Add immutable item-bank, protocol, model-panel, calibration and release artifacts; strict measurements, script-aware comparisons, response-matrix export, coverage auditing and complete artifact transfer. Include before/after diagrams, candidate foundation source pins, schemas and a separate synthetic end-to-end example. All 190 automated tests and extracted-package workflow checks pass. Real experiment collection and IRT fitting remain external; no real benchmark results are claimed.
23 days ago
Add item-level response comparisons in toolkit 1.2.0
Add paired question reports with complete original and rephrased responses, correctness transitions, recorded scores, source references, and unmatched items. Include searchable offline HTML, CSV and JSONL exports, updated documentation, and a clearly labeled synthetic example. Validated with 138 tests and an extracted-toolkit end-to-end check.
Add the versioned archive descriptor, runnable filesystem toolkit, schema documentation, run template, and pinned references to the existing Belebele and Global MMLU rephrasing inputs. Preserve original JSONL outputs when real runs are added. No measured model results are included in this bootstrap; the toolkit demo is explicitly synthetic.
