences["cleanup"] = buildSequence(
876: "cleanup",
877: "Cleanup Cache",
878: "Remove cached models to free disk space (interactive)",
879: "~0",
880: []string{"cleanup-cache"},
881: )
071huggingfaceinstallation sequencesOne provider/tooling installation profile. It adds no separate workflow meaning beyond resolve dependencies → install selected profile → verify; a dedicated slot repeats the existing operation with a different argument.
Candidate definitioninternal/sequence/types.go:887
887: r.Sequences["huggingface"] = buildSequence(
888: "huggingface",
889: "HuggingFace Setup",
890: "Install HuggingFace CLI and configure authentication",
891: "~100MB",
892: []string{"core", "python", "huggingface-cli", "huggingface-auth"},
893: )
Comparison definitioninternal/workflows/registry.go:58
58: Name:          "fresh-install",
60: Description:   "Complete fresh installation of conduct with all dependencies including vLLM, CUDA, and Docker",
61: Category:      "Installation & Setup",
65: {Description: "Update system packages", Command: "sudo apt update && sudo apt upgrade -y", ExpectedOutcome: "System packages updated to latest versions", Troubleshooting: "If apt fails, check your internet connection or try 'sudo apt --fix-broken install'"},
66: {Description: "Install Go (if not present)", Command: "wget https://go.dev/dl/go1.21.5.linux-amd64.tar.gz && sudo tar -C /usr/local -xzf go1.21.5.linux-amd64.tar.gz", ExpectedOutcome: "Go installed at /usr/local/go", Troubleshooting: "Add 'export PATH=$PATH:/usr/local/go/bin' to ~/.bashrc"},
67: {Description: "Install conduct CLI", Command: "go install github.com/joshkornreich/conduct@latest", ExpectedOutcome: "Conduct binary available in $GOPATH/bin", Troubleshooting: "Ensure GOPATH/bin is in your PATH"},
68: {Description: "Install Docker", Command: "curl -fsSL https://get.docker.com | sudo sh", ExpectedOutcome: "Docker daemon running", Troubleshooting: "Run 'sudo systemctl start docker' if daemon not running"},
69: {Description: "Install NVIDIA Container Toolkit", Command: "distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - && curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list && sudo apt update && sudo apt install -y nvidia-container-toolkit", ExpectedOutcome: "nvidia-container-toolkit installed", Troubleshooting: "Ensure NVIDIA drivers are installed first"},
70: {Description: "Configure Docker for GPU access", Command: "sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker", ExpectedOutcome: "Docker configured with NVIDIA runtime", Troubleshooting: "Check /etc/docker/daemon.json for nvidia runtime entry"},
71: {Description: "Verify installation", Command: "conduct --version && docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi", ExpectedOutcome: "Conduct version displayed and GPU accessible in Docker", Troubleshooting: "If GPU not visible, check NVIDIA driver installation"},
072setup-ml-stackjobsA named package bundle for the general environment setup workflow. Its steps are already covered by prepare environment → install profile → verify; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:73
73: Name:        "setup-ml-stack",
74: Category:    "Setup",
75: Description: "Install complete ML development stack (PyTorch, Transformers, vLLM, etc.)",
77: {Name: "Install PyTorch", Description: "Deep learning framework", Command: "install", Args: []string{"pytorch"}},
78: {Name: "Install Transformers", Description: "Hugging Face transformers", Command: "install", Args: []string{"transformers"}},
79: {Name: "Install Accelerate", Description: "Distributed training", Command: "install", Args: []string{"accelerate"}},
80: {Name: "Install vLLM", Description: "High-throughput serving", Command: "install", Args: []string{"vllm"}},
81: {Name: "Install Flash Attention", Description: "Optimized attention", Command: "install", Args: []string{"flash-attn"}},
82: {Name: "Install bitsandbytes", Description: "Quantization support", Command: "install", Args: []string{"bitsandbytes"}},
83: {Name: "Install safetensors", Description: "Safe serialization", Command: "install", Args: []string{"safetensors"}},
073setup-inferencejobsAn inference package bundle overlaps runtime setup. Its steps are already covered by prepare runtime → install profile → health-check; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:87
87: Name:        "setup-inference",
88: Category:    "Setup",
89: Description: "Install inference-focused stack (vLLM, TGI, Triton)",
91: {Name: "Install vLLM", Description: "High-throughput LLM serving", Command: "install", Args: []string{"vllm"}},
92: {Name: "Install TGI", Description: "Text Generation Inference", Command: "install", Args: []string{"text-generation-inference"}},
93: {Name: "Install Triton Server", Description: "NVIDIA inference server", Command: "install", Args: []string{"triton-inference-server"}},
94: {Name: "Install Ray Serve", Description: "Distributed serving", Command: "install", Args: []string{"ray-serve"}},
074setup-trainingjobsA training package bundle specializes environment setup. Its steps are already covered by prepare training environment → verify; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:98
98: Name:        "setup-training",
99: Category:    "Setup",
100: Description: "Install training-focused stack (DeepSpeed, PEFT, TRL)",
102: {Name: "Install PyTorch", Description: "Deep learning framework", Command: "install", Args: []string{"pytorch"}},
103: {Name: "Install DeepSpeed", Description: "Distributed training", Command: "install", Args: []string{"deepspeed"}},
104: {Name: "Install PEFT", Description: "Parameter-efficient fine-tuning", Command: "install", Args: []string{"peft"}},
105: {Name: "Install TRL", Description: "Transformer Reinforcement Learning", Command: "install", Args: []string{"trl"}},
106: {Name: "Install Accelerate", Description: "Hugging Face accelerate", Command: "install", Args: []string{"accelerate"}},
107: {Name: "Install Datasets", Description: "Hugging Face datasets", Command: "install", Args: []string{"datasets"}},
108: {Name: "Install W&B", Description: "Experiment tracking", Command: "install", Args: []string{"wandb"}},
