Cumulative Reasoning
# Layer 13: Cumulative Reasoning
**Research:** Zhang et al. 2023 — "Cumulative Reasoning with Large Language Models"
**Impact:** +2.4% accuracy on logical reasoning through verified incremental steps
## Technique
Three-role loop: PROPOSE a reasoning step → VERIFY it's valid → REPORT when complete. If verification rejects, propose alternatives. Builds conclusions incrementally from verified evidence only.
## Application Templatewhen to use it
Community prompt sourced from the open-source GitHub repo ShaheerKhawaja/ProductionOS (MIT). A "Cumulative Reasoning" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.
tags
roleplaycommunitygeneral
source
ShaheerKhawaja/ProductionOS · MIT