EyeTrustAIDefects4J · Completed method investigation
When the score cannot support the risk target.
A completed development audit asks whether a software-bug scorer has enough signal to justify a protected calibration experiment.
Audit outcome
Insufficient signalNo supported threshold in the fixed development procedure.
Protected calibration
Not runThe development finding stopped progression.
Protected test access
0No locked outcomes were used.
A calibration method cannot manufacture a useful ranking from inadequate evidence. A negative readiness result is a scientific result, not a successful validation claim.
What we tested
The fixed development procedure evaluated whether the scorer and action definition could support a certified threshold. It found none and retained all cases for review. No post-audit tuning or protected-use model training followed.
What we found
| Stage | Recorded result |
|---|---|
| Method engine | Verified |
| Development representation | Provenance verified |
| Scorer signal | Insufficient |
| Protected calibration | Not executed |
| Locked evaluation | Not executed |
Keeping this boundary prevents an implemented method from being presented as a demonstrated automation benefit.
What the evidence supports
No protected calibration certificate was produced. The result is a negative development feasibility audit, explicitly separate from the twelve completed benchmark studies.
The statistical statement, in more detail
Development readiness, mathematical method verification and successful held-out validation are different forms of evidence. Passing an implementation check does not establish a scientifically effective scoring signal.
Observed evaluation error, calibration support and practical output size measure different things. The study’s declared population and reference determine how its result should be read.
Scope of the result
The finding concerns this scorer, representation and action design. It is not a claim that software defect prediction cannot be calibrated. No protected calibration or locked-test source/outcome access occurred.
Sources & data
This summary draws on the completed experiment records. The downloadable table contains the aggregate results shown here. A standalone public reproduction package for this investigation is not linked.
Download the result table (CSV)Numbers are rounded for reading. These are research results, not measured customer outcomes. Dataset examples and individual records are not redistributed on this page.