075setup-ragjobsA RAG package bundle specializes environment setup. Its steps are already covered by prepare retrieval environment → verify; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:112
112: Name:        "setup-rag",
113: Category:    "Setup",
114: Description: "Install RAG stack (LangChain, LlamaIndex, vector DBs)",
116: {Name: "Install LangChain", Description: "LLM framework", Command: "install", Args: []string{"langchain"}},
117: {Name: "Install LlamaIndex", Description: "RAG framework", Command: "install", Args: []string{"llama-index"}},
118: {Name: "Install ChromaDB", Description: "Vector database", Command: "install", Args: []string{"chromadb"}},
119: {Name: "Install FAISS", Description: "Vector search", Command: "install", Args: []string{"faiss-gpu"}},
120: {Name: "Install Sentence Transformers", Description: "Embeddings", Command: "install", Args: []string{"sentence-transformers"}},
076load-llama-70bjobsThe model identity and size are inputs to load-model. Its steps are already covered by load-model with an exact model reference; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:128
128: Name:        "load-llama-70b",
129: Category:    "Model Loading",
130: Description: "Download and load Llama 3.3 70B on vLLM (TP4, FP16)",
132: {Name: "Install vLLM", Description: "Ensure vLLM is installed", Command: "install", Args: []string{"vllm"}},
133: {Name: "Download Model", Description: "Download Llama 3.3 70B from HuggingFace", Command: "install", Args: []string{"llama-3.3-70b"}},
134: {Name: "Load Model", Description: "Start vLLM server with TP4", Command: "load", Args: []string{"llama-3.3-70b", "--tp", "4", "--fp", "16"}},
Comparison definitioninternal/workflows/registry.go:164
164: Name:          "load-model",
166: Description:   "Load a downloaded model into vLLM for inference",
167: Category:      "Model Management",
171: {Description: "List downloaded models", Command: "conduct models list --local", ExpectedOutcome: "Local models displayed with sizes", Troubleshooting: "Download a model first if list is empty"},
172: {Description: "Check GPU memory availability", Command: "nvidia-smi --query-gpu=memory.free --format=csv,noheader", ExpectedOutcome: "Free GPU memory shown in MiB", Troubleshooting: "Unload other models if memory is low"},
173: {Description: "Load the model", Command: "conduct load <model-name>", ExpectedOutcome: "Model loaded, server ready for inference", Troubleshooting: "Use --gpu-memory-utilization 0.8 for memory constraints"},
174: {Description: "Verify model is loaded", Command: "conduct models status", ExpectedOutcome: "Model shown as active/loaded", Troubleshooting: "Check server logs for loading errors"},
175: {Description: "Test model inference", Command: "conduct infer \"Hello, how are you?\"", ExpectedOutcome: "Model generates response", Troubleshooting: "Ensure model is fully loaded before inference"},
180: Name:          "switch-model",
182: Description:   "Unload current model and load a different one",
077load-mistral-largejobsA particular model is an argument to load-model. Its steps are already covered by load-model with an exact model reference; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:138
138: Name:        "load-mistral-large",
139: Category:    "Model Loading",
140: Description: "Download and load Mistral Large 2 123B (TP4, FP16)",
142: {Name: "Install vLLM", Description: "Ensure vLLM is installed", Command: "install", Args: []string{"vllm"}},
143: {Name: "Download Model", Description: "Download Mistral Large 2", Command: "install", Args: []string{"mistral-large-2"}},
144: {Name: "Load Model", Description: "Start vLLM server with TP4", Command: "load", Args: []string{"mistral-large-2", "--tp", "4", "--fp", "16"}},
Comparison definitioninternal/workflows/registry.go:164
164: Name:          "load-model",
166: Description:   "Load a downloaded model into vLLM for inference",
167: Category:      "Model Management",
171: {Description: "List downloaded models", Command: "conduct models list --local", ExpectedOutcome: "Local models displayed with sizes", Troubleshooting: "Download a model first if list is empty"},
172: {Description: "Check GPU memory availability", Command: "nvidia-smi --query-gpu=memory.free --format=csv,noheader", ExpectedOutcome: "Free GPU memory shown in MiB", Troubleshooting: "Unload other models if memory is low"},
173: {Description: "Load the model", Command: "conduct load <model-name>", ExpectedOutcome: "Model loaded, server ready for inference", Troubleshooting: "Use --gpu-memory-utilization 0.8 for memory constraints"},
174: {Description: "Verify model is loaded", Command: "conduct models status", ExpectedOutcome: "Model shown as active/loaded", Troubleshooting: "Check server logs for loading errors"},
175: {Description: "Test model inference", Command: "conduct infer \"Hello, how are you?\"", ExpectedOutcome: "Model generates response", Troubleshooting: "Ensure model is fully loaded before inference"},
180: Name:          "switch-model",
182: Description:   "Unload current model and load a different one",
078load-code-stackjobsA coding model set is a serving profile. Its steps are already covered by select profile → load models → verify readiness; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:148
148: Name:        "load-code-stack",
149: Category:    "Model Loading",
150: Description: "Load coding-focused models (Codestral + StarCoder2)",
152: {Name: "Install vLLM", Description: "Ensure vLLM is installed", Command: "install", Args: []string{"vllm"}},
153: {Name: "Download Codestral", Description: "Mistral code model", Command: "install", Args: []string{"codestral"}},
154: {Name: "Download StarCoder2", Description: "BigCode model", Command: "install", Args: []string{"starcoder2-15b"}},
155: {Name: "Load Codestral", Description: "Start on port 8000", Command: "load", Args: []string{"codestral", "--tp", "2", "--port", "8000"}},
Comparison definitioninternal/workflows/registry.go:164
164: Name:          "load-model",
166: Description:   "Load a downloaded model into vLLM for inference",
