Spaces:
Running
Running
| # Data and evaluation provenance | |
| **Original source:** Academic Rhetorical Intent Mini v1.1 from the existing research portfolio; 1,026 synthetic-curated examples in English, Spanish and Portuguese. Source splits and four baseline engines were recovered from ModelOps Research Lab v2.0 and ModelGuard Observatory. The training split has 842 rows, the controlled IID split 130, and novel-pattern challenge 54. Class labels: background, objective, method, result, limitation, implication. | |
| The included `source_modelops_engine.js` is a baseline algorithm implementation, not a separately hosted model or the former Python classifier. The evaluation outputs were recomputed with `node generate_artifacts.js`, not typed by hand. Every row links to original ID, source text, transformation and scores. SHA-256 records of source and output data are stored in `benchmark_manifest.json`. | |
| **Research copyright:** Original synthetic-curated dataset and derivatives distributed under CC BY 4.0 with credit; application source under MIT. The source dataset is not text scraped from scholarly publications, students or private records. | |
| **Caveat:** Academic-style source texts are artificially authored. No external real-world validation, demographic fairness, annotation-interrater agreement, causal claims or production monitoring is implied. | |