Compression ratio 2.8x below the ~9x measured on real payloads
Everything is in Postgres on purpose. That only holds while gzip, content-addressing and the interval model all still work — this page is how you notice one of them stopped.
| Metric | Value | Why |
|---|---|---|
| slot rows | 187 | one row per slot, written on arrival and departure — not per poll |
| slot rows live | 187 | vanished_at is null |
| slot table | 0 B | partitioned monthly; expiry is DROP PARTITION, never DELETE |
| slot_sync table | 48.0 KB | one cursor row per (clinic, service) — what an unchanged pass writes |
| raw_snapshot rows | 59 | content-addressed; a repeat payload bumps a counter instead |
| raw_snapshot on disk | 2.0 MB | gzipped bytea, storage external |
| raw_snapshot uncompressed | 5.5 MB | what the same payloads would have cost verbatim |