167: Category:      "Model Management",
171: {Description: "List downloaded models", Command: "conduct models list --local", ExpectedOutcome: "Local models displayed with sizes", Troubleshooting: "Download a model first if list is empty"},
172: {Description: "Check GPU memory availability", Command: "nvidia-smi --query-gpu=memory.free --format=csv,noheader", ExpectedOutcome: "Free GPU memory shown in MiB", Troubleshooting: "Unload other models if memory is low"},
173: {Description: "Load the model", Command: "conduct load <model-name>", ExpectedOutcome: "Model loaded, server ready for inference", Troubleshooting: "Use --gpu-memory-utilization 0.8 for memory constraints"},
174: {Description: "Verify model is loaded", Command: "conduct models status", ExpectedOutcome: "Model shown as active/loaded", Troubleshooting: "Check server logs for loading errors"},
175: {Description: "Test model inference", Command: "conduct infer \"Hello, how are you?\"", ExpectedOutcome: "Model generates response", Troubleshooting: "Ensure model is fully loaded before inference"},
180: Name:          "switch-model",
182: Description:   "Unload current model and load a different one",
079load-multimodaljobsModality is a serving profile and acceptance requirement. Its steps are already covered by load models → verify required modalities; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:159
159: Name:        "load-multimodal",
160: Category:    "Model Loading",
161: Description: "Load multimodal vision-language models",
163: {Name: "Install vLLM", Description: "Ensure vLLM is installed", Command: "install", Args: []string{"vllm"}},
164: {Name: "Download Llama Vision", Description: "Llama 3.2 11B Vision", Command: "install", Args: []string{"llama-3.2-11b-vision"}},
165: {Name: "Download Pixtral", Description: "Mistral vision model", Command: "install", Args: []string{"pixtral-12b"}},
166: {Name: "Load Llama Vision", Description: "Start vision model", Command: "load", Args: []string{"llama-3.2-11b-vision", "--tp", "1", "--port", "8000"}},
Comparison definitioninternal/workflows/registry.go:164
164: Name:          "load-model",
166: Description:   "Load a downloaded model into vLLM for inference",
167: Category:      "Model Management",
171: {Description: "List downloaded models", Command: "conduct models list --local", ExpectedOutcome: "Local models displayed with sizes", Troubleshooting: "Download a model first if list is empty"},
172: {Description: "Check GPU memory availability", Command: "nvidia-smi --query-gpu=memory.free --format=csv,noheader", ExpectedOutcome: "Free GPU memory shown in MiB", Troubleshooting: "Unload other models if memory is low"},
173: {Description: "Load the model", Command: "conduct load <model-name>", ExpectedOutcome: "Model loaded, server ready for inference", Troubleshooting: "Use --gpu-memory-utilization 0.8 for memory constraints"},
174: {Description: "Verify model is loaded", Command: "conduct models status", ExpectedOutcome: "Model shown as active/loaded", Troubleshooting: "Check server logs for loading errors"},
175: {Description: "Test model inference", Command: "conduct infer \"Hello, how are you?\"", ExpectedOutcome: "Model generates response", Troubleshooting: "Ensure model is fully loaded before inference"},
180: Name:          "switch-model",
182: Description:   "Unload current model and load a different one",
080download-all-flagshipjobsA mutable flagship collection is data, not a stable action meaning. Its steps are already covered by resolve model set → download → verify manifest; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:174
174: Name:        "download-all-flagship",
175: Category:    "Download",
176: Description: "Download all flagship models from the catalog (70B+ class)",
178: {Name: "Download Llama 3.3 70B", Description: "Meta flagship", Command: "install", Args: []string{"llama-3.3-70b"}},
179: {Name: "Download Mistral Large 2", Description: "Mistral flagship", Command: "install", Args: []string{"mistral-large-2"}},
180: {Name: "Download Command R+", Description: "Cohere flagship", Command: "install", Args: []string{"command-r-plus-08-2024"}},
181: {Name: "Download Jamba 1.5 Large", Description: "AI21 hybrid model", Command: "install", Args: []string{"jamba-1.5-large"}},
182: {Name: "Download Nemotron 70B", Description: "NVIDIA model", Command: "install", Args: []string{"nemotron-70b"}},
081download-all-efficientjobsAn efficiency-selected collection is a model-set parameter. Its steps are already covered by resolve model set → download → verify manifest; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:186
186: Name:        "download-all-efficient",
187: Category:    "Download",
188: Description: "Download all efficient models (7-14B class)",
190: {Name: "Download Phi-4", Description: "Microsoft 14B", Command: "install", Args: []string{"phi-4"}},
191: {Name: "Download Mistral Nemo", Description: "Mistral 12B", Command: "install", Args: []string{"mistral-nemo"}},
192: {Name: "Download Gemma 2 9B", Description: "Google 9B", Command: "install", Args: []string{"gemma-2-9b"}},
193: {Name: "Download Yi 1.5 9B", Description: "01.AI 9B", Command: "install", Args: []string{"yi-1.5-9b"}},
194: {Name: "Download InternLM 2.5 7B", Description: "Alibaba 7B", Command: "install", Args: []string{"internlm2.5-7b"}},
195: {Name: "Download Jamba Mini", Description: "AI21 12B hybrid", Command: "install", Args: []string{"jamba-1.5-mini"}},
082download-all-codejobsA coding collection is a model-set parameter. Its steps are already covered by resolve model set → download → verify manifest; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:199
199: Name:        "download-all-code",
200: Category:    "Download",
201: Description: "Download all code-focused models",
203: {Name: "Download Codestral", Description: "Mistral 22B code", Command: "install", Args: []string{"codestral"}},
204: {Name: "Download StarCoder2 15B", Description: "BigCode 15B", Command: "install", Args: []string{"starcoder2-15b"}},
205: {Name: "Download StarCoder2 7B", Description: "BigCode 7B", Command: "install", Args: []string{"starcoder2-7b"}},
