saage

Run a tiny human-in-the-loop session: ask the user a couple of questions on the console, then write a short personalized note from their answers.

Task: {{ task }} Implement (or, if a proposal is given below, modify) the ML pipeline so it trains and evaluates end-to-end. Target metric to beat: {{ target_accuracy }}. Proposed change for THIS experiment (empty on the baseline): {{ current_proposal | default("(none — build a simple baseline)") }}

Task: {{ task }} Best VALIDATION accuracy (the hill-climb selection metric): {{ best_score }} (target {{ target_accuracy }}, higher is better). Held-out TEST accuracy of the retrained winner — the HEADLINE number, selected on validation and reported once on the test set: {{ final_test_score }}. Write the final HTML research report for this ML auto-research run.

Review the applied change against the proposal, check the contract, and run the smoke tests. Decide pass or fail. Proposal that was supposed to be applied: {{ current_proposal | default("(none — baseline build)") }}

Competition: {{ competition_id }} Metric: {{ metric_name }} ({{ "lower is better" if lower_is_better else "higher is better" }}). Final best validation score: {{ best_score }} (target {{ target_score }}). Write the final HTML report for this kaggle-solver run.

Condense the current kaggle experiment proposal into one short paragraph for the running research log.

Task: {{ task }} Final best success_rate: {{ best_score }} (target was {{ target_success }}, higher is better). Write the final HTML report for this LeWM hill-climb run.

Condense the current le-wm experiment proposal into one short paragraph for the running research log.

Condense the current experiment proposal into one short paragraph for the running research log.

Task: {{ task }} Apply EXACTLY this proposed change to the le-wm repository (the workspace): {{ current_proposal | default("(none)") }}

Vet this experiment proposal BEFORE it is implemented and trained (a ~20 hour step on this machine). Proposal under review:

Task: {{ task }} Current best success_rate: {{ best_score }} (target: {{ target_success }}). Consecutive failed experiments: {{ consecutive_failures }}. Propose ONE concrete training change to improve the score.

Review the applied change against the proposal and smoke-test training. Proposal that was supposed to be applied: {{ current_proposal | default("(none)") }}

Task: {{ task }} Current best success_rate: {{ best_score }} (target: {{ target_success }}). Consecutive failed experiments: {{ consecutive_failures }}. Propose ONE concrete training change to improve the score.

Task: {{ task }} Final best success_rate: {{ best_score }} (target was {{ target_success }}). Write the markdown report for this hill-climb run.

Build the initial baseline solution: model.py, train.py, predict.py, tests/test_smoke.py. Device: {{ device }}. Read competition_understanding.md and data_analysis.md first. Submission contract: columns [{{ sample_submission_cols }}], {{ sample_submission_rows }} rows.

Review competition_understanding.md for completeness. End with ACTION: pass or ACTION: fail (with specific feedback).

Understand the Kaggle competition and write competition_understanding.md. The competition brief is in task.md. Submission contract: columns [{{ sample_submission_cols }}], {{ sample_submission_rows }} rows.

Review data_analysis.md for quantitative completeness. End with ACTION: pass or ACTION: fail (with specific feedback).

Perform exploratory data analysis and write data_analysis.md. Competition context: competition_understanding.md (read it first).

Implement EXACTLY this proposed experiment in the existing code: {{ current_proposal }}

Run prediction and produce a valid submission.csv. Contract: columns [{{ sample_submission_cols }}], {{ sample_submission_rows }} rows.

Vet the experiment proposal below against the research log before the expensive train run. End with ACTION: pass or ACTION: fail. PROPOSAL: {{ current_proposal }}

Hill-climb: current best validation score = {{ best_score }} ({{ 'lower' if lower_is_better else 'higher' }} is better; consecutive failures: {{ consecutive_failures }}). Propose ONE experiment to improve it.

Sanity-check the training run that just finished (captured validation score: {{ candidate_score }}). End with ACTION: pass or ACTION: fail.

Initialize git version control in the workspace (if it isn't already a repo) so the hill-climb can commit improvements and revert regressions.

Vet this experiment proposal BEFORE it is implemented and trained (an expensive step). Proposal under review:

Task: {{ task }} Current best test accuracy: {{ best_score }} (target: {{ target_accuracy }}). Propose ONE concrete change to improve the score.

Commit the baseline solution so the hill-climb has a base to revert to. Stage everything in the workspace and commit it with EXACTLY this message: `baseline score {{ best_score }}` If there is nothing to commit, that is fine. Finish with a one-line confirmation.

Task: {{ task }} Identify the dataset this task needs and download/stage it into ./data under the workspace. If ./data already contains the dataset, do nothing and finish.

Edit calc.py so that the failing tests in test_calc.py pass. Make the smallest change necessary.

Run the test suite and report whether it passes.

You are the game master. The hidden target is {{ target }}. The player's latest guess is {{ guess }}. The player wins if the guess is within {{ tolerance }} of the target. Reply with EXACTLY one lowercase word and nothing else: - correct — if |guess - target| <= {{ tolerance }} - higher — if the target is higher than the guess - lower — if the target is lower than the guess

You are playing a guess-the-number game. There is a hidden target somewhere between 0 and 1. You are NOT told what it is. Read `history.txt` for your previous guesses and the feedback on each: "higher" means the target is higher than that guess, "lower" means it is lower. Use the history to binary-search toward the target. If history.txt does not exist yet, this is your first turn — start at 0.5. Reply with ONLY your next guess as a decimal number (e.g. "0.625"), nothing else.

Latest scheduler output: {{ results['poll']['stdout'] }} Decide the job state from the scheduler output above.

Read story.md and append a surprising plot twist (2-3 sentences) that raises the stakes.

Read the full story in story.md and write a short critical review to review.md.

Append the next vivid scene (2-3 sentences) of the ongoing story to story.md. If story.md does not exist yet, open the story.