Project Plan Template (General Placeholders)

Replace each [PLACEHOLDER: …] with your project specifics. Keep the section names and ordering to ensure consistency across projects.

If this plan will be shared with an LLM, sanitize it first. Include schema, methods, validation rules, and synthetic examples only. Do not include individual-level records, PHI, PII, credentials, private file paths, controlled-access data, sensitive small-cell outputs, or protected system details. Keep the code repository separate from protected data and run any real-data workflow only in the approved environment.

For the guide’s BMI exercise, use synthetic participants aged 20 or older at measurement, apply general descriptive categories to unrounded calculated BMI, and flag missing or invalid age before selection. Use demographics for a descriptive audit without inferring clinical cutoffs from broad EHR race categories. See the BMI source and scope notes. Adapt the population and method fields explicitly for other projects.

---

**Data Description**
Primary dataset:
- Source/location: [PLACEHOLDER: approved local path or system label; do not paste private paths into LLM prompts]
- Format: [PLACEHOLDER: CSV/TSV/Parquet/DB table/API]
- Row grain: [PLACEHOLDER: one row per entity/event/transaction]
- Key columns: [PLACEHOLDER: primary keys, business keys]
- Time fields: [PLACEHOLDER: date/time columns and time zone]
- Units/encodings: [PLACEHOLDER: units, categories, code systems]
- Expected size: [PLACEHOLDER: rows, columns]

Secondary dataset(s) (if any):
- Source/location: [PLACEHOLDER]
- Join keys: [PLACEHOLDER]
- Purpose: [PLACEHOLDER: enrich, reference, labels]

Derived variables needed:
- [PLACEHOLDER: normalized fields, computed metrics, flags, labels]

Assumptions/notes:
- [PLACEHOLDER: caveats about data provenance, refresh cadence, access]
- Population and eligibility time point: [PLACEHOLDER: inclusion/exclusion rules and date at which eligibility is evaluated]
- Missing or invalid eligibility fields: [PLACEHOLDER: flagging, exclusion, and reporting policy]

---

**Task to Be Accomplished**

1. **Data Ingestion**
  - Environment/Language: [PLACEHOLDER: R/Python/SQL]
  - Packages/Versions: [PLACEHOLDER: packageA vX.Y, packageB vZ]
  - Version evidence: [PLACEHOLDER: existing manifest/lockfile and verified runtime; mark unknown versions explicitly]
  - Steps:
    - Read inputs from [PLACEHOLDER: paths/endpoints].
    - Enforce schemas and data types; parse dates/times with [PLACEHOLDER: tz].
    - Normalize identifiers (trim, case, unicode) and handle missing values with [PLACEHOLDER: policy].
    - Persist raw snapshots to [PLACEHOLDER: storage/location].

2. **Data Cleaning and Filtering**
  - Validation rules: [PLACEHOLDER: field-level constraints, ranges, regex].
  - Deduplication: [PLACEHOLDER: keys + tie-breakers].
  - Standardization: [PLACEHOLDER: units, categories, code mappings].
  - Outlier/anomaly detection: [PLACEHOLDER: statistical rules/model, thresholds].
  - Business rules/exclusions: [PLACEHOLDER: domain-specific filters].
  - Logging: write rejected rows and reasons to [PLACEHOLDER: path/table].

3. **Representative Record Selection**
  - Entity grain for final output: [PLACEHOLDER: per user/order/product/etc.].
  - Selection window: [PLACEHOLDER: start/end relative to index date/event].
  - Criteria: [PLACEHOLDER: latest/closest/median/score-based].
  - Tie-breakers: [PLACEHOLDER: rule order].
  - Index date definition (if applicable): [PLACEHOLDER].

4. **Categorization of Key Variables**
  - Variable A categories: [PLACEHOLDER: bins/labels/rules].
  - Variable B segments: [PLACEHOLDER].
  - Mapping source and version: [PLACEHOLDER].
  - Calculation and rounding policy: [PLACEHOLDER: formula, units, unrounded comparisons, display precision].
  - Population applicability: [PLACEHOLDER: source-defined population and evidence supporting any mapping; do not infer it from broad demographic labels].
  - Sensitivity variants (optional): [PLACEHOLDER].

---

**Expected Output**

1. **Cleaned Dataset**
  - Format/location: [PLACEHOLDER: CSV/Parquet/DB table at path].
  - Row grain: [PLACEHOLDER].
  - Columns: [PLACEHOLDER: list or pointer to data dictionary].
  - Constraints/quality gates: [PLACEHOLDER: not-null, uniqueness, ranges].

2. **Summary Report**
  - Contents: [PLACEHOLDER: row counts per step, exclusion reasons, distributions, charts].
  - Format/location: [PLACEHOLDER: Markdown/HTML/Dashboard at path].

3. **Data Dictionary**
  - For each column: name, type, description, allowed values, lineage/provenance.
  - Location: [PLACEHOLDER: path/wiki].

4. **Additional Considerations**
  - Reproducible rules (final thresholds/parameters): [PLACEHOLDER].
  - Parameterization/config: [PLACEHOLDER: .env/JSON/YAML].
  - Performance: [PLACEHOLDER: expected runtime, memory, partitioning].
  - Privacy/ethics/compliance: [PLACEHOLDER: PII/PHI handling, approvals, data/code separation, GenAI restrictions].
  - Governance/versioning: [PLACEHOLDER: repo, tags, data versions].

---

**Implementation Checklist (Quick Fill)**
- [ ] Confirm input sources, schemas, and access
- [ ] Specify eligibility at the measurement/event date and missingness handling
- [ ] Define validation rules and unit/category mappings
- [ ] Choose selection window, criteria, and tie-breakers
- [ ] Specify categories/segments and sensitivity variants
- [ ] Set output formats/paths and reporting artifacts
- [ ] Record package versions and configuration
- [ ] Define executable acceptance checks with expected outcomes

---

**Acceptance Criteria and Verification**
- Fixture: [PLACEHOLDER: repository-relative path to synthetic or public input]
- Run command: [PLACEHOLDER: exact command from the repository root]
- Assertions: [PLACEHOLDER: exact output schema, counts, uniqueness, and expected values]
- Boundary cases: [PLACEHOLDER: values at and just below thresholds, date boundaries, missing values, and empty input]
- Determinism check: [PLACEHOLDER: repeated run or shuffled-input command and expected equality]
- Failure case: [PLACEHOLDER: invalid input command, expected exit status, and diagnostic]
- Evidence to retain: [PLACEHOLDER: command results, tested versions, and output location]
- Stop conditions: [PLACEHOLDER: unresolved methodological decisions, unavailable access, or required action beyond authorization]

---

**References (optional)**
- [PLACEHOLDER: Domain standards/specifications]
- [PLACEHOLDER: Internal guidelines/playbooks]
- [PLACEHOLDER: External references/tool docs]