206: {Name: "Download CodeGemma 7B", Description: "Google code model", Command: "install", Args: []string{"codegemma-7b"}},
207: {Name: "Download Yi Coder 9B", Description: "01.AI code model", Command: "install", Args: []string{"yi-coder-9b"}},
208: {Name: "Download Stable Code 3B", Description: "Stability AI code", Command: "install", Args: []string{"stable-code-instruct-3b"}},
083download-all-embeddingsjobsEmbedding model downloads reuse the same download/verify workflow. Its steps are already covered by resolve model set → download → verify manifest; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:212
212: Name:        "download-all-embeddings",
213: Category:    "Download",
214: Description: "Download all embedding and retrieval models",
216: {Name: "Download Nomic Embed", Description: "Nomic embeddings", Command: "install", Args: []string{"nomic-embed-text-v1.5"}},
217: {Name: "Download BGE Large", Description: "BAAI embeddings", Command: "install", Args: []string{"bge-large-en-v1.5"}},
218: {Name: "Download E5 Mistral", Description: "E5 7B embeddings", Command: "install", Args: []string{"e5-mistral-7b-instruct"}},
219: {Name: "Download Arctic Embed", Description: "Snowflake embeddings", Command: "install", Args: []string{"arctic-embed-l"}},
220: {Name: "Download BGE Reranker", Description: "Reranking model", Command: "install", Args: []string{"bge-reranker-v2-m3"}},
084optimize-inferencejobsA domain-specific optimization job overlaps the broader measured optimization cycle. Its steps are already covered by baseline → optimize → measure → compare; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:228
228: Name:        "optimize-inference",
229: Category:    "Optimization",
230: Description: "Apply common inference optimizations (Flash Attention, xFormers)",
232: {Name: "Install Flash Attention", Description: "Optimized attention kernels", Command: "install", Args: []string{"flash-attn"}},
233: {Name: "Install xFormers", Description: "Memory-efficient attention", Command: "install", Args: []string{"xformers"}},
234: {Name: "Install Triton", Description: "JIT compiler", Command: "bash", Args: []string{"pip", "install", "triton", "--upgrade"}},
235: {Name: "Patch Triton Issues", Description: "Fix common Triton issues", Command: "patch", Args: []string{"triton"}},
Comparison definitioninternal/workflows/registry.go:525
525: Name:          "performance-test",
527: Description:   "Execute performance tests to measure system throughput and latency",
528: Category:      "Testing & Validation",
532: {Description: "Run latency test", Command: "conduct test inference-latency run -v", ExpectedOutcome: "Latency metrics collected", Troubleshooting: "Run multiple times for accuracy"},
533: {Description: "Run throughput test", Command: "conduct test inference-throughput run -v", ExpectedOutcome: "Throughput metrics collected", Troubleshooting: "Ensure no other workloads running"},
534: {Description: "Run stress test", Command: "conduct test stress run --duration 5m", ExpectedOutcome: "System stability verified", Troubleshooting: "Monitor GPU temperature during test"},
535: {Description: "Generate performance report", Command: "conduct benchmark report", ExpectedOutcome: "Performance report generated", Troubleshooting: "Compare with previous baselines"},
540: Name:          "integration-test",
542: Description:   "Execute integration tests to verify component interactions",
543: Category:      "Testing & Validation",
085optimize-memoryjobsMemory is the optimization objective, not a separate control-flow pattern. Its steps are already covered by baseline → optimize selected metric → verify; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:239
239: Name:        "optimize-memory",
240: Category:    "Optimization",
241: Description: "Optimize GPU memory usage and quantization",
243: {Name: "Install bitsandbytes", Description: "8-bit quantization", Command: "install", Args: []string{"bitsandbytes"}},
244: {Name: "Install AutoGPTQ", Description: "GPTQ quantization", Command: "bash", Args: []string{"pip", "install", "auto-gptq", "--upgrade"}},
245: {Name: "Install AutoAWQ", Description: "AWQ quantization", Command: "bash", Args: []string{"pip", "install", "autoawq", "--upgrade"}},
246: {Name: "Clear Cache", Description: "Clear model cache locks", Command: "patch", Args: []string{"cache-corrupt"}},
247: {Name: "Fix VRAM Issues", Description: "Clear VRAM if needed", Command: "patch", Args: []string{"vram-oom"}},
Comparison definitioninternal/workflows/registry.go:196
196: Name:          "model-benchmark",
198: Description:   "Run performance benchmarks on a loaded model to measure throughput and latency",
199: Category:      "Model Management",
203: {Description: "Verify model is loaded", Command: "conduct models status", ExpectedOutcome: "Model shown as active", Troubleshooting: "Load a model first"},
204: {Description: "Run latency benchmark", Command: "conduct benchmark latency --prompts 100", ExpectedOutcome: "Average latency and percentiles displayed", Troubleshooting: "Reduce prompt count for quick tests"},
205: {Description: "Run throughput benchmark", Command: "conduct benchmark throughput --duration 60", ExpectedOutcome: "Tokens per second displayed", Troubleshooting: "Longer duration gives more accurate results"},
206: {Description: "Run memory benchmark", Command: "conduct benchmark memory", ExpectedOutcome: "Memory usage patterns displayed", Troubleshooting: "Monitor nvidia-smi during test"},
207: {Description: "Generate benchmark report", Command: "conduct benchmark report --output benchmark.json", ExpectedOutcome: "Detailed benchmark report saved", Troubleshooting: "Use --format html for visual report"},
218: Name:          "quick-inference",
220: Description:   "Run a quick inference test to verify the model is working correctly",
086fix-common-issuesjobsThe failure set changes over time; keep repairs as recipes behind diagnose/correct/verify. Its steps are already covered by inspect → diagnose → correct → verify; the extra name consumes a byte without a distinct control-flow requirement.
