← Review site # Q2/Q3 disagreement review This is a separate copy. It displays only current explicit Q2 OR Q3 label disagreements in BOTH Medly and ScratchMath, using the latest targeted-review coder13 and targeted Opus-copy exports. No Q1 or Q4 rating controls appear. | Dataset | Criterion | Disputed ratings | |---|---|---:| | Medly | Q2 Mathematical correctness | 31 | | Medly | Q3 Task relevance | 9 | | ScratchMath | Q2 Mathematical correctness | 26 | | ScratchMath | Q3 Task relevance | 22 | | Total | Q2/Q3 only | 88 | The union is72 response entries across32 distinct inputs. Medly has36 selected response entries:27 Q2-only,5 Q3-only,4 both. ScratchMath has36:14 Q2-only,10 Q3-only,12 both. A response with only one disagreement shows only that criterion; a response disputed on both shows both. Source questions, images, model responses, response IDs, fixed order and original anonymous response numbers are unchanged. Benchmark/Input/Response menus and all Next/Previous controls contain only selected entries. The selection is frozen from the two named source exports; resolving an entry does not make it disappear during discussion. The current Opus copy contains the targeted reassessment of Medly Q3; Q2 and ScratchMath Q3 comparison labels were preserved from the earlier Opus export. The label “Current Opus copy” identifies the exact supplied comparison file, not a claim that every criterion was freshly rejudged. No additional calls or rescoring were made for this site. The prefilled editable starting point is coder13’s ordinary300-record export. Hidden criteria and other records remain intact for compatibility and are not reassessed by this focused workflow. Import/export retains the exact field set, fingerprint, orderId, all300records and prior history. The download is a coder13 JSON only; no unrelated Opus download is included. Use coder ID13. Drafts use the separate pstutor-q2q3-disagreement-review-20261030-v1 namespace, preserving original study/review workspaces. No old site or source file is modified. Read the actual response and evidence, decide only the displayed criteria, give reasons for Fail/Cannot determine, save, and export a backup. Labels may agree or disagree; agreement is not a target. admin/SELECTION.json records exact selected IDs, criteria and source SHA256 hashes. admin/VALIDATION.json records tests. The site is a standard static folder under annotator/.