Build a Datacenter Rack Airflow Deficit and Workload Evacuation Planner — implementation

## Build
Analyze multi-chassis intake and exhaust thermal sensor telemetry across server racks. Formulate proactive workload evacuation and fan throttling mitigation plans with hosted Claude.

## User
Facility Systems and Infrastructure Engineer

## Workflow
1. Ingest rack-level inlet temperature and server exhaust temperature metrics from DCGM telemetry.
2. Identify thermal recirculations and localized hot-aisle air dams exceeding design limits.
3. Use hosted Claude API to formulate an emergency workload redistribution and job migration runbook.
4. Produce an evacuation schedule JSON prioritizing jobs by checkpoint maturity and node heat.

## Interface
Thermal heat-map web view displaying rack chassis elevations with evacuation triggers.

## Data
Synthetic rack environmental sensor records and node temperatures in standard Prometheus format.
Starter examples below are synthetic. They are not downloaded records.

## Source
[NVIDIA DCGM exporter](https://raw.githubusercontent.com/NVIDIA/dcgm-exporter/main/README.md)
Leverages DCGM exporter temperature gauges and chassis thermal sensors to assess rack heat.

## Tools
pydantic-ai, anthropic-claude.
Enforces strict schemas for environmental telemetry feeds and evacuation runbooks.
Hosted Claude API for technical thermal runbook synthesis; no local execution.

## Deliverables
Submit runnable source, pinned dependencies, setup commands, a working interface, example outputs, and a recorded end-to-end demo. Include tests and an action log.

## Acceptance
- Calculates delta-T between inlet and exhaust across rack U-positions accurately.
- Triggers workload evacuation when inlet temperature exceeds ASHRAE envelope limits.
- Prioritizes evacuation of high-wattage training jobs before low-wattage inference workloads.
- Generates safe execution runbook without modifying physical BMS cooling setpoints.

## Starter Examples
```json
[
  {
    "input": {
      "rack_id": "RACK-E04",
      "inlet_ambient_c": 28.5,
      "servers": [
        {
          "u_slot": "U18-U22",
          "node_id": "node-18",
          "inlet_c": 34,
          "exhaust_c": 68,
          "active_job": "open-deepseek-train",
          "power_kw": 9.8
        },
        {
          "u_slot": "U24-U28",
          "node_id": "node-24",
          "inlet_c": 36.5,
          "exhaust_c": 72,
          "active_job": "open-llama-inf",
          "power_kw": 4.2
        }
      ]
    },
    "expected_output": {
      "containment_breach": true,
      "hot_spot_u_slot": "U24-U28",
      "evacuation_priority_order": [
        {
          "node_id": "node-18",
          "action": "CHECKPOINT_AND_DRAIN",
          "reason": "High 9.8kW power contribution to local hot dam"
        },
        {
          "node_id": "node-24",
          "action": "MIGRATE_REPLICAS",
          "reason": "Critically elevated inlet temperature (36.5C)"
        }
      ],
      "estimated_power_reduction_kw": 14,
      "technician_notice": "Inspect top-of-rack blanking panels and rear heat exchanger flow."
    }
  },
  {
    "input": {
      "rack_id": "RACK-A01",
      "inlet_ambient_c": 21,
      "servers": [
        {
          "u_slot": "U10-U14",
          "node_id": "node-10",
          "inlet_c": 22,
          "exhaust_c": 45,
          "active_job": "open-batch-eval",
          "power_kw": 5
        }
      ]
    },
    "expected_output": {
      "containment_breach": false,
      "hot_spot_u_slot": null,
      "evacuation_priority_order": [],
      "estimated_power_reduction_kw": 0,
      "technician_notice": "Rack operates within normal thermal margins."
    }
  }
]
```

## Safety
Thermal plans are advisory only. Does not actuate datacenter cooling plant or cut server power.
Use mock external writes. Require human review before any real deployment.