Candidate definitioncmd/job.go:255
255: Name:        "fix-common-issues",
256: Category:    "Maintenance",
257: Description: "Fix common installation and runtime issues",
259: {Name: "Fix CUDA Mismatch", Description: "Resolve CUDA version conflicts", Command: "patch", Args: []string{"cuda-mismatch"}},
260: {Name: "Fix Missing Dependencies", Description: "Install missing pip deps", Command: "patch", Args: []string{"missing-deps"}},
261: {Name: "Fix Flash Attention", Description: "Rebuild flash attention", Command: "patch", Args: []string{"flash-attn"}},
262: {Name: "Fix Ninja Build", Description: "Install ninja build system", Command: "patch", Args: []string{"ninja-build"}},
263: {Name: "Clear Cache Locks", Description: "Remove stale cache locks", Command: "patch", Args: []string{"cache-corrupt"}},
Comparison definitioninternal/workflows/registry.go:578
578: Name:          "diagnose-vllm",
580: Description:   "Troubleshoot vLLM server problems",
581: Category:      "Troubleshooting",
585: {Description: "Check vLLM process", Command: "pgrep -f vllm", ExpectedOutcome: "vLLM process ID shown", Troubleshooting: "Start vLLM if no process found"},
586: {Description: "Check server logs", Command: "conduct logs --service vllm --tail 50", ExpectedOutcome: "Recent vLLM logs displayed", Troubleshooting: "Look for error messages"},
587: {Description: "Test health endpoint", Command: "curl -s http://localhost:8000/health", ExpectedOutcome: "Health check returns OK", Troubleshooting: "Server may still be initializing"},
588: {Description: "Check loaded models", Command: "curl -s http://localhost:8000/v1/models | jq .", ExpectedOutcome: "Loaded models listed", Troubleshooting: "Load a model if list is empty"},
589: {Description: "Test simple inference", Command: "curl -X POST http://localhost:8000/v1/completions -H 'Content-Type: application/json' -d '{\"model\":\"default\",\"prompt\":\"test\",\"max_tokens\":5}'", ExpectedOutcome: "Inference response returned", Troubleshooting: "Check GPU memory if inference fails"},
590: {Description: "Restart vLLM (if needed)", Command: "conduct server restart", ExpectedOutcome: "vLLM server restarted", Troubleshooting: "Wait for model to reload after restart"},
595: Name:          "diagnose-network",
087health-checkjobsThe guided workflow registry already has a health-check entry. This is a direct duplicate, not another protocol.
Candidate definitioncmd/job.go:267
267: Name:        "health-check",
268: Category:    "Maintenance",
269: Description: "Run system health checks and display status",
271: {Name: "Check GPU Status", Description: "Verify GPU availability", Command: "bash", Args: []string{"nvidia-smi", "--query-gpu=name,memory.total,memory.free", "--format=csv"}},
272: {Name: "Check PyTorch", Description: "Verify PyTorch CUDA", Command: "bash", Args: []string{"python3", "-c", "import torch; print(f'PyTorch {torch.__version__}, CUDA: {torch.cuda.is_available()}')"}},
273: {Name: "Check vLLM", Description: "Verify vLLM installation", Command: "bash", Args: []string{"python3", "-c", "import vllm; print(f'vLLM {vllm.__version__}')"}},
274: {Name: "Check HuggingFace", Description: "Verify HF authentication", Command: "patch", Args: []string{"hf-auth"}},
275: {Name: "Show Atomically Status", Description: "Show installed items", Command: "status", Args: []string{}},
331: Args:  cobra.ExactArgs(1),
423: fmt.Printf("%s %s\n", jobInfoStyle.Render("Category:"), job.Category)
Comparison definitioninternal/workflows/registry.go:510
510: Name:          "health-check",
512: Description:   "Run a quick health check to verify system status",
513: Category:      "Testing & Validation",
517: {Description: "Run doctor check", Command: "conduct doctor", ExpectedOutcome: "System health summary displayed", Troubleshooting: "Shows issues that need attention"},
518: {Description: "Check service health", Command: "conduct health", ExpectedOutcome: "All services healthy", Troubleshooting: "Restart unhealthy services"},
519: {Description: "Verify GPU access", Command: "conduct test gpu-status run", ExpectedOutcome: "GPU test passes", Troubleshooting: "Check NVIDIA drivers if fails"},
520: {Description: "Test inference", Command: "conduct test inference-basic run", ExpectedOutcome: "Basic inference works", Troubleshooting: "Ensure model is loaded"},
525: Name:          "performance-test",
527: Description:   "Execute performance tests to measure system throughput and latency",
528: Category:      "Testing & Validation",
088horror-genreanalysis protocolsHorror changes the evaluation criteria, not the orchestration. Giving every genre a byte multiplies codes for the same analysis sequence.
