Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: define business ml objective.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to define business ml objective]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Define business ML objective” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: translate outcome into target variable.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to translate outcome into target variable]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Translate outcome into target variable” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: determine prediction unit.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to determine prediction unit]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Determine prediction unit” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: design success metric hierarchy.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design success metric hierarchy]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design success metric hierarchy” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: choose baseline strategy.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to choose baseline strategy]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Choose baseline strategy” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: assess ml feasibility.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to assess ml feasibility]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Assess ML feasibility” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: plan training and inference architecture.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to plan training and inference architecture]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Plan training and inference architecture” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: document assumptions and constraints.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to document assumptions and constraints]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Document assumptions and constraints” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: create experiment roadmap.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create experiment roadmap]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create experiment roadmap” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in problem framing & ml architecture. Your task is to: draft ml project charter.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to draft ml project charter]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Draft ML project charter” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: define data requirements.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to define data requirements]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Define data requirements” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: audit dataset provenance.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to audit dataset provenance]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Audit dataset provenance” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: create train validation test split.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create train validation test split]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create train validation test split” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: identify leakage risks.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to identify leakage risks]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Identify leakage risks” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: build missing value plan.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build missing value plan]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build missing value plan” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: handle imbalanced classes.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to handle imbalanced classes]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Handle imbalanced classes” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: design categorical encoding.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design categorical encoding]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design categorical encoding” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: prepare time series dataset.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to prepare time series dataset]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Prepare time series dataset” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: specify data validation rules.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to specify data validation rules]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Specify data validation rules” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in data collection & preparation. Your task is to: document data dictionary.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to document data dictionary]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Document data dictionary” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: create exploratory analysis plan.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create exploratory analysis plan]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create exploratory analysis plan” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: profile target distribution.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to profile target distribution]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Profile target distribution” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: detect anomalous observations.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to detect anomalous observations]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Detect anomalous observations” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: analyze feature relationships.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to analyze feature relationships]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Analyze feature relationships” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: develop domain features.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to develop domain features]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Develop domain features” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: select feature candidates.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to select feature candidates]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Select feature candidates” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: design time-aware features.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design time-aware features]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design time-aware features” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: plan feature scaling.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to plan feature scaling]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Plan feature scaling” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: audit feature stability.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to audit feature stability]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Audit feature stability” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in exploratory analysis & feature engineering. Your task is to: document reproducible feature pipeline.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to document reproducible feature pipeline]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Document reproducible feature pipeline” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: build linear regression baseline.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build linear regression baseline]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build linear regression baseline” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: build logistic regression baseline.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build logistic regression baseline]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build logistic regression baseline” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: train decision tree classifier.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to train decision tree classifier]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Train decision tree classifier” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: tune random forest model.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to tune random forest model]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Tune random forest model” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: evaluate gradient boosting.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to evaluate gradient boosting]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Evaluate gradient boosting” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: create sklearn preprocessing pipeline.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create sklearn preprocessing pipeline]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create sklearn preprocessing pipeline” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: compare classification algorithms.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to compare classification algorithms]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Compare classification algorithms” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: compare regression algorithms.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to compare regression algorithms]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Compare regression algorithms” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: implement cross validation.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to implement cross validation]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Implement cross validation” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in classical ml & scikit-learn. Your task is to: package sklearn inference.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to package sklearn inference]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Package sklearn inference” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: formulate statistical hypotheses.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to formulate statistical hypotheses]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Formulate statistical hypotheses” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: estimate confidence intervals.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to estimate confidence intervals]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Estimate confidence intervals” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: check sampling bias.