Inspect time windows

Compare the authored record shape at each window stage.

How to use this explorer

Move between stages with the arrows or the numbered stage list. Each stage shows the input it receives on the left and the output it produces on the right, with the lines that changed marked. The configuration that makes the change sits below the comparison, and the final stage shows the complete pipeline.

The 5 stages in order

1. Original High-Frequency Events
   A synthetic sensor record before any window key or aggregate field is added.
2. Tumbling Window Aggregation
   The authored mapping assigns readings to fixed one-minute buckets and shows summary fields.
3. Sliding Window Trends
   The authored mapping places a summary into overlapping five-minute lookback windows.
4. Session-Based Activity Clustering
   The authored mapping groups fixture events using a five-minute inactivity threshold.
5. Multi-Level Configuration
   A review configuration sketches sensor, location, and global keys with external cache and output dependencies.

Interactive pipeline

TIME WINDOWS

Curated aggregation walkthrough

Stage1. Original High-Frequency Events2. Tumbling Window Aggregation3. Sliding Window Trends4. Session-Based Activity Clustering5. Multi-Level Configuration

Stage 1 of 5

Original High-Frequency Events

A synthetic sensor record before any window key or aggregate field is added.

0HighlightsAuthored emphasis only—not a computed diff.

Highlights only

Copy & download

Input

 { "sensor_id": "temp_001", "temperature": 72.3, "humidity": 45.2, "timestamp": "2025-01-15T10:23:45.123Z" }

Output

 { "sensor_id": "temp_001", "temperature": 72.3, "humidity": 45.2, "timestamp": "2025-01-15T10:23:45.123Z" }

Complete pipeline01-original-high-frequency-events.yaml

# No processing - just input pass-through
input:
  file:
    paths: ["sensor-data.jsonl"]

pipeline:
  processors: []  # No transformation

output:
  stdout:
    codec: lines