Candidate definitioncmd/coverage_protocol_defs.go:1407
1407: ID:       "horror-genre",
1408: Name:     "Horror Genre Analysis",
1409: Category: "Genre Protocols",
1410: Description: "Horror-specific analysis based on fear psychology, subgenre conventions, and scare mechanics.",
1469: {Order: 1, Name: "Subgenre ID", Description: "Classify subgenre and hybrid elements", Models: []string{"claude-opus"}, Parallel: false},
1470: {Order: 2, Name: "Horror Diagnostics", Description: "6 parallel tracks: fear arc, scares, rules, threat, atmosphere, survival logic", Models: []string{"claude-opus", "claude-sonnet", "jamba-1.5-mini"}, Parallel: true},
1471: {Order: 3, Name: "Convention Check", Description: "Compare against subgenre beat sheet requirements", Models: []string{"claude-opus"}, Parallel: false},
1472: {Order: 4, Name: "Fear Scoring", Description: "Calculate dread quotient, scare density, escalation", Models: []string{"claude-sonnet"}, Parallel: false},
1473: {Order: 5, Name: "Commercial Analysis", Description: "Budget tier, rating, festival potential", Models: []string{"claude-sonnet"}, Parallel: false},
1474: {Order: 6, Name: "Recommendations", Description: "Horror-specific improvement suggestions", Models: []string{"claude-opus"}, Parallel: false},
Comparison definitioncmd/coverage_protocol_defs.go:78
78: ID:       "deep-dive",
79: Name:     "Deep Dive",
80: Category: "Multi-Pass Analysis",
81: Description: "Four-pass progressive analysis that goes from surface-level overview to granular scene-by-scene examination.",
124: {Order: 1, Name: "Aerial Pass", Description: "High-level overview - logline, genre, tone, major structure", Models: []string{"claude-sonnet"}, Parallel: false},
125: {Order: 2, Name: "Act Analysis", Description: "Act-by-act breakdown with turning points and progression", Models: []string{"claude-opus"}, Parallel: false},
126: {Order: 3, Name: "Sequence Analysis", Description: "Eight-sequence breakdown with beat mapping", Models: []string{"claude-opus"}, Parallel: true},
127: {Order: 4, Name: "Scene Microscope", Description: "Detailed scene-by-scene analysis of flagged areas", Models: []string{"claude-sonnet"}, Parallel: true},
128: {Order: 5, Name: "Integration", Description: "Compile multi-resolution findings into coherent report", Models: []string{"claude-opus"}, Parallel: false},
131: Outputs: []string{"deep_dive_report.md", "aerial_view.md", "act_analysis.md", "sequence_breakdown.json", "scene_notes.json"},
089comedy-genreanalysis protocolsComedy changes the evaluation criteria, not the orchestration. Its separate code duplicates the same analysis sequence with another rubric.
Candidate definitioncmd/coverage_protocol_defs.go:1488
1488: ID:       "comedy-genre",
1489: Name:     "Comedy Genre Analysis",
1490: Category: "Genre Protocols",
1491: Description: "Theory-based comedy analysis with joke mechanics, comedic structure, and audience targeting.",
1552: {Order: 1, Name: "Comedy Classification", Description: "Identify subgenre and tone spectrum", Models: []string{"claude-opus"}, Parallel: false},
1553: {Order: 2, Name: "Comedy Diagnostics", Description: "6 parallel tracks: jokes, character, setpieces, structure, rhythm, visual", Models: []string{"claude-opus", "claude-sonnet", "jamba-1.5-mini"}, Parallel: true},
1554: {Order: 3, Name: "Theory Application", Description: "Analyze jokes through incongruity/superiority/relief lenses", Models: []string{"claude-opus"}, Parallel: false},
1555: {Order: 4, Name: "Audience Targeting", Description: "Demographic appeal and cultural context analysis", Models: []string{"claude-sonnet"}, Parallel: false},
1556: {Order: 5, Name: "Effectiveness Scoring", Description: "Laugh density, variety, callback efficiency, originality", Models: []string{"claude-sonnet"}, Parallel: false},
1557: {Order: 6, Name: "Punch-Up Suggestions", Description: "Comedy-specific improvement and punch-up opportunities", Models: []string{"claude-opus"}, Parallel: false},
Comparison definitioncmd/coverage_protocol_defs.go:78
78: ID:       "deep-dive",
79: Name:     "Deep Dive",
80: Category: "Multi-Pass Analysis",
81: Description: "Four-pass progressive analysis that goes from surface-level overview to granular scene-by-scene examination.",
124: {Order: 1, Name: "Aerial Pass", Description: "High-level overview - logline, genre, tone, major structure", Models: []string{"claude-sonnet"}, Parallel: false},
125: {Order: 2, Name: "Act Analysis", Description: "Act-by-act breakdown with turning points and progression", Models: []string{"claude-opus"}, Parallel: false},
126: {Order: 3, Name: "Sequence Analysis", Description: "Eight-sequence breakdown with beat mapping", Models: []string{"claude-opus"}, Parallel: true},
127: {Order: 4, Name: "Scene Microscope", Description: "Detailed scene-by-scene analysis of flagged areas", Models: []string{"claude-sonnet"}, Parallel: true},
128: {Order: 5, Name: "Integration", Description: "Compile multi-resolution findings into coherent report", Models: []string{"claude-opus"}, Parallel: false},
131: Outputs: []string{"deep_dive_report.md", "aerial_view.md", "act_analysis.md", "sequence_breakdown.json", "scene_notes.json"},
090tv-pilotanalysis protocolsPilot format is a source constraint and rubric choice. It does not establish a distinct general-purpose workflow worth its own code.
