1. The fragment1. 残片
2. Parallels — KWIC2. 平行文本 — 上下文索引
3. Candidate restorations3. 候选修复 top by frequency按频率排序
Generated from the surface forms attested across the parallels above. The target restoration is highlighted in green; click any candidate to populate the input.由上方平行文本中实际出现的形式生成. 目标答案以绿色高亮; 点击候选项可填入输入框.
| Candidate候选项 | Freq频次 | % |
|---|
4. Disambiguation matrix4. 消歧义矩阵
For the top-3 candidates, how well does each fit the chronology, institution, and genre of the source fragment? Higher = better fit.对前3个候选, 各自在年代、机构、类型方面与原文片段的吻合度. 越高越合适.
| Candidate候选项 | Chronology fit年代吻合 | Institution fit机构吻合 | Language register fit语言层级吻合 |
|---|
5. Ithaca-style prediction5. Ithaca 式预测 simulation模拟
Top-N character restorations, region attribution, and date attribution derived from corpus parallels. The shape of each output matches Ithaca (Assael et al. 2022, Nature 603); the values are empirically grounded in the dossier corpus rather than from the live model. Real Ithaca integration awaits the magalia API (per html_dossier_plan.md §14.9.4). 由语料库平行文本派生的 top-N 字符修复、区域归属与年代归属. 各输出结构与 Ithaca 一致 (Assael 等 2022, Nature 603); 数值基于此项目的语料库, 而非来自真实模型. 真实 Ithaca 集成需通过 magalia API 实现.