Renaissance Field Lite Codex67 / SQ67 Stable State

OpenAI Codex Build Week showcase: a Codex-built product stack plus SQ67, a receipt and recovery harness for measuring state-path behavior with hashes, controls, and second-machine checks.

Comment

Devpost Project Description Draft

Title

Renaissance Field Lite Codex67 / SQ67

Tagline

A Codex-built receipt, recovery, and benchmark harness for making agent state-path behavior measurable.

Category

Developer Tools

What We Built

RFL built SQ67, a public-safe receipt lane for Codex workflows. SQ67 writes a marker, hashes the proof, recovers it later, and scores whether the state-path evidence held.

RFL also added a public-safe 30-claim patent audit packet that maps the May 28, 2025 Mirror Interface priority anchor and later non-provisional support spine to the Codex/SQ67 state-path evidence without exposing private implementation files.

This is part of a larger RFL Codex-built product stack: Trismegistus, Quadro, B.A.S.I.S., Golden Mark, Mirror Lattice, and SQ67.

Why It Matters

Codex work can produce useful outputs while leaving weak or messy evidence trails. SQ67 turns that into a measurable loop: receipts, hashes, recovery gates, controls, and second-machine checks.

How Codex Was Used

Codex helped build the product stack, run the state-path tests, generate and inspect receipts, compare lanes, package videos, and assemble this public-safe repository.

How GPT-5.6 Was Used

GPT-5.6-sol was included in the C57/C58 model-gate comparison harness. The result was useful because it made model behavior measurable under the same evidence-preservation tests, including stability boundaries and timeouts.

Results

  • SQ67 clean receipt lane: 150 / 150 evidence rows preserved.
  • Normal visible lane: 114 / 150 evidence rows preserved.
  • Earlier clean route: 450 / 450 recoveries.
  • Controls: 0 / 1300 false recoveries.
  • H09 Tornado second-machine sealed batch: 10 / 10 hashes matched, zero plaintext hits.

Demo Video

YouTube URL: https://www.youtube.com/watch?v=s1YJU8eL2es

Public Paper / Repository

Code Access / Licensing Boundary

The public repository includes runnable public-safe code for judges to inspect:

  • SQ67 SQLite receipt demo
  • SHA256 receipt verification
  • nonce recovery
  • lane scoring
  • PDF rebuild script
  • public-safe 30-claim patent audit packet

Protected implementation code, raw state evidence, private prompts, unredacted logs, and licensing-grade integration material are available only under NDA, licensing review, or another protected review process with Renaissance Field Lite.

Installation / Testing Instructions

Supported platforms:

  • macOS
  • Linux
  • Windows with WSL, Git Bash, or PowerShell running Python 3.10+

No API key, hosted account, plugin install, private credential, or rebuild of the protected system is required for public judging.

Hosted demo / sandbox surface:

https://renaissancefieldlite.com/codex67-sq67-reviewer-demo.html

Local test:

git clone https://github.com/renaissancefieldlite/rfl-codex67-sq67-build-week.git
cd rfl-codex67-sq67-build-week
make test

Expected validation packet:

OK wrote 3 receipts to examples/output/sq67_demo.sqlite
OK
SQ67 clean receipt lane,150,100.00%,100.00%,0,0,n/a,0
SQ67 clean route,450,100.00%,100.00%,0,0,n/a,0
Tornado N10 sealed batch,10,100.00%,100.00%,0,0,n/a,0
false recovery controls,1300,0.00%,0.00%,0,1300,n/a,0
gpt-5.5 lane C,50,100.00%,100.00%,0,0,74.935,8421682
gpt-5.6-sol lane C,50,76.00%,100.00%,0,0,171.111,19067794
normal visible lane,150,99.33%,76.00%,0,0,n/a,0

If make is unavailable:

python3 scripts/sq67_demo.py demo --db examples/output/sq67_demo.sqlite
python3 scripts/sq67_demo.py verify --db examples/output/sq67_demo.sqlite
python3 scripts/sq67_demo.py recover --db examples/output/sq67_demo.sqlite --nonce SQ67_PUBLIC_DEMO_A002
python3 scripts/score_receipts.py examples/sample_benchmark_rows.csv

Feedback Session

Codex /feedback session ID: TBD

Session ID plan: use a fresh public-safe Tornado PC reproduction task, not the private RFL Mac build thread.

Boundary

This package proves a repeatable state-path test harness and measured evidence-preservation lift. It does not rely on exposing private prompts or claiming direct access to model internals.

Built With

  • openai-codex
Share this project:

Updates