CODEX: Testing Machine Learning with the Coverage of Data Explorer Tool
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Artificial Intelligence (AI) and the subfield machine learning (ML) deployed in systems with direct impact on human lives necessitates assurance of these systems through rigorous test and evaluation (T&E) methodologies such as those applied to testing software systems. Yet, ML differs from other software in important ways, limiting the unmodified adoption of software T&E approaches. A key difference between software systems that are programmed and ML systems is that ML derives logic by learning from data samples. This data dependence makes T&E of the training data an important part of AI testing. Combinatorial testing (CT) is an approach matured within software testing that has been adapted to measure coverage of the data input space. CT-based metrics and methods facilitate measuring the dimensions of a model’s operating envelope, understanding the input space distance between domains for model transfer, identifying the top information gain data points for fine-tuning a model on a new domain, designing test sets that distinguish a model’s performance on learned domains from its generalization to new domains, and measuring imbalance in training data frequency coverage and determining impact on resulting model performance. The Coverage of Data Explorer (CODEX) tool implements these functionalities for the primary purpose of research experimentation and validation of the methods and metrics. This tool paper describes CODEX’s functionalities and maps them to AI T&E applications as well as gives an overview of the workflow of the tool. CODEX is available in a public open-source repository hosted at https://github.com/vtnsi/codex/.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it