Autonomous Software Engineering Under Bounded Authority: A ZERO-DAY-DREAD Case Study of Independent Review, Reproducibility, and Evidence-Preserving Recovery

This technical report presents QSOL-IBAE-ZDD-001, a single-project observational study of AI-assisted autonomous software engineering under human-controlled integration and publication authority. The experiment examines development of ZERO-DAY-DREAD, a deterministic, offline-capable cybersecurity satire application implemented using HTML, CSS and vanilla JavaScript. The research separates a retrospective baseline (PR #1) from a prospectively registered engineering phase (PR #2). The first phase documented six Sourcery findings and a subsequent Codex identity-collision finding. The second phase documented two additional Sourcery findings, corrective implementations and independent Codex reviews across successive revisions. The engineering process included deterministic advisory generation, YAML 1.2 interoperability checks, Unicode normalization, evidence identity verification, cross-platform testing, browser validation and GitHub Pages deployment-order analysis. Final validation included 24 automated Node tests, 1,120 YAML round-trip checks, a bounded 56,000-record identity corpus and successful GitHub Actions validation across Linux, macOS, Windows, Chromium, Firefox and WebKit. The final main-branch workflow completed 12 successful jobs, including deployment. The study additionally documents unsuccessful and interrupted workflow executions, subsequent corrections, evidence recovery after session interruption, and the boundaries separating automated verification, independent review and human authorization. The accompanying research archive contains the manuscript, source snapshots, selected GitHub evidence, original supporting attachments, derived event records, reproducibility instructions and SHA-256 manifests. Findings support the feasibility of an inspectable, human-governed autonomous engineering workflow in this case. They do not establish universal software correctness, model superiority, machine-checked formal verification or execution within the native QSOL-IBAE runtime.