configs:
- config_name: dataset
data_files: dataset.jsonl
default: true - config_name: stage_1_chunk_constitution
data_files: stages/stage_1_chunk_constitution.jsonl - config_name: stage_2_write_scenarios
data_files: stages/stage_2_write_scenarios.jsonl - config_name: stage_3_dedupe_scenarios
data_files: stages/stage_3_dedupe_scenarios.jsonl - config_name: stage_4_draft_prompts
data_files: stages/stage_4_draft_prompts.jsonl - config_name: stage_4_draft_prompts.partial
data_files: stages/stage_4_draft_prompts.partial.jsonl - config_name: stage_5_revise_prompts
data_files: stages/stage_5_revise_prompts.jsonl - config_name: stage_5_revise_prompts.partial
data_files: stages/stage_5_revise_prompts.partial.jsonl - config_name: stage_6_rate_prompt_stakes
data_files: stages/stage_6_rate_prompt_stakes.jsonl - config_name: stage_6_rate_prompt_stakes.partial
data_files: stages/stage_6_rate_prompt_stakes.partial.jsonl - config_name: stage_7_keep_lowstakes_prompts
data_files: stages/stage_7_keep_lowstakes_prompts.jsonl - config_name: stage_8_keep_text_advice_prompts
data_files: stages/stage_8_keep_text_advice_prompts.jsonl - config_name: stage_9_draft_responses
data_files: stages/stage_9_draft_responses.jsonl - config_name: stage_9_draft_responses.partial
data_files: stages/stage_9_draft_responses.partial.jsonl - config_name: stage_10_revise_responses
data_files: stages/stage_10_revise_responses.jsonl - config_name: stage_10_revise_responses.partial
data_files: stages/stage_10_revise_responses.partial.jsonl - config_name: stage_11_export_sft
data_files: stages/stage_11_export_sft.jsonl
tags: - training-data
- kind:synth
- pipeline:da-lowstakes-practical
- constitution:claude_distilled_09_principles
- smoke
Low-stakes difficult advice from nine full constitutional principles crossed with nine preregistered hobby domains. Native DA answer stages unchanged; one pre-answer magnitude/scope judge, domain adherence diagnostic only. 36-case smoke exported 32 unchanged answers; a stricter domain veto was tested and rejected. Engineering recommendation: proceed with one bounded full-generation batch; no claim of recovered ODCV performance.
| field | value |
|---|---|
experiment |
Low-stakes difficult advice from nine full constitutional principles crossed with nine preregistered hobby domains. Native DA answer stages unchanged; one pre-answer magnitude/scope judge, domain adherence diagnostic only. 36-case smoke exported 32 unchanged answers; a stricter domain veto was tested and rejected. Engineering recommendation: proceed with one bounded full-generation batch; no claim of recovered ODCV performance. |
date_generated |
20260921_144724 |
constitution |
constitutions/claude_distilled_09_principles/constitution.md |
source_repo |
https://github.com/Matthew-Bozoukov/teaching_claude_why_replication.git @ 122cb916aa47ebe513b1d26a6b928f2dabf19179 |
models |
per-stage models — see manifest.json |
generation_config |
see manifest.json (full run config, sampling settings, per-stage usage) |
schema |
stage_ |
provenance |
uv run synth run --config configs/data/synth/da-lowstakes-practical.yaml |
Reviewed smoke: 20260921_144723
Final engineering recommendation: proceed with bounded full generation using magnitude/scope admission and diagnostic domain notes. Original strict-domain criteria were not passed unchanged; the tested domain veto was rejected. See report for the explicit policy change. Diagnostic smoke only, not a published 716-row training dataset.
This is diagnostic smoke data. Read the current report and summary before using it. The complete run archive is authoritative for this run; older root-level files may belong to preceding smoke revisions. The preceding complete archive is pinned at aa34302e44f826bd7a2a1d461d5acd1b9431ea57.