| <!doctype html><html lang='en'><meta charset='utf-8'><meta name='viewport' content='width=device-width,initial-scale=1'> |
| <title>Midway annotation sample</title><style>body{font:17px/1.6 system-ui;max-width:850px;margin:40px auto;padding:0 24px}li{margin:14px 0}small{color:#555}</style> |
| <h1>Midway annotation sample</h1><p>6 selected threads across 3 pairs: bridal sales training, UAT test planning, and SWE-chat code review. Each pair has one single-turn and one multi-turn thread.</p> |
| <p>Read each full thread and relevant expandable tool events. Use a named copy of <a href='annotation_template.csv'>annotation_template.csv</a> for your annotations. <a href='annotation_schema.json'>Annotation schema</a>. No human judgments are prefilled in this snapshot.</p><ul> |
| <li><a href='bridal_threads.html#P01'>P01 · CollabSkill bridal sales training · single</a><br><small>1 human message(s), 19 released events</small></li> |
| <li><a href='bridal_threads.html#P02'>P02 · CollabSkill bridal sales training · multi</a><br><small>3 human message(s), 25 released events</small></li> |
| <li><a href='collabskill_threads.html#P07'>P07 · CollabSkill UAT · single</a><br><small>1 human message(s), 43 released events</small></li> |
| <li><a href='collabskill_threads.html#P08'>P08 · CollabSkill UAT · multi</a><br><small>3 human message(s), 14 released events</small></li> |
| <li><a href='swe_chat_threads.html#SWE-P01'>SWE-P01 · SWE-chat code review · single</a><br><small>1 human message(s), 43 released events</small></li> |
| <li><a href='swe_chat_threads.html#SWE-P02'>SWE-P02 · SWE-chat code review · multi</a><br><small>2 human message(s), 89 released events</small></li> |
| </ul><p>Counts refer to human messages. Each CollabSkill pair shares an assigned task. SWE-chat examples share a theme across different projects and agents. Missing attachments or workspace context may limit assessment. The SWE-chat single-turn release has tool calls without separate tool-result events.</p><p>This exposed midway practice set is for rubric development and review. Keep its sessions and related split groups out of a later blind holdout. Reading views redact common contact and credential patterns. Exact private source records are retained separately in scenarios.jsonl.</p></html> |
|
|
