Public engineering evidence

Case studies, with the status left visible.

A concise record of the problem, turnaround, verification and outcome. Submitted work is not presented as accepted work, and insufficient evidence is not turned into a claim.

Case 01 · TypeScript

Generic propagation in unjs/ofetch

A controlled engineering comparison in which both the baseline and GDN Creator initially produced the same plausible but incorrect abstraction.

Verified controlled result

Problem

The proposed repair passed ordinary repository checking but still failed the frozen acceptance harness with five diagnostics.

Turnaround

A bounded repair was triggered during the same controlled evaluation after the independent verifier rejected the first candidate.

Verification

Frozen controls, verified hashes, repository checks and a preserved upstream commit created a reproducible evidence trail.

Outcome

The repaired GDN candidate passed the frozen requirements. The baseline candidate failed. Classification: GDN win for this single controlled case.

Scope: one controlled case—not a claim of general superiority, production certification or security certification.

Inspect evidence →

Case 02 · Autonomous aircraft

Terrain-based arming regression in ArduPilot

A regression path in genuine open-source autopilot software was converted into a targeted, reviewable test contribution.

Upstream review

Problem

Terrain-based arming behaviour needed a reproducible regression test so the expected safety behaviour could be checked automatically.

Turnaround

The issue was investigated, narrowed and converted into a focused contribution during a short engineering sprint.

Verification

The change was exercised against the relevant test path and submitted with the reproduction logic available for maintainer review.

Outcome

A public pull request was opened. Final acceptance remains with ArduPilot maintainers and is intentionally not claimed here.

Status shown as upstream review until the project records a final disposition.

View public review →

Case 03 · Flight-control communications

Cached MAVLink FTP replies in PX4

A duplicate-reply path was examined at the serialization boundary where a compact cached FTP reply was expanded into a full MAVLink message.

Contribution submitted

Problem

Unused payload bytes could survive cache reuse and enter duplicate-reply serialization instead of being deterministically cleared.

Turnaround

The defect boundary was isolated and a focused repair was prepared without expanding the surrounding protocol surface.

Verification

The compact cache is materialised into a zero-initialised full message, unused bytes are cleared, and regression coverage checks the expansion boundary.

Outcome

A reviewable upstream contribution was opened. Merge, acceptance and release status remain decisions for PX4 maintainers.

Public project reference retained; no claim of PX4 endorsement or deployment is made.

View issue context →

Case 04 · Scientific Python

Raw annotation span conversion in MNE-Python

A bounded API addition converts a raw annotation index into its corresponding minimum and maximum time span.

Contribution submitted

Problem

The Raw interface needed a direct, reviewable conversion from an annotation index to the associated time boundaries.

Turnaround

The API behaviour, implementation and regression tests were assembled as one focused contribution.

Verification

Ten targeted tests passed; one Qt-dependent test was skipped because Qt was unavailable. Ruff checks and formatting checks passed.

Outcome

The change was submitted for upstream review. The skipped dependency and pending maintainer decision remain explicit.

Test results describe the submitted environment, not an upstream acceptance decision.

View issue context →

Disclosure: OC Labs uses AI-assisted engineering tools within the GDN workflow. GDN is model-independent and applies separate verification, evidence and human-approval controls. Public repositories and maintainers referenced above are not clients and do not endorse OC Labs unless explicitly stated.