Veridical Tech

// Research Docs

SPM Overview

Public-safe summary of what SPM is for, what failure modes it is evaluated against, and how this site frames proof.

What SPM is

StellarPath Memory System (SPM-Core) is a precision long-context memory system for AI. It is evaluated here as a context optimization program for fact-sensitive long-horizon work. SPM-Core is the primary public system. Standard RAG, hybrid RAG, and Mem0 are used as comparison baselines.

What SPM is tested against

The public test surface focuses on conditions that matter in production: early facts buried under noise, similar-fact confusion, environment separation, configuration lookup, long debug context, and long-range planning constraints.

  • Does the system preserve the right fact under long noisy history?

  • Does it keep similar facts separated instead of contaminating answers?

  • Does it explicitly abstain when a question cannot be answered?

  • Does it remain useful when history size becomes operationally extreme?

Public proof standard

This site uses a black-box standard. The public discussion stays on benchmark setup, datasets, systems, and results. Internal implementation detail is intentionally out of scope.

LHCSB primary score

85.57%

BABILong full-250

99.60%

T5 abstention accuracy

100%

Formal test samples

1,461 across 5 benchmarks