Build the application

## Build
Evaluate multi-stage cycle time variances from simulated Sparkplug B device streams. Generate dynamic pallet buffer rebalancing proposals to alleviate manufacturing bottlenecks.

## User
Industrial Operations Planning Lead

## Workflow
1. Consume simulated Sparkplug B DDATA metric updates capturing station execution durations.
2. Compute moving average cycle times to detect process drift and buffer accumulation.
3. Synthesize buffer rebalancing proposals to divert in-progress pallets to auxiliary cells.
4. Generate a supervisor sign-off sheet with projected throughput recovery metrics.

## Interface
Line balance histogram with drag-and-drop buffer allocation overrides and sign-off panel.

## Data
Create normalized synthetic JSON files of Sparkplug B DDATA payloads representing multi-station takt times. Store in local sandbox directory.
Starter examples below are synthetic. They are not downloaded records.

## Source
[Eclipse Sparkplug](https://raw.githubusercontent.com/eclipse-sparkplug/sparkplug/master/README.md)
Sparkplug B DDATA metric records and property metadata for process duration tracking described in Source 2.

## Tools
ollama, pydantic, fastapi.
Ollama runs efficient local models on edge plant hardware without sending operational data to cloud infrastructure.
Local quantized open model (e.g. 8B parameter class) capable of tabular numerical comparison.

## 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
- Detects stations operating at greater than 120% of target takt time.
- Proposes valid pallet buffer diversions without overfilling target auxiliary buffers.
- Calculates expected hourly production recovery under proposed dispatch routes.

## Starter Examples
```json
[
  {
    "input": {
      "line_takt_sec": 60,
      "stations": [
        {
          "station_id": "ST_PRESS",
          "current_cycle_sec": 85,
          "buffer_count": 8,
          "buffer_max": 10
        },
        {
          "station_id": "ST_PRESS_AUX",
          "current_cycle_sec": 55,
          "buffer_count": 2,
          "buffer_max": 10
        }
      ],
      "inbound_pallets": [
        "P_101",
        "P_102",
        "P_103"
      ]
    },
    "expected_output": {
      "bottleneck_identified": "ST_PRESS",
      "cycle_overrun_pct": 41.7,
      "diversion_proposals": [
        {
          "pallet_id": "P_101",
          "divert_to": "ST_PRESS_AUX"
        },
        {
          "pallet_id": "P_102",
          "divert_to": "ST_PRESS_AUX"
        }
      ],
      "projected_takt_recovery_sec": 62
    }
  },
  {
    "input": {
      "line_takt_sec": 60,
      "stations": [
        {
          "station_id": "ST_PRESS",
          "current_cycle_sec": 58,
          "buffer_count": 3,
          "buffer_max": 10
        }
      ],
      "inbound_pallets": [
        "P_104"
      ]
    },
    "expected_output": {
      "bottleneck_identified": "NONE",
      "cycle_overrun_pct": 0,
      "diversion_proposals": [],
      "projected_takt_recovery_sec": 58
    }
  }
]
```

## Safety
Read-only queue analysis. Output plans require industrial engineer approval before floor execution.
Use mock external writes. Require human review before any real deployment.