10M persistent-memory benchmark run
A persistent-memory streaming run configured with 10,000,000 sample facts. Its 10,833,323 learned-fact count is cumulative because the run resumed prior state; both values count facts, not model parameters.
RUN-LEVEL BENCHMARK
Original run decision
The original 10M persistent-memory run obtained PASS according to its own recorded criteria. The measurements on this page belong to that run and are not combined into a global Elena score.
Later extended-suite result
A subsequent extended suite obtained an overall FAIL. That later result remains separate and prevents the earlier run from being presented simply as a ‘10M suite PASS’.
Correct interpretation
10M means 10,000,000 configured sample facts. It is a fact-count label, not a parameter-count label. The final learned-fact count is cumulative because the run continued from earlier state.
Runtime boundary
The benchmark documents a run-level persistent-memory milestone. It does not establish that SenzaAdam, Big Brain, or Clean Speak is the universal current runtime.
Recorded run metrics
- Configured sample facts
- 10,000,000
- Cumulative learned facts
- 10,833,323
- Cells created
- 394,856
- Cell reuses
- 10,438,467
- Recall
- 0.9883
- Anti-forgetting
- 1.0000
- Average active cells
- 1.00
- Brain file
- 2,358.25 MB
- Adam
- Absent
- Global backward
- Disabled