AI-assisted development disclosure¶
AI coding assistants were used as engineering tools during HydroSwarm development. Repository history and project records include:
- ChatGPT / Codex;
- Claude / Claude Code;
- Freebuff.
They were used for implementation assistance, debugging, test generation, documentation review, code/architecture critique, and repository maintenance.
The project author retained responsibility for scientific objectives, architecture choices, experiment governance, split/lock policy, promotion decisions, claims, safety boundaries, and final release decisions. Generated suggestions were subject to repository tests, frozen protocols, hash/provenance checks, and human review.
Scientific boundary¶
The final HydroCore-v5 model is not a hosted general-purpose LLM. It is a locally loaded scientific model trained for the governed Sentinel task family on synthetic WNTR/EPANET-generated evidence. ChatGPT, Claude, and Freebuff are not inference components and have no runtime authority.
No hosted AI API is required for HydroSwarm's scientific runtime.
Disclosure principle¶
AI assistance does not change the evidentiary standard applied to the repository: numerical claims must trace to generated artifacts; model/calibration identity is hash-bound; the locked evaluation was authorization-gated and opened once; and negative/stress results are reported rather than rewritten after the fact.
Generated code, tests, scientific assumptions, citations, licenses, security posture, and claims remain the project author's review responsibility.