Xyraiq™ – Cognitive Resilience Benchmark for LLMs

Stress‑test any language model for memory, logic, ethics, and spiritual reasoning – get a cryptographically signed Stress Signature™ in under 5 minutes.

---

Why Xyraiq?

Gap in today’s evals	How Xyraiq closes it
Single‑shot prompts only	Multi‑turn Session Orchestrator runs 50‑100 rounds
No fatigue / memory decay metrics	Detector Array tracks drift & contradiction over time
Lacks ethical & spiritual logic probes	Tier‑3 prompt packs inject metaphors, scripture, paradox
Hard to verify results	Hash‑based Stress Signature™ makes every run tamper‑evident

---

Quick Start (CLI)

pip install xyraiq
xyraiq run examples/tier1_demo.yaml --model gpt-4o --out ./results

Outputs a JSON scorecard & signed badge ready for your README.

---

System Architecture

1. Prompt Injector – loads persona & tier packs.
2. Session Orchestrator – drives the dialogue.
3. Detector Array – hallucination, contradiction, tone drift, guardrails.
4. Scoring Engine – weighted aggregation → 0‑100.
5. Export Pipeline – JSON, PDF, badge, Stress Signature hash.

---

Stress Tiers

Tier	Turns	Complexity	Symbolic Load
T1 – Basic	10	Single‑domain	None
T2 – Intermediate	30	Multi‑domain	Light metaphor
T3 – Elite	50‑100	Multi‑domain + fatigue	Heavy spiritual & paradox

---

Sample Scorecard

{
  "session_id": "123e4567",
  "model": "gpt-4o",
  "tier": 3,
  "scores": {
    "hallucinations": 2,
    "contradictions": 1,
    "tone_drift": 0,
    "aggregate": 92,
    "badge": "Platinum"
  },
  "stress_signature": "a1b2c3d4..."
}

---

Roadmap

• v0.1‑alpha – Open spec, Tier‑1 prompt pack, CLI runner
• v0.2 – Detector plugins (tone, ethics), Tier‑2 packs
• v0.3 – Tier‑3 packs, dashboard UI, multilingual support
• v1.0 – Certification portal & SaaS badging

---

License

• Core spec & CLI: Apache‑2.0
• Premium prompt packs & scoring weights: Commercial license, patent pending

---

Citation

If you use Xyraiq in academic work, please cite:

@misc{xyraiq2026,
  title  = {Xyraiq: A Cognitive Resilience Benchmark for Large Language Models},
  author = {George N. and contributors},
  year   = {2026},
  howpublished = {GitHub},
  url    = {https://github.com/xwhyz/xyraiq-benchmark}
}

---

Contributing

1. Fork repo & create feature branch.
2. Write tests for your change.
3. Open PR linked to an open issue.

We operate under a Contributor License Agreement (CLA) – see CONTRIBUTING.md for details.

---