Developer storage scenario suite
Six fresh, video-recorded scenarios map the storage advantage envelope instead of presenting one universal compression number.
What developers should take away
- Repeated structured facts: current-state factoring is consistently strong in these tests.
- Mixed data: savings narrow as value diversity rises.
- Unique/high-entropy data: factoring can expand storage substantially.
- Vector comparisons: the 4.4 GB Level-3 representation crosses a float32-per-entity vector baseline at 22 dimensions.
- History/provenance: retained temporal evidence can dominate storage; compare it to current-state databases only with that semantic difference stated.
- Known v0.41 temporal-cycle limitation: repeated logical bond identities can lose an earlier interval; tracked in ENTITY #116.
Scenario A - scale curve
Vacuumed SQLite current state versus ADAM native factored current state. Positive = ADAM smaller.
| Records | Repetitive | Mixed | Unique |
|---|---|---|---|
| 100 | 51.11% | 26.81% | -25.04% |
| 500 | 51.43% | 19.66% | -55.94% |
| 1,500 | 49.67% | 8.24% | -75.67% |
| 5,000 | 50.34% | 4.06% | -82.11% |
Raw JSON | Transcript | Script
Scenario B - value-cardinality break-even
3,000 records with progressively fewer reusable values. First non-saving tested point: 1,000 distinct values.
| Value cardinality | Factored vs SQLite |
|---|---|
| 1 | 42.18% |
| 3 | 41.36% |
| 10 | 41.44% |
| 30 | 39.35% |
| 100 | 35.36% |
| 300 | 24.14% |
| 1000 | -6.24% |
| 3000 | -75.10% |
Raw JSON | Transcript | Script
Scenario C - relationship density / fanout
25 million actual compact bonds were generated at each fanout. All 500 sampled lookups passed for every row.
| Fanout | Entities | Measured bytes | Index share | vs 768-d float32 |
|---|---|---|---|---|
| 1 | 25,000,000 | 600 MB | 33.33% | 99.22% |
| 2 | 12,500,000 | 500 MB | 20.00% | 98.70% |
| 5 | 5,000,000 | 440 MB | 9.09% | 97.14% |
| 10 | 2,500,000 | 420 MB | 4.76% | 94.53% |
| 20 | 1,250,000 | 410 MB | 2.44% | 89.32% |
Raw JSON | Transcript | Script
Scenario D - vector-dimension crossover
Derived from the freshly measured 4.4 GB / 50M-entity Level-3 representation. Baseline is one float32 vector per entity and excludes vector-database metadata/index overhead.
| Vector dims | Vector bytes | ENTITY reduction | Relation |
|---|---|---|---|
| 16 | 3.2 GB | -37.50% | VECTOR_SMALLER |
| 22 | 4.4 GB | 0.00% | EQUAL |
| 32 | 6.4 GB | 31.25% | ENTITY_SMALLER |
| 64 | 12.8 GB | 65.62% | ENTITY_SMALLER |
| 128 | 25.6 GB | 82.81% | ENTITY_SMALLER |
| 384 | 76.8 GB | 94.27% | ENTITY_SMALLER |
| 768 | 153.6 GB | 97.14% | ENTITY_SMALLER |
| 1536 | 307.2 GB | 98.57% | ENTITY_SMALLER |
| 3072 | 614.4 GB | 99.28% | ENTITY_SMALLER |
Raw JSON | Transcript | Script
Scenario E - retained provenance / history cost
200 entities with monotonic unique updates. SQLite stores final current state only; ADAM also retains event history. The current-state factored projection and authoritative journal are reported separately.
| Updates/entity | SQLite current | Factored current | Factored vs SQLite | Journal / SQLite | Initial time travel |
|---|---|---|---|---|---|
| 0 | 24,576 | 16,886 | 31.29% | 18.7x | PASS |
| 1 | 28,672 | 24,616 | 14.15% | 28.6x | PASS |
| 3 | 28,672 | 38,576 | -34.54% | 53.9x | PASS |
| 10 | 28,672 | 89,815 | -213.25% | 142.0x | PASS |
| 25 | 28,672 | 172,345 | -501.09% | 327.9x | PASS |
Raw JSON | Transcript | Script
Scenario F - known temporal-cycle limitation
This is a targeted regression, not a storage advantage. It reproduced the v0.41 interval-history limitation when a revoked deterministic bond identity is later reused.
EXPECTED_INITIAL={"_id": "0a01152898bb82aed9ffedfbded2f2f81855d14a19fa802ef56e86e3eac25112", "_key": "P-1", "_type": "project", "cost": 100, "note": "initial", "status": "active"}
ACTUAL_AT_INITIAL_SEQ={"_id": "0a01152898bb82aed9ffedfbded2f2f81855d14a19fa802ef56e86e3eac25112", "_key": "P-1", "_type": "project", "cost": 100, "note": "initial"}
DIFFERENCES={"status": {"actual": null, "expected": "active"}}status: active interval was no longer reconstructable after the same logical bond identity cycled. Tracked as ENTITY issue #116.Raw JSON | Transcript | Script
Evidence bundle
Machine-readable suite record | SHA-256 manifest