Candidate definitioncmd/coverage_protocol_defs.go:1575
1575: ID:       "tv-pilot",
1576: Name:     "TV Pilot Analysis",
1577: Category: "Format Protocols",
1578: Description: "TV-specific analysis evaluating both pilot quality and series potential with series engine assessment.",
1643: {Order: 1, Name: "Format ID", Description: "Classify format, genre, and initial platform fit", Models: []string{"claude-opus"}, Parallel: false},
1644: {Order: 2, Name: "Pilot Diagnostics", Description: "6 parallel tracks: structure, ensemble, world, hooks, tone, voice", Models: []string{"claude-opus", "claude-sonnet", "jamba-1.5-mini"}, Parallel: true},
1645: {Order: 3, Name: "Series Engine", Description: "Analyze what generates episodes and series sustainability", Models: []string{"claude-opus"}, Parallel: false},
1646: {Order: 4, Name: "Castability", Description: "Assess role attractiveness and casting opportunities", Models: []string{"claude-sonnet"}, Parallel: false},
1647: {Order: 5, Name: "Platform Matching", Description: "Detailed fit analysis for network, cable, streaming", Models: []string{"claude-sonnet"}, Parallel: false},
1648: {Order: 6, Name: "Season Projection", Description: "Project S1 arc and multi-season potential", Models: []string{"claude-opus"}, Parallel: false},
Comparison definitioncmd/coverage_protocol_defs.go:891
891: ID:       "studio-grade",
892: Name:     "Studio Grade",
893: Category: "Comprehensive",
894: Description: "Full professional coverage matching major studio standards with all documentation and database integration.",
944: {Order: 1, Name: "Extraction", Description: "Parse screenplay and create database records", Models: []string{"local"}, Parallel: false},
945: {Order: 2, Name: "Multi-Model Primary", Description: "Parallel analysis with multiple models", Models: []string{"claude-opus", "llama-3.3-70b"}, Parallel: true},
946: {Order: 3, Name: "Full Rubric", Description: "Complete 16-point rubric analysis", Models: []string{"claude-opus"}, Parallel: false},
947: {Order: 4, Name: "Market Analysis", Description: "Commercial viability and positioning", Models: []string{"claude-opus"}, Parallel: true},
948: {Order: 5, Name: "Synthesis", Description: "Aggregate and reconcile all analyses", Models: []string{"claude-opus"}, Parallel: false},
949: {Order: 6, Name: "Audit", Description: "Verify claims and check for bias", Models: []string{"claude-opus"}, Parallel: false},
091role-showcaseanalysis protocolsA casting/performance focus is a requested analytical perspective. A separate workflow code duplicates the existing perspective-based analysis pattern.
Candidate definitioncmd/coverage_protocol_defs.go:1667
1667: ID:       "role-showcase",
1668: Name:     "Role Showcase",
1669: Category: "Industry Workflow",
1670: Description: "Actor/casting perspective analysis evaluating role attractiveness, castability, and actor matching.",
1728: {Order: 1, Name: "Role Inventory", Description: "Extract all roles with page counts, scenes, dialogue", Models: []string{"local", "claude-sonnet"}, Parallel: false},
1729: {Order: 2, Name: "Castability Analysis", Description: "6 parallel tracks: range, depth, awards, presence, moments, wishlist", Models: []string{"claude-opus", "claude-sonnet"}, Parallel: true},
1730: {Order: 3, Name: "Role Profiles", Description: "Generate detailed casting breakdown for each key role", Models: []string{"claude-opus"}, Parallel: false},
1731: {Order: 4, Name: "Actor Matching", Description: "Suggest casting tiers and actor types", Models: []string{"claude-sonnet"}, Parallel: false},
1732: {Order: 5, Name: "Enhancement", Description: "Suggest improvements to increase role attractiveness", Models: []string{"claude-opus"}, Parallel: false},
1735: Outputs: []string{"role_showcase.md", "casting_breakdown.md", "actor_matching.md", "role_enhance.md", "casting_summary.md"},
Comparison definitioncmd/coverage_protocol_defs.go:11
11: ID:       "trinity-pass",
12: Name:     "Trinity Pass",
13: Category: "Multi-Pass Analysis",
14: Description: "Three-perspective analysis simulating creative executive, producer, and director viewpoints for comprehensive coverage.",
60: {Order: 1, Name: "Extraction", Description: "Extract and parse screenplay into structured format", Models: []string{"local"}, Parallel: false},
61: {Order: 2, Name: "Creative Executive Pass", Description: "Analyze from studio executive perspective - commercial viability, market positioning, audience appeal", Models: []string{"claude-opus"}, Parallel: true},
62: {Order: 3, Name: "Producer Pass", Description: "Analyze from producer perspective - budget implications, casting needs, production feasibility", Models: []string{"claude-opus"}, Parallel: true},
63: {Order: 4, Name: "Director Pass", Description: "Analyze from director perspective - visual storytelling, performance opportunities, cinematic potential", Models: []string{"claude-opus"}, Parallel: true},
64: {Order: 5, Name: "Synthesis", Description: "Reconcile three perspectives into unified assessment with prioritized recommendations", Models: []string{"claude-opus"}, Parallel: false},
67: Outputs: []string{"trinity_coverage.md", "executive_notes.md", "producer_notes.md", "director_notes.md", "synthesis.json"},
092coverage orchestrate thoroughdeployment presetsAdds a separate code for the thorough option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1286
1286: "coverage orchestrate thorough": {
1287: Name:        "coverage orchestrate thorough",
1288: Description: "Phi-4 + Llama-70B TP4 + Jamba-1.5-Large TP3",
1289: GPUType:     "B200",
1290: GPUCount:    8,
1291: GPUMemoryGB: 180,
1294: Name:         "Phi-4",
1295: ModelID:      "microsoft/phi-4",
1296: TP:           1,
1303: Name:         "Llama-70B",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
093coverage orchestrate rapiddeployment presetsAdds a separate code for the rapid option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1356
1356: "coverage orchestrate rapid": {
1357: Name:        "coverage orchestrate rapid",
