ResearchBenchmarksQuestion Answering on SQuAD 200-passage (val)Follow50.3QA AccuracySEAL Baseline, 'Rewrite'2.56414.95727.3539.743Jan 20, 2026Evaluation ResultsMethodMethodLinksQA AccuracySEAL Baseline, 'Rewrite'Iteration=0Iteration=02026.0150.3SEAL Baseline, 'Rewrite'Iteration=1Iteration=12026.0146.9SEAL Baseline, 'Rewrite'Iteration=2Iteration=22026.0145.6Learned-Template Self-Editing (With Archive)Iteration=2Iteration=22026.0145.1SEAL Baseline, 'Rewrite'Iteration=3Iteration=32026.0143.6SEAL Baseline, 'Rewrite'Iteration=4Iteration=42026.0140.7Learned-Template Self-Editing (With Archive)Iteration=1Iteration=12026.0139.5Learned-Template Self-Editing (With Archive)Iteration=3Iteration=32026.0137.7Learned-Template Self-Editing (With Archive)Iteration=4Iteration=42026.0136.5Learned-Template Self-Editing (No Archive)Iteration=4Iteration=42026.0128.7Learned-Template Self-Editing (No Archive)Iteration=3Iteration=32026.0128.1Learned-Template Self-Editing (No Archive)Iteration=1Iteration=12026.0127.9Learned-Template Self-Editing (No Archive)Iteration=2Iteration=22026.0127.9SEAL Baseline, 'Implications'Iteration=3Iteration=32026.0127.9SEAL Baseline, 'Implications'Iteration=4Iteration=42026.0127.4Learned-Template Self-Editing (No Archive)Iteration=0Iteration=02026.0126.3SEAL Baseline, 'Implications'Iteration=1Iteration=12026.0126.3SEAL Baseline, 'Implications'Iteration=2Iteration=22026.0123.9SEAL Baseline, 'Implications'Iteration=0Iteration=02026.0120.3Learned-Template Self-Editing (With Archive)Iteration=0Iteration=02026.014.4