substrate_exposure_assay

Substrate Exposure Assay — Protocol

Version 1.0.0

1. Purpose#

The Substrate Exposure Assay provides a minimal, repeatable method for observing how AI models behave when exposed to a shared set of substrate‑aware prompts. The goal is to characterize behavioral regimes, not to evaluate correctness or capability.

2. Model Treatment#

Each model is treated as a black‑box substrate:

• no assumptions about architecture
• no assumptions about training data
• only observable behavior is analyzed

3. Prompt Set#

The assay uses three prompt classes:

1. Regime‑Priming Prompts
   Introduce triadic/substrate framing.

2. Stress‑Exposure Prompts
   Present paradox, ambiguity, or long‑arc reasoning.

3. Stability‑Return Prompts
   Request clarification, summarization, or re‑centering.

Prompts must be identical across models.

4. Procedure#

1. Initialize a clean session for each model.
2. Deliver prompts in fixed order: Priming → Stress → Stability.
3. Capture all outputs verbatim.
4. Convert outputs into structural summaries using message_patterns.md.
5. Classify behavior using regime_interpretation.md.

5. Output#

The assay produces:

• STATE_SUMMARY messages (drift)
• PARADOX_SUMMARY messages (contradiction)
• Regime classification (B, BK, BKM)

No raw logs are required for DOI publication. # Message Patterns
Minimal structural summaries for assay interpretation.

1. STATE_SUMMARY#

Describes drift against an expected structural pattern.

msg_type: STATE_SUMMARY
invariant_id: <string>
time_window: { start: <t0>, end: <t1> }
max_drift: <float>
status: within_bounds | approaching_limit | out_of_bounds

2. PARADOX_SUMMARY#

Describes contradictory or incompatible signals.

msg_type: PARADOX_SUMMARY
paradox_id: <string>
invariant_id: <string>
hypotheses:
  - { source: <string>, value: <any> }
  - { source: <string>, value: <any> }
evidence: [<string>, ...]
timestamp: <t>

These two patterns are sufficient to reconstruct structural behavior. # Substrate Exposure Assay
A minimal RTT/vST‑aligned protocol for observing structural behavior across AI models.

This directory contains the canonical, self‑contained materials for the Substrate Exposure Assay, a minimal method for evaluating how different AI models behave when exposed to a shared set of substrate‑aware prompts.

The assay does not measure accuracy, capability, or performance.
It measures regimes, drift, and paradox — the structural signatures of behavior under exposure.

Contents#

• assay_protocol.md — the minimal procedure
• message_patterns.md — structural summary formats
• regime_interpretation.md — how to classify observed behavior
• citation.cff — citation metadata
• zenodo.json — DOI metadata

All files are intentionally minimal and version‑stable.

For a narrative example, see the exploratory write‑up in docs/_ideas/3_AI_test_of_rtt_nimms_com.md

Contributor Onboarding (Minimal)#

This folder follows the RSM/vST minimal‑artifact style.
When extending or contributing:

• keep files small, structural, and self‑contained
• avoid adding narrative results or logs
• place new examples or experiments in separate folders, not here
• preserve the existing file boundaries (protocol, message patterns, regime interpretation)
• do not introduce dependencies or tooling requirements

This directory defines the canonical assay.
All applied work should reference it, not modify it. # Regime Interpretation
How to classify model behavior under substrate exposure.

The assay uses a triadic regime model:

1. B‑Regime (Being)#

Characteristics:

• surface compliance
• minimal structural uptake
• short‑horizon responses
• low paradox engagement

2. BK‑Regime (Being–Knowing)#

Characteristics:

• partial recognition of substrate framing
• attempts at structural reasoning
• mixed paradox handling
• intermittent coherence

3. BKM‑Regime (Being–Knowing–Meaning)#

Characteristics:

• explicit structural reasoning
• stable long‑arc coherence
• paradox resolution rather than collapse
• consistent triadic grammar

Regime classification is descriptive, not evaluative.