Research / Internal study

What retrieved context changes.

A 20-question internal comparison of answer correctness with and without retained project knowledge.

Context check · n = 2028 Sep 2026

Without context

15%

3 / 20 correct

With Enkeon

85%

17 / 20 correct

Answer correctness: 3 of 20 versus 17 of 200%25%50%75%100%Without contextWith Enkeon+70 pp
Answer correctness14 additional correct answers

Method

Twenty questions were generated from retained notes. The same model answered each question without recalled context and with Enkeon retrieval. A grader, not told which condition produced an answer, checked correctness against the retained note.

The 28 September 2026 report records 3 of 20 correct without context and 17 of 20 with Enkeon: a difference of 70 percentage points.

Scope and limitations

  • The retrieved study summary does not identify the model version.
  • This measures correctness for knowledge-specific questions, not general writing quality or coding ability.
  • Questions derived from retained knowledge favor a system with access to that knowledge; this is the intended comparison.
  • A single internal check does not establish uncertainty across repeated runs.
  • The chart connector expresses the difference between conditions, not measurements over time.

All research and methods ↗