Probabilistic Systems Engineering Proof

Proof boundary

What is proven here

The strongest claim did not survive. What is supported here is narrower: explicit source-of-truth artifacts materially reduce drift and improve invariant preservation in AI-assisted iteration, especially when guarantees span multiple mutation surfaces.

What is not proven here is that contracts themselves become the live authority surface. This site argues for a repeatable mechanism and a bounded research claim, not a universal proof about every model, team, or software domain.

What held

The strongest positive result is narrower than the earlier framing. When important guarantees span multiple mutation surfaces, explicit source-of-truth artifacts help preserve invariant scope during AI-assisted iteration. Start with Contract Authority Under AI v0.7.3 and Contract-Centered Iterative Stability v4.7.3.

What did not hold

Contracts did not become the place where the system actually permits, refuses, or constrains action. Once operationalized, effective control moved into downstream machinery such as manifests, fixtures, and evaluators.

Type of evidence

The evidence here is mechanism evidence: repeated structured runs, comparative tracks, clause-level evaluation, and explicit failure localization across deltas. Start with Contract-Centered Iterative Stability v4.7.3, the compact framing in Thesis & Experimental Methodology v3.0.1, and the broader engineering framing in Contract-Centered Engineering v2.17.

What remains bounded

  • This is a bounded engineering research corpus, not a universal benchmark claim.
  • The current evidence is strongest on cross-surface invariant-scope drift under iterative change.
  • The research supports explicit source-of-truth artifacts reducing drift, not contracts as the final authority surface.

What this work does not claim

  • It does not claim every AI-assisted workflow fails.
  • It does not claim all prompts are useless.
  • It does not claim contracts alone bind models or systems.
  • It does not claim statistical closure across every model, organization, or domain.

The closure move here is explicit boundedness: a clear claim, a clear evidence type, and a clear limit surface. Portable rerun material remains available in Comparison Summary v0.2 without becoming the front door.