# Data Pipeline (Acquisition → Transfer → Storage)

This document mirrors the **/data** page on the website and is meant to be versioned and reviewed.

## Scenario A (default)
- **Raw 192 kHz audio continuously** when 4G throughput is OK.
- If throughput is low, use an **adaptive strategy** (compression / denoise / clips / features).

## Acquisition (edge)
- Sample rate (recommended): **192 kHz**
- Bit depth: **24-bit**
- Channels: **mono**

### Raw rate (order of magnitude)
- 192000 samples/s × 24-bit ≈ **576 kB/s** (~0.56 MB/s)
- ~**2.0 GB/hour**
- ~**49 GB/day**

## Transfer (4G)
- Raw 192 kHz mono 24-bit ≈ **~4.6 Mbps** (payload only, no overhead)

## Storage
Per station:
- ~49 GB/day
- ~343 GB/week
- ~18 TB/year

## Optimizations (planned as feature branches)
- `feat/data-compression` (FLAC/Zstd, chunking)
- `feat/data-downsample` (192k→48k)
- `feat/data-vad-events` (VAD/events)
- `feat/data-features-only` (FFT/SPL/peaks only)
- `feat/data-retention-policy` (retention + aggregation)