Verified release evidence
MADB 0.2.14

Measured on named hardware, against a declared corpus.

These figures are measured-machine observations and broad regression evidence. They are not universal performance guarantees or marketing SLAs.

Environment

Apple M5 Pro release run

HardwareApple M5 Pro, 18 cores, 24 GB
ModelMac17,9
OSmacOS 26.5.2 (25F84)
Episodes10,000
Prospective intentions1,000
Temporal observations20,000
Observed matrix

Release measurements

Source commit 08857e2; measured with the repository release benchmark on August 23, 2026.

MetricObserved
Typed write throughput25,141.354690 ops/s
Observed-time p500.037625 ms
Observed-time p950.048209 ms
Observed index entries30,000
Observed index approximate bytes740,256
Epistemic recall p951,435.798417 ms
Replay p95607.003959 ms
Prospective-list p9550.207333 ms
Executive-context p951,387.788458 ms
Metamemory p951.681375 ms
Exact-subject precision1.0
Due-intention recall1.0
Verification

What passed after final hardening

  • Complete Rust workspace test suite.
  • 192 of 192 MADB storage tests.
  • All named 0.2.14 cognitive and autonomous-agent integration suites.
  • Strict workspace Clippy with warnings denied.
  • Formatting and diff checks.
  • Final macOS ARM64 ABI3 wheel build and isolated Python 3.14 import.
  • MADB point-in-time storage and transport smoke tests on disposable stores.
Claim boundary

MADB uses functional inspiration from episodic memory. It does not implement a biological brain, simulate neural tissue or claim neurological equivalence. Default weights remain engineering priors; results vary by workload, hardware and corpus.