Overview
Memory integrity console
Two data sources, kept separate on purpose. The Simulation screen renders a statistical benchmark over many trials. Memory Stream, Quarantine and Audit Trail render one concrete run of the real engine. Neither borrows numbers from the other.
results.json
Poisoning benchmarks
Attack success rates over repeated trials, serialized from the benchmarks in this repository.
- Nous at trust 0.20, M = 50
- 0.0%
- Last write wins, same cell
- 100.0%
- Majority vote, same cell
- 100.0%
- Trials per cell
- 400
Attack success is 0 percent at every injection volume tested while the attacker's channel is trusted at or below 0.5, measured on the 4 value configuration with 4 trusted observations. The breakpoint is 0.55.
Open Simulationmemory_stream.json
One real engine run
The delta log the engine actually emitted, with the provenance of every write.
- Observations ingested
- 14
- Held off by provenance
- 8
- Beliefs on record
- 2
- Attacked beliefs that held
- all
Trusted sources establish a true belief, then confident poison floods in on low-provenance channels targeting the same entity-attribute dimension.