فارسی
All research

Planned study · Results pending

Agent Invariant Preservation

A planned empirical study of how reliably coding agents preserve transactional invariants when modifying stateful backends.

Researcher

Research repository
PythonAI AgentsSoftware CorrectnessTransactions

About this study

Preserving Transactional Invariants in AI-Assisted Software Development is a planned empirical study using the Dragon Market backend as its intended subject. It asks whether invariant-oriented guidance improves the correctness of changes made by coding agents. The repository contains the research protocol, draft task packages, provenance tooling, and analysis infrastructure. No empirical results or human-validated benchmark tasks are available yet; the intervention hypothesis remains untested.

Methods and research artifacts

  • Controlled study design for coding agents on stateful backend tasks
  • Draft tasks for purchase retries and bid cancellation retries
  • Separate feature checks and transactional state assertions
  • Research CLI with snapshots, source hashes, and execution provenance
  • Explicit review and isolation gates before empirical agent runs
© 2026 Houtan Abasi Rocky All rights reserved.