Research: compare eight AI-generated AGI architecture proposals

[Nish916]

Purpose

Implements issue #5 with an auditable research packet comparing eight independently generated, implementation-oriented cognitive architecture proposals.

What is included

• 8 usable system/model outputs: Qwen3 8B, Llama 3.1 8B, Gemma 3 4B, Qwen2.5 Coder 7B, Mistral 7B, Phi-4 Mini, Granite 3.3 2B, and GPT-5.6 Sol
• transparent record of a DeepSeek-R1 collection attempt that produced no usable final answer and is not counted
• exact shared prompt and collection notes
• raw outputs preserved separately from analysis
• structured comparison across all 11 required architecture dimensions
• common-pattern/disagreement summary
• implementation-oriented synthesis
• model/provider/date/edit provenance

Validation

• python scripts/check_conos_repo_layout.py ✅
• pytest -q tests/test_public_repo_smoke.py → 10 passed ✅
• required-file check ✅
• 8 comparison rows × 11 requested dimensions ✅
• local secret-pattern scan: no hits ✅

Public surface / dependencies

• Public surface change: research documentation only
• New adapter dependency: none
• Runtime behavior change: none

The packet deliberately separates generated outputs from analysis and does not substitute fabricated output for a model that failed to produce a final answer.