airmosaic-ai

Fixed synthetic generation request

English

Apply the workflow prompt and generate generated_regression.py plus the explicit public task config/examples/ml-regression/workflow-synthetic.json for the following bounded task:

Use only the manifest-listed assets/data/03_atmospheric_transport/SyntheticCausal/v1/rows.csv; do not inspect original research data or personal settings.
Predict outcome from x1, x2 and treatment; the name treatment is a synthetic predictor here, not a causal-effect request. Use the released dataset labels and all 400/100/100 train/validation/test rows, with id as the key.
Compare the mean baseline, a depth-5 tree and a 48-tree forest with fixed seed 20261028, one thread and a 180-second execution cap. Select by validation RMSE and freeze before final test evaluation. Request bounded permutation/PDP/ICE diagnostics.
Generate only thin orchestration around ToolRunnerService.run('ml_regression_workflow', ...), using explicit result_scope='workflow'. Default to preflight; execution requires --execute. No project ID, arbitrary command, automatic installation or network access.
Independently check saved prediction keys/truths and metric formulas, model reload, source integrity and the fixture-specific baseline criterion. Keep a new result batch, per-run gen_script code/prompt/config materials, environment and hashes under the workflow’s ignored results tree.
Preserve failures, report sanitized status/counts and promote only reviewed generic example source. Do not claim scientific acceptance, automatic model generality or deterministic regeneration from the prompt.

Expected deliverables: an inspectable program, schema-valid task, negative engineering checks and actual synthetic execution evidence. The agent authors/reviews code in the task session; no extra code-generation service is implied.

Follow the common execution/process-record contract. New runs use a compact YYYYMMDDHHmm timestamp plus a public task name; the internal run ID only correlates receipts. Preserve gen_script/, events.jsonl, paired retrospective documents and overall workflow.json. No new source_snapshot/ is created; verified redundant synthetic batches are cleaned only with explicit authorization. Generation basis is an explicit design summary, not an uncaptured historical generation transcript.

This site is open source. Improve this page.