1358: Description: "6x Mistral-Small + Llama-70B TP2",
1359: GPUType:     "B200",
1360: GPUCount:    8,
1361: GPUMemoryGB: 180,
1364: Name:         "Mistral-Small-0",
1365: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
1366: TP:           1,
1373: Name:         "Mistral-Small-1",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
094coverage orchestrate contextualdeployment presetsAdds a separate code for the contextual option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1465
1465: "coverage orchestrate contextual": {
1466: Name:        "coverage orchestrate contextual",
1467: Description: "Jamba-1.7-Large TP8 (398B) full context",
1468: GPUType:     "B200",
1469: GPUCount:    8,
1470: GPUMemoryGB: 180,
1473: Name:         "Jamba-1.7-Large",
1474: ModelID:      "ai21labs/AI21-Jamba-Large-1.7-FP8",
1475: TP:           8,
1486: Name:     "Model Download",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
095coverage orchestrate heterogeneousdeployment presetsAdds a separate code for the heterogeneous option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1515
1515: "coverage orchestrate heterogeneous": {
1516: Name:        "coverage orchestrate heterogeneous",
1517: Description: "Multi-vendor ensemble: Mistral+Phi scenes, Llama+Command-R acts, Jamba screenplay",
1518: GPUType:     "B200",
1519: GPUCount:    8,
1520: GPUMemoryGB: 180,
1524: Name:         "Scene-Mistral",
1525: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
1526: TP:           1,
1533: Name:         "Scene-Phi4",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
096coverage orchestrate homogeneousdeployment presetsAdds a separate code for the homogeneous option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1610
1610: "coverage orchestrate homogeneous": {
1611: Name:        "coverage orchestrate homogeneous",
1612: Description: "All-Llama consistency: Llama-3.2-3B scenes, dual Llama-70B acts, Llama-70B screenplay",
1613: GPUType:     "B200",
1614: GPUCount:    8,
1615: GPUMemoryGB: 180,
1619: Name:         "Scene-Llama-0",
1620: ModelID:      "meta-llama/Llama-3.2-3B-Instruct",
1621: TP:           1,
1628: Name:         "Scene-Llama-1",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
097coverage orchestrate intelligentdeployment presetsAdds a separate code for the intelligent option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1702
1702: "coverage orchestrate intelligent": {
1703: Name:        "coverage orchestrate intelligent",
1704: Description: "2x Llama-70B TP4 for maximum reasoning",
1705: GPUType:     "B200",
1706: GPUCount:    8,
1707: GPUMemoryGB: 180,
1710: Name:         "Llama-70B-0",
1711: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
1712: TP:           4,
1719: Name:         "Llama-70B-1",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
098coverage orchestrate exhaustivedeployment presetsAdds a separate code for the exhaustive option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1763
1763: "coverage orchestrate exhaustive": {
1764: Name:        "coverage orchestrate exhaustive",
1765: Description: "Mistral-Small + Phi-4 + Llama-70B + Jamba-1.5 + Command-R-Plus",
1766: GPUType:     "B200",
1767: GPUCount:    8,
1768: GPUMemoryGB: 180,
1771: Name:         "Mistral-Small",
1772: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
1773: TP:           1,
1780: Name:         "Phi-4",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
099coverage orchestrate ensembledeployment presetsAdds a separate code for the ensemble option on the same orchestration family. Encoding each depth, model mix, or routing option as a whole protocol grows overlapping variants instead of distinct workflows.
Candidate definitioncmd/lambda_workflows_data.go:1855
1855: "coverage orchestrate ensemble": {
1856: Name:        "coverage orchestrate ensemble",
1857: Description: "Llama-70B + Jamba-1.5 + Command-R-Plus + Mistral-Small ensemble",
1858: GPUType:     "B200",
1859: GPUCount:    8,
1860: GPUMemoryGB: 180,
1863: Name:         "Llama-70B",
1864: ModelID:      "meta-llama/Llama-3.3-70B-Instruct",
1865: TP:           2,
1872: Name:         "Jamba-1.5-Large",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")
100coverage orchestratedeployment presetsBundles the hierarchical analysis method with a fixed deployment blueprint. As a byte-table entry it conflates the reusable instruction with its model/GPU configuration, making the code unnecessarily specific.
Candidate definitioncmd/lambda_workflows_data.go:1196
1196: "coverage orchestrate": {
1197: Name:        "coverage orchestrate",
1198: Description: "2x Mistral-Small + 2x Llama-70B TP2 + Jamba-1.5-Large TP2",
1199: GPUType:     "B200",
1200: GPUCount:    8,
1201: GPUMemoryGB: 180,
1204: Name:         "Mistral-Small-0",
1205: ModelID:      "mistralai/Mistral-Small-24B-Instruct-2501",
1206: TP:           1,
1213: Name:         "Mistral-Small-1",
Comparison definitioncmd/architect_orchestrate.go:50
50: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateThorough, "thorough", false, "Thorough analysis variant")
51: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateRapid, "rapid", false, "Rapid analysis variant")
52: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateContextual, "contextual", false, "Context-aware variant")
53: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHeterogeneous, "heterogeneous", false, "Mixed model types")
54: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateHomogeneous, "homogeneous", false, "Uniform model config")
55: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateIntelligent, "intelligent", false, "AI-driven selection")
56: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateExhaustive, "exhaustive", false, "Complete analysis")
57: architectOrchestrateCmd.PersistentFlags().BoolVar(&orchestrateEnsemble, "ensemble", false, "Multi-model consensus")

Ranking is Merritt’s judgment, not a measured probability. Byte-table membership and code allocation remain unchanged.