Coding and Analysis Workflow
Use AI to change code or analyze data through scoped plans, executable checks, and reversible recovery.
By the end
You will be able to
- Prepare a bounded coding or analysis task with data and environment constraints.
- Execute changes in small inspectable increments.
- Verify behavior with tests, data checks, and direct evidence.
- Recover safely from incorrect code, invalid analysis, and tool or environment failure.
Prepare the task and safe workspace
State the expected behavior or analytical question, relevant files or fields, constraints, acceptance tests, and what must not change. Inspect the actual code, schema, and data before proposing a solution. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Use a branch, copy, sandbox, or read-only data source appropriate to the risk. Remove secrets and unnecessary personal data; never paste production credentials or unrestricted datasets into a prompt. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Outcome/question: [BEHAVIOR OR DECISION]
Inputs: [FILES, SCHEMA, DATA SAMPLE]
Constraints: [SCOPE, PERFORMANCE, PRIVACY]
Do not change: [INVARIANTS]
Checks: [TESTS, QUERIES, RECONCILIATION]
Workspace: [BRANCH, SANDBOX, READ-ONLY]
Rollback: [HOW TO RESTORE]Execute small, inspectable increments
Ask for or create a short plan, then change one coherent behavior at a time. Review diffs, queries, assumptions, and intermediate results before expanding scope. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
For analysis, preserve raw inputs and make transformations explicit. For code, prefer existing project patterns and avoid unrequested dependencies, permissions, or architecture changes. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
{
"contractVersion": "1.0",
"executionBoundary": "bounded-task",
"inputValidation": {
"maxTokens": 4096,
"allowedModalities": ["text", "json"]
},
"outputSchema": {
"type": "object",
"required": ["status", "evidence", "confidence"],
"properties": {
"status": { "type": "string", "enum": ["verified", "rejected"] },
"evidence": { "type": "array", "items": { "type": "string" } },
"confidence": { "type": "number", "minimum": 0.0, "maximum": 1.0 }
}
}
}Verify with the system, not the explanation
Run the relevant formatter, type checks, unit tests, integration tests, and production build. Exercise the changed behavior and important failure paths. A model's explanation is not evidence that code compiles or works. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
For analysis, validate types, missingness, units, row counts, joins, filters, outliers, and reconciliation totals. Review whether the sample represents the population and whether the conclusion exceeds the data. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Recover without destroying evidence
When a check fails, preserve the error and smallest reproduction, identify whether the cause is code, data, environment, or expectation, and correct the narrowest layer. Do not delete tests or weaken assertions merely to obtain green output. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Use version control or a verified backup to reverse unsafe changes. Stop before destructive, external, costly, or permission-expanding actions unless the authorized workflow explicitly allows them. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Review security, privacy, and downstream impact
Inspect authentication, authorization, secrets, input validation, logs, dependencies, data exposure, and failure behavior. Generated code can be syntactically valid while introducing insecure or biased behavior. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Record the task, diff or query, tests, data-quality results, reviewer, deployment boundary, and rollback evidence. Commit only the verified scoped change. In production environments, engineers must account for token utilization patterns, context window pressure, and deterministic execution boundaries. When system constraints or rate limits are approached, explicit retry strategies with exponential backoff and jitter prevent cascading failures across downstream dependencies.
Practice activity
Complete a reversible coding or analysis change
- Choose a low-risk code defect or small dataset question and write the work order with invariants, checks, workspace, data policy, and rollback.
- Implement one coherent change or transformation while preserving a diff or query log.
- Run at least three relevant checks, including one failure or boundary case.
- Inject or identify one failure, preserve its evidence, recover narrowly, and confirm the original checks still pass.
What to produce
- A work order, diff or query, and check output tied to the acceptance criteria.
- A failure-and-recovery record containing reproduction, cause, correction, regression check, and rollback path.
Reflect before continuing
Which result looked correct until you checked the real code, data, or failure path?
Evidence
Sources and verification
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence ProfileNIST · verified 2026-07-25
- Model guidanceOpenAI · verified 2026-07-25
- Prompting best practicesAnthropic · verified 2026-07-25
Knowledge check
Make it stick.
Choose the strongest answer for each question. Your attempts become part of your account transcript.