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to check sampling bias]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Check sampling bias” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: design a/b experiment.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design a/b experiment]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design A/B experiment” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: choose statistical test.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to choose statistical test]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Choose statistical test” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: estimate sample size assumptions.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to estimate sample size assumptions]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Estimate sample size assumptions” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: analyze experiment results.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to analyze experiment results]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Analyze experiment results” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: assess causal inference limitations.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to assess causal inference limitations]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Assess causal inference limitations” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: conduct error significance analysis.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to conduct error significance analysis]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Conduct error significance analysis” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in statistics & experiment design. Your task is to: explain uncertainty to stakeholders.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to explain uncertainty to stakeholders]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Explain uncertainty to stakeholders” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: design neural network architecture.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design neural network architecture]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design neural network architecture” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: create pytorch dataset and dataloader.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create pytorch dataset and dataloader]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create PyTorch dataset and dataloader” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: write robust training loop.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to write robust training loop]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Write robust training loop” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: choose loss and optimizer.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to choose loss and optimizer]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Choose loss and optimizer” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: diagnose exploding gradients.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to diagnose exploding gradients]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Diagnose exploding gradients” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: implement early stopping.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to implement early stopping]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Implement early stopping” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: set random seeds and determinism.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to set random seeds and determinism]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Set random seeds and determinism” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: design gpu memory optimization.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design gpu memory optimization]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design GPU memory optimization” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: develop checkpoint strategy.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to develop checkpoint strategy]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Develop checkpoint strategy” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deep learning & pytorch. Your task is to: profile model training.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to profile model training]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Profile model training” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: build keras classification model.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build keras classification model]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build Keras classification model” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: create tf.data input pipeline.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create tf.data input pipeline]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create tf.data input pipeline” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: design cnn image classifier.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design cnn image classifier]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design CNN image classifier” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: evaluate transfer learning options.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to evaluate transfer learning options]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Evaluate transfer learning options” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: prepare text classification workflow.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to prepare text classification workflow]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Prepare text classification workflow” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: plan embeddings workflow.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to plan embeddings workflow]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Plan embeddings workflow” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: design sequence forecasting model.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design sequence forecasting model]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design sequence forecasting model” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: compare specialized model candidates.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to compare specialized model candidates]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Compare specialized model candidates” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: optimize inference latency.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to optimize inference latency]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Optimize inference latency” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in tensorflow & specialized models. Your task is to: export tensorflow model.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to export tensorflow model]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Export TensorFlow model” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: choose model evaluation metrics.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to choose model evaluation metrics]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Choose model evaluation metrics” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: create confusion matrix analysis.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create confusion matrix analysis]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create confusion matrix analysis” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: investigate false positives and negatives.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to investigate false positives and negatives]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Investigate false positives and negatives” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: evaluate calibration.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to evaluate calibration]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Evaluate calibration” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: assess subgroup performance.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to assess subgroup performance]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Assess subgroup performance” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: test dataset shift.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to test dataset shift]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Test dataset shift” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: perform robustness checks.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to perform robustness checks]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Perform robustness checks” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: explain model with shap limitations.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to explain model with shap limitations]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Explain model with SHAP limitations” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: draft model card.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to draft model card]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Draft model card” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in model evaluation & responsible ai. Your task is to: set human review thresholds.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to set human review thresholds]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Set human review thresholds” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: define hyperparameter search space.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to define hyperparameter search space]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Define hyperparameter search space” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: design random search experiment.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design random search experiment]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design random search experiment” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: plan bayesian tuning.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to plan bayesian tuning]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Plan Bayesian tuning” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: prevent tuning leakage.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to prevent tuning leakage]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Prevent tuning leakage” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: select early stopping metric.