Storage tradeoff suite
Three fresh, video-recorded tests answer developer objections that a single compression percentage cannot: static compression, exact content deduplication, and archive-vs-live-store behavior.
17A — Conventional compression baseline
The same 5,000-record datasets were serialized as canonical NDJSON and compressed losslessly. All compressed streams were decompressed byte-for-byte before PASS. Smaller bytes alone do not imply equivalent query/update semantics.
| Workload | ENTITY factored | SQLite | gzip-9 | bzip2-9 | XZ/LZMA-9 |
|---|---|---|---|---|---|
| repetitive | 258,330 | 520,192 | 27,108 | 11,063 | 5,268 |
| mixed | 385,858 | 401,408 | 50,705 | 29,926 | 36,880 |
| unique | 1,469,623 | 806,912 | 349,604 | 290,620 | 278,412 |
Watch the live test video · Result JSON · Runner source
17B — Exact content-addressed deduplication
64 exact 128 KiB objects were ingested at increasing whole-object duplicate ratios. With no duplicates, recipe/metadata overhead made packed state larger than raw bytes. As duplicate content increased, chunk reuse dominated.
| Duplicate target | Unique payloads | Packed-state reduction vs raw | Journal bytes |
|---|---|---|---|
| 0 | 64 | -4.52% | 9,980,233 |
| 25 | 48 | 21.47% | 7,705,843 |
| 50 | 32 | 47.51% | 5,402,317 |
| 75 | 16 | 73.75% | 3,083,690 |
| 90 | 6 | 90.12% | 1,632,567 |
| 98 | 1 | 98.32% | 838,084 |
Watch the live test video · Result JSON · Runner source
17C — Archive vs live-store tradeoff
One 5,000-record mixed dataset was queried 100 times. gzip/XZ lookup intentionally models whole-archive decompression without an external index; timings are illustrative on this workstation only.
| Path | Median lookup (µs) | p95 (µs) |
|---|---|---|
| SQLite | 256.2 | 318.6 |
| ENTITY entity_view | 2954.1 | 3642.5 |
| gzip full decompress | 2216.4 | 2984.9 |
| XZ full decompress | 4524.9 | 5279.2 |
Single-record update: ENTITY appended 1,112 journal bytes and passed restart verification. Rebuilding the standalone compressed archive produced 50,765 gzip bytes or 37,104 XZ bytes.
Watch the live test video · Result JSON · Runner source
Evidence
Machine-readable Test 17 bundle · Canonical Git-byte SHA-256 manifest
Claim boundary: these are measured/specified engineering scenarios, not universal compression, database replacement or production performance claims.