ENTITY Documentation Portal
ENTITY test 15 | fresh execution | 2026-10-01

Fresh storage proof with contemporaneous video

Two fresh storage qualifications were executed on the BTG workstation and recorded while they ran. The Level-3 test measures the compact bond+index representation against a defined 768-d float32 vector baseline. The installed-storage audit measures actual ADAM development-state bytes across repetitive, mixed and unique workloads.

Overall: PASS. Level-3: 250,000,000 bonds, 50,000,000 entities, 4,400,000,000 measured bytes, 153,600,000,000-byte vector baseline, 97.1354166667% lower, 1,000/1,000 sampled lookups correct. Installed audit: 20/20 checks PASS.

A. Level-3 250M-bond proof

The VP8 WebM was generated contemporaneously by the runner while the benchmark child process wrote the 4.0 GB bond file and 0.4 GB index. Video SHA-256: ABEB2A86D422D3011FC2232FD577AC900FE4781747268754C71D410750D7DD78.

MeasureFresh result
Bonds / entities250,000,000 / 50,000,000
Bond file4,000,000,000 bytes
Entity index400,000,000 bytes
Total representation4,400,000,000 bytes
768-d float32 baseline153,600,000,000 bytes
Controlled reduction97.1354166667%
Sampled retrievals1,000 / 1,000 correct

Bond SHA-256 17403382dd33f2ae573d1c35f91584530de57e0e3c2a13b6d7b7a7094a2c2f9e
Index SHA-256 c7aa7fa07ed15b597064b345dd2f925b4636a5d6c82dd79483f117b80e583e93

Those two hashes were recomputed after the run by a second verifier and matched both the fresh result and the historical Level-3 qualification.

B. Installed ADAM storage audit

Fresh installed audit: 20/20 checks PASS. Video SHA-256: 090B8BD91D6517AA9B8E50DCFD5484F3B46EC406A614746DAAA741B870AD2EC8.

WorkloadSQLite bytesADAM factored bytesReduction
Repetitive167,93685,12349.312238% lower
Mixed131,072121,0837.621002% lower
Unique253,952446,45375.802120% larger

Every workload also passed exact-view and restart-recovery checks. The separate high-entropy exactness case expanded by 20.240402% on this run.

Evidence bundle

The machine-readable bundle binds the two runs, hashes and source boundary to ENTITY main 33eb9bd53870126a84c9c5529183faa5d0289a45.

Claim boundary: The 97.1354166667% figure is proved only for the controlled compact bond+index representation versus the defined 768-d float32 vector baseline. The installed audit independently demonstrates that storage behavior is workload-dependent and may expand for unique/high-entropy data. This is not universal compression and not proof that every ENTITY workload saves 97%.