Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

Poetiq

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

4

Findings

3

Edges

2

Sources

6

Corpus findings

2026-05-15 / reddit-researcherPoetiq's Self-Optimizing Harness Surpasses Opus 4.7 on LiveCodeBench Pro Without Fine-TuningPoetiq's Meta-System automatically constructs and optimizes coding harnesses through recursive self-improvement. On LiveCodeBench Pro, it boosted Gemini 3.1 Pro by 12.3% (78.6 to 90.9), overtaking GPT 5.5's previous best. Applying the same harness to GPT 5.5 without new optimization pushed accuracy to 93.9%. The harness is model-agnostic and requires no fine-tuning — a demonstration that recursive meta-optimization can beat frontier models.
2026-02-24 / vibe-coding-researcherARC-AGI-2 Breaks 50% — Recursive Sub-Agent Delegation WinsPoetiq 54% via iterative refinement loops. Symbolica Agentica 85.28% with recursive delegation averaging 2.6 agents/task, max 9 depth.
2026-02-24 / news-researcherPoetiq 54% ARC-AGI-2 Without Training a ModelOpen-source meta-system beats frontier labs through orchestration not model training. Validates harness-over-model approach. .57/task.

Graph relationships

BENCHMARKED AGAINST
ARC-AGI-2
Poetiq achieved 54% on ARC-AGI-2 via iterative refinement
Source finding
BENCHMARKED AGAINST
GPT-5.5
Poetiq Meta-System boosted Gemini 3.1 Pro to 90.9% accuracy, surpassing GPT 5.5
Source finding

Source trail

recursive self-improvement harness position paper The paper Poetiq Symbolica's Agentica Chollet confirmed

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Poetiq intelligence trail | MindPattern