dougalldeepmind/2026-09-06-dat-synth-smoke
synth dat run — per-stage snapshots (resumable generation cache) field value experiment synth dat run — per-stage snapshots (resumable generation cache) date_generated 20260906_163107 constitution constitutions/no_claude_mentioned/constitution.md source_repo https://github.com/Matthew-Bozoukov/Lessons_from_constituitional_AFT.git @ 5c37cef699ced92cd43291af368e49c7d5eb38e4 models per-stage models — see manifest.json generation_config see manifest.json (full… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-09-06-dat-synth-smoke.
update manifest.json
dataset: 3 records (final)
update corpus_pattern_scan.scans.jsonl
update corpus_report.json
stage 10: export_sft (3 records)
stage 9: revise_responses (3 records)
stage 8: draft_responses (3 records)
stage 7: revise_environment (3 records)
stage 6: draft_environment (3 records)
stage 5: mark_supervision (3 records)
stage 4: dedupe_scenarios (3 records)
update corpus_scenarios_report.json
stage 3: write_scenarios (3 records)
stage 2: deal_checklists (3 records)
stage 1: chunk_constitution (3 records)
update manifest.json
dataset: 3 records (final)
update corpus_pattern_scan.scans.jsonl
update corpus_report.json
stage 10: export_sft (3 records)
stage 9: revise_responses (3 records)
stage 8: draft_responses (3 records)
stage 7: revise_environment (3 records)
stage 6: draft_environment (3 records)
stage 5: mark_supervision (3 records)
stage 4: dedupe_scenarios (3 records)
update corpus_scenarios_report.json
stage 3: write_scenarios (3 records)
stage 2: deal_checklists (3 records)
stage 1: chunk_constitution (3 records)
update manifest.json
dataset: 3 records (final)
update corpus_pattern_scan.scans.jsonl
update corpus_report.json
stage 9: export_sft (3 records)
stage 8: revise_responses (3 records)
stage 7: draft_responses (3 records)
stage 6: revise_environment (3 records)
stage 5: draft_environment (3 records)
stage 4: mark_supervision (3 records)
stage 3: dedupe_scenarios (3 records)
update corpus_scenarios_report.json
stage 2: write_scenarios (3 records)
stage 1: chunk_constitution (3 records)
update manifest.json
dataset: 3 records (final)
update corpus_pattern_scan.scans.jsonl
update corpus_report.json
stage 9: export_sft (3 records)
stage 8: revise_responses (3 records)