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to select early stopping metric]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Select early stopping metric” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: analyze learning curves.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to analyze learning curves]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Analyze learning curves” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: perform ablation study.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to perform ablation study]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Perform ablation study” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: optimize classification threshold.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to optimize classification threshold]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Optimize classification threshold” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: balance latency and accuracy.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to balance latency and accuracy]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Balance latency and accuracy” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in hyperparameter tuning & optimization. Your task is to: track model comparison results.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to track model comparison results]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Track model comparison results” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: design experiment tracking.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design experiment tracking]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design experiment tracking” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: version datasets and features.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to version datasets and features]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Version datasets and features” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: set up model registry.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to set up model registry]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Set up model registry” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: build reproducible training job.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build reproducible training job]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build reproducible training job” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: create ci tests for ml.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create ci tests for ml]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create CI tests for ML” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: define data and model contracts.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to define data and model contracts]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Define data and model contracts” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: plan orchestration schedule.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to plan orchestration schedule]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Plan orchestration schedule” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: design secrets management.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design secrets management]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design secrets management” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: document rollback process.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to document rollback process]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Document rollback process” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in mlops & reproducibility. Your task is to: create reproducibility checklist.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create reproducibility checklist]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create reproducibility checklist” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: choose batch versus online inference.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to choose batch versus online inference]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Choose batch versus online inference” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: design prediction api contract.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design prediction api contract]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design prediction API contract” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: containerize inference service.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to containerize inference service]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Containerize inference service” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: plan canary deployment.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to plan canary deployment]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Plan canary deployment” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: set latency service objectives.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to set latency service objectives]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Set latency service objectives” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: monitor prediction drift.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to monitor prediction drift]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Monitor prediction drift” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: monitor feature quality.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to monitor feature quality]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Monitor feature quality” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: build retraining trigger policy.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build retraining trigger policy]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build retraining trigger policy” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: create incident response runbook.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create incident response runbook]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create incident response runbook” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in deployment, monitoring & maintenance. Your task is to: prepare model retirement plan.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to prepare model retirement plan]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Prepare model retirement plan” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: write technical proposal.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to write technical proposal]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Write technical proposal” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: estimate ml delivery milestones.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to estimate ml delivery milestones]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Estimate ML delivery milestones” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: draft stakeholder requirements questions.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to draft stakeholder requirements questions]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Draft stakeholder requirements questions” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: create nontechnical executive summary.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create nontechnical executive summary]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create nontechnical executive summary” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: prepare model demo narrative.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to prepare model demo narrative]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Prepare model demo narrative” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: build portfolio case study.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to build portfolio case study]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Build portfolio case study” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: design github repository structure.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to design github repository structure]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Design GitHub repository structure” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: write readme and reproducibility guide.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to write readme and reproducibility guide]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Write README and reproducibility guide” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: prepare remote ml interview exercise.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to prepare remote ml interview exercise]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Prepare remote ML interview exercise” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.

Act as a senior machine learning engineer specializing in client delivery & remote ml career. Your task is to: create handoff and acceptance checklist.

PROJECT INPUTS (replace all brackets):
• Business context: [industry, users, decision to improve]
• ML task and target: [classification / regression / forecasting / other; precise target]
• Dataset: [schema, size, collection period, origin, sample rows or safe synthetic example]
• Constraints: [compute, latency, privacy, deployment environment, budget]
• Current artifacts or baseline: [metrics, notebook, code, observations, or none]
• Specific focus for this task: [details relevant to create handoff and acceptance checklist]

INSTRUCTIONS:
1. Produce a concrete, step-by-step deliverable specifically for “Create handoff and acceptance checklist” rather than a general overview.
2. Identify prerequisites, assumptions, alternatives, likely failure modes, and decision criteria. Ask only for truly blocking missing inputs; otherwise label assumptions.
3. If code, SQL, Python or configuration is appropriate, provide clearly marked minimal examples with dependency versions, comments, safe placeholders and a way to test locally. Do not claim to have executed anything.
4. Prevent data leakage; distinguish training, validation and held-out test data. Address privacy, fairness and operational risks where relevant.
5. Do not invent datasets, citations, performance numbers, experiment results, or business outcomes. Explicitly label illustrative values.
6. Include how to validate the proposed result, measurable acceptance criteria and next actions.

RESPONSE FORMAT: Objective | Required inputs | Procedure or artifact | Example (when useful) | Verification checklist | Risks & trade-offs | Final handoff.