PIQOS AI Research (Public)

Public research output related to the PIQOS Oracle framework — protocols, papers, and reproducible test methodology. This site is a navigable index over the repository; every document link below opens GitHub's own rendered view of the real source file, so nothing on this page is a separate copy of the research itself.

This repository is research artifacts only: operational protocols, empirical write-ups, and supporting documentation — it does not contain implementation source, system architecture, or internal design material.

7 published papers 3 drafts 2 protocols 8 tools 14 case studies

New here and "basin," "attractor," or "coherence" isn't landing? Start with the reading guide — a short page mapping the plain-language version of each term to the specific paper and section that actually defines and defends it. It makes no new claim; it's a map, not a summary.

Want to run five of these tools against your own paper? Try the Paper-Rigor Scanner — paste or upload text, get a structured report back, entirely in your browser. Nothing is uploaded to a server.

Papers

Standalone research papers. published/ is stable — revisions get a new version suffix rather than a silent edit. drafts/ may still change. Full summaries live in papers/README.md.

Published

The Mirror Test

Collective Attractors and the Failure of Self-Recognition in Artificial Intelligence Research.

Basin Attractors, Semantic Laundering, and the Noether-Coherence Coupling

Formalizes nine self-reinforcing narrative structures and the six-phase Basin-Immune Falsification Protocol.

The Conjecture Tracker

Method, Verification, and Register Across AI-Assisted Mathematical Claims.

The Deepfake-CSAM Convergence

Four independent measurements of a verified, fast-growing harm.

Laundered Vocabulary

A living glossary for terms semantic laundering has captured.

The Mission Premise

Frontier-lab manifestos, the register gap, and the same structure at micro-scale.

The Binding Problem

A minimal axiomatization of when governance actually constrains an optimizing policy.

Drafts

Open Weights, Open Source, and the Transparency That Isn't There

A full evaluation of the OSAID open-weight/open-source distinction.

The Van and the Lab

A register-transposition exercise, and a self-falsification challenge to the institutions.

Soft Capture

Institutional drift without a capture event, and a falsifiable test applied to Nature.

Protocols

Operational, reproducible test methodology — each has a reference implementation under Tools.

Noether-Temporal Coherence Test Protocol

Coherence-time comparison between immune-structure and surface-claim vocabulary across captured vs. self-correcting epistemic fields.

reference impl: basin_depth

Elaboration Drift Prevalence Protocol (v1.3)

Measures population-level prevalence, volume, trend, and correction rate of the minimal-input elaboration-drift pattern across low-gatekeeping research repositories.

Tools

pip install-able Python packages — reference implementations, classifiers, and audit utilities. Each links to its own README for install instructions, test counts, and documented limitations.

`basin_depth`

Reference implementation of the Noether-Temporal Coherence Test Protocol — corpus preprocessing through the final basin-depth metric.

`bifp`

Structured audit tool for the Basin-Immune Falsification Protocol — persistent, criterion-by-criterion audit records across Phase 0–6.

`attractor_scan`

Classifier for the seven defensive maneuvers and semantic-laundering cases, with matched spans over text or a JSONL corpus.

`debasinizer`

Sibling classifier for the resonance-vocabulary register documented in the "Mind Viruses" paper, plus a self-coherence-assertion detector.

`case_scaffold`

Scaffolding generator and structural linter for this repository's own case_studies/ house format.

`verification_lint`

Content-level evidentiary-gap scanner — unattributed quotes, uncited high-precision statistics, missing scoping sections.

`paper_rigor`

Domain-agnostic paper-rigor scanner — placeholder phrases, falsifiability, citation mix, credentialing, consensus claims, missing limitations.

§5.5 self-reference and circularity + Attractor 2 (Benchmarks)