← Review site # Focused disagreement review This separate local site displays only current explicit label disagreements between the second-review coder-13 file and the unchanged original Opus file. | Dataset | Criteria shown | Disputed ratings | Response entries | |---|---|---:|---:| | Medly | Q2 Mathematical correctness | 31 | 31 | | ScratchMath | Q2 Mathematical correctness | 26 | — | | ScratchMath | Q3 Task relevance | 22 | — | | ScratchMath combined | Only disputed Q2 and/or Q3 per response | 48 | 36 | | Total | Selected criteria only | 79 | 67 | There are29 distinct inputs. ScratchMath has14 Q2-only responses,10 Q3-only responses, and12 responses with both disputed. Questions, source images, model response text, IDs, and fixed anonymous response numbers are unchanged. Only selected responses and criteria appear in the workflow. No other model/dataset/criterion is offered in its menus or Next/Previous navigation. The original300-record ordinary JSON is retained for compatibility and seeded as the editable starting point. Hidden labels are preserved; they are not assessed by this focused workflow. Import/export retains the original fingerprint, orderId, fields, records and prior history. Each save still validates all four stored criteria, so an imported full backup must keep its original hidden fields. The site uses a separate browser-storage namespace and does not read or overwrite original study or earlier review drafts. The original Opus and coder13 source files are unchanged. The selection is fixed from current round2 disagreements; subsequent edits do not remove entries, preventing cases from disappearing during discussion. Open annotator/index.html through the task browser or a supported static host. Use coder13; ordinary full-study imports remain compatible. Export through Export backup. No server submissions occur. admin/SELECTION.json contains exact selected IDs, criteria and source SHA256 hashes. No new judgment call or rescore has been made in this focused-site task.