Skip to main content
LatticeAG

Poly Series

PolyGnosis

Hermes Skill

Hermes Agent skill for adversarial multi-model consensus — independent solve, critique, formal scoring, verified synthesis.

Problem

Models can agree on the wrong answer. Surface-level consensus hides flawed reasoning and produces unreliable outputs for high-stakes tasks.

Solution

PolyGnosis runs an adversarial multi-model consensus protocol — independent solves, adversarial critique, formal scoring with RRF and Borda, verified synthesis, and a quality gate with meta-review.

How it works

01

Orchestrator

02

Solvers

03

Critic

04

Scoring

05

Synthesis

06

Quality Gate

07

Meta-Review

When to use

  • - High-stakes answers where agreement is not enough.
  • - Research synthesis that needs explicit critique.
  • - Comparing multiple model outputs with a formal scoring protocol.

Not for

  • - Latency-critical chat completions.
  • - Tasks where there is no way to score or verify candidate answers.

Features

Independent candidate solutions

Adversarial model critique

RRF + Borda scoring

Quality gate and meta-review

Install / usage

usage.sh
lattice polygnosis solve \
  --question "Which answer is best supported?" \
  --protocol adversarial-consensus \
  --scoring rrf,borda

Architecture

Problem

Models can agree on the wrong answer. Surface-level consensus hides flawed reasoning and produces unreliable outputs for high-stakes tasks.

Solution

PolyGnosis runs an adversarial multi-model consensus protocol — independent solves, adversarial critique, formal scoring with RRF and Borda, verified synthesis, and a quality gate with meta-review.

View on GitHub

Related

More from this series