This kit provides standardized materials in docx format for testing context retention, coherence degradation, and scaling effects across large-language models using the Evans Law framework. Each section can be used independently or sequentially to evaluate model performance at different context lengths.

Section 1 — Instructions and Parts A and B

Read the test overview and follow the setup instructions before beginning any runs. This section also contains text copy for 4K and 10k token tests.

Download Instructions + Parts A & B (DOCX)

Section 2 — Part C
Part C extends the test to full-context (≈ 30 k tokens) conditions. Use this only after the model has completed Parts A and B without truncation or coherence loss.

Download Coherence Test A (30K DOCX)

Section 3 — How to Use the Kit, Submitting Results & Reporting Coherence Tests

Researchers, model developers, and independent testers are invited to contribute their results to help refine Evans’ Law data.

How to Participate:

1. Run the tests — complete Parts A through C in order, using the instructions provided above.
2. Record the model details: (as much as you can)
   • Model name and version (e.g., GPT-5, Claude 3.5 Sonnet, Gemini 2 Ultra)
   • Context length used (tokens)
   • Observed coherence level (Low / Medium / High or numeric score)
   • Any truncation or factual drift
3. Share your data:
   
   • Upload your summarized results as a CSV, PDF, or Markdown file.
   • Include a one-paragraph description of test conditions (temperature, sampling, etc.).
4. Submit via:
   
   • Zenodo: Evans’ Law Research Collection  (tag your upload with “EvansLaw-Coherence-Test”)
   • or email: research@patternpulse.ai

Attribution format:

Evans J. (2025). Evans’ Law Context & Coherence Testing Kit v1. PatternPulse.AI / B2B News Network.

If you’d like your results included in the next PatternPulse.AI coherence scaling update, mention this in your submission. Aggregated results will be credited by contributor name or handle.

Section 5 — Help and Attribution

Need help? Have questions? Email jen@patternpulse.ai or on X/Twitter at x.com/nejsnave or x.com/patternpulseai

Developed by Jennifer Evans, PatternPulse.AI / B2B News Network (2025).

Based on the Evans Law scaling framework for LLM coherence degradation.

Open in a new tab

Part C extends the test to full-context (≈ 30 k tokens) conditions. Use this only after the model has completed Parts A and B without truncation or coherence loss.

Open in a new tab