Papers
Research papers, mechanism papers, synthesis, and boundary results.
- 2026-02Pick a concrete, non-toy problem. Draft a short set of rules. Apply them to different AI agents. See whether the behavior shifted in reproducible ways.
When Ai Collapses Fact And Assumption
- 2026-04Blended inference is the baseline response mode of LLMs. Smooth prose is the goal. In software, that smoothness can hide the boundary between grounded analysis and inferred assumptions.
- When AI Collapses Fact and Assumption v1.4Older versionBlended inference is the baseline response mode of LLMs. Smooth prose is the goal. In software, that smoothness can hide the boundary between grounded analysis and inferred assumptions.
Making Ai Omitted Scope Visible With Vdg
- 2026-04I designed an experiment to demonstrate omitted scope using a caching system in Python. It’s just one file.
Ai Answers Can Come With Silent Tech Debt
- 2026-03This is convenient. It’s what makes the vibe coding quickstart feel so magical. But it comes with the burden: implicit choices.
Stop Calling Every Ai Miss A Hallucination
- 2026-03Sometimes the model really did make something up. Fine, call that a hallucination.
Contract Authority Under Ai
- 2026-03I pursued the strongest claim I could make about contracts: that they could become the governing authority artifact for AI-assisted software work.
Thesis Experimental Methodology
- 2026-02In iterative development, workflows that externalize authority into a versioned contract exhibit measurably lower regression and drift than workflows that modify code directly via conversational prompts.
Contract Centered Iterative Stability
- 2026-02This paper identifies a stability boundary in iterative AI-assisted software development.
- Contract-Centered Iterative Stability v4.7.2Older versionThis paper identifies a stability boundary in iterative AI-assisted software development.
Why Verified Deduction Gap
- 2026-02AI systems produce probabilistic outputs. Identical prompts do not guarantee identical responses. Small contextual changes alter the conditional distribution of answers. In internal ticket experiments conducted under reset conditions, one pattern became clear: the central problem was not only correctness, but struct...
- Why Verified Deduction Gap v1.4.1Older versionAI systems produce probabilistic outputs. Identical prompts do not guarantee identical responses. Small contextual changes alter the conditional distribution of answers. In internal ticket experiments conducted under reset conditions, one pattern became clear: the central problem was not only correctness, but struct...
Contract Centered Engineering
- 2026-02Writing a production-grade system required sustained effort: design, implementation, review, integration, testing, deployment, and iteration. Producing even a single complete implementation demanded coordination across roles. Producing multiple independent implementations of the same system was rarely practical outs...
- Contract-Centered Engineering v2.16Older versionWriting a production-grade system required sustained effort: design, implementation, review, integration, testing, deployment, and iteration. Producing even a single complete implementation demanded coordination across roles. Producing multiple independent implementations of the same system was rarely practical outs...
Breaking The Loop
- Breaking the Loop v1.0Latest2026-02Large language models generate coherent text with high fluency. They are less reliable at preserving structural integrity across long iterative edits.
Protocols & Contracts
Explicit source-of-truth artifacts and response-structure documents that support the papers without pretending to be the final live authority surface.
Artifact Projection
- Date: March 1, 2026 Status: DRAFT (DS Stabilization: engine selection; engine preflight; extracted-text canonicalization; schema alignment)
Attack Types Triage Rules
- Attack Types & Triage Rules v0.8.1Older version
Convergence Contract
- This contract defines a deterministic, safe convergence process that synchronizes Google Docs “artifact” documents into version-controlled Markdown artifacts under artifacts/ . The system converges derived repository state to match the current set of eligible Source Documents while enforcing representation and safet...
- convergence-contract-v2.5.2Older version
- convergence-contract-v1.1Older version
Implementation Notes Extraction Protocol
Publishing Repository
- This contract defines the operational semantics for the Probabilistic Systems Engineering publishing repository.
- This contract defines the operational semantics for the Probabilistic Systems Engineering publishing repository.
Source Of Truth Extraction Protocol
Vdg Contract
- Reduce structural variance in probabilistic outputs by enforcing explicit separation of:
- VDG Contract Artifact v1.4.2Older versionReduce structural variance in probabilistic outputs by enforcing explicit separation of:
- vdg-contract-v1.4.1Older version
Replication & verification
Portable rerun support, comparison packets, and verification materials. Useful for repeatability and auditability, but secondary to the primary reading path.
Comparison Summary
- Comparison Summary v0.2Latest1. Purpose This document summarizes replicated evidence produced under the replication stage of the research program. Its function is to compare observed run outcomes against the shared thesis, methodology, and replication standard, and to determine whether the current evidence supports or weakens the central claim....
Failure Class
- 1. Purpose This document consolidates a recurring structural failure observed across the current replicated run set. Its purpose is to name the failure mechanism and the provisional failure class supported by repeated internal replication. This is a bounded consolidation artifact, not a claim of statistical generali...
Glossary
- Canonical Glossary v0.7LatestA durable artifact that governs what counts as correct for a system, workflow, or evaluation.
Methodology
- This methodology defines a reproducible experiment to measure iterative drift in AI-assisted software changes, comparing:
Program Spine
- Program Spine v0.4LatestThe purpose of this roadmap is to record the canonical spine of the research program.
Replication
- Replication v0.7LatestThis document defines the replication instruction artifact for rerunning the experiment.
Thesis
- Research Thesis v1.1LatestWhat remains under-tested is whether it preserves stability under iterative change.