| name | validation-import-alignment |
| description | Implement or improve reference-data imports, timestamp alignment, validation workflows, error summaries, and experiment tracking for treadmill, IMU, video-derived, pressure, or force-plate comparisons. Use this when working on validation pipelines or import tools. |
Validation Import Alignment
Use this skill when the task involves:
- importing reference CSV files
- timestamp alignment
- validation runs
- session comparison workflows
- error summary reports
- experimental vs validated labeling
- mapping external data to internal sessions
Objective
Make validation workflows reproducible, explicit, and scientifically honest.
Reference systems may include:
- treadmill console exports
- IMU CSV
- video-derived CSV
- pressure insole CSV
- force plate CSV
These data sources differ in:
- timestamps
- sampling frequency
- coordinate systems
- units
- derived metrics
- noise characteristics
Validation logic must preserve those differences instead of hiding them.
Workflow
When implementing validation:
- Import and store raw metadata for the external dataset.
- Link the import to the session and station context.
- Parse timestamps and units explicitly.
- Document any assumptions used for alignment.
- Compute comparable summaries only for metrics that are meaningfully comparable.
- Generate validation reports with errors, caveats, and confidence context.
- Preserve whether the output remains experimental or becomes station-validated / externally validated.
Comparison rules
Good comparisons include:
- cadence estimate vs cadence reference
- step interval estimate vs reference timing metric
- session trend vs session trend
- stage summary vs stage summary
Avoid forcing false precision on comparisons that are not directly comparable.
Required outputs
A validation run should capture:
- sessionId
- reference source type
- import timestamp
- alignment method
- preprocessing notes
- compared metric definitions
- error summaries
- validation outcome
- model version
- caveats
Scientific rules
Do not treat:
- a rough agreement plot
- a single session correlation
- an imported CSV without time alignment
as proof of broad validity.
The workflow must make limitations visible.
Always update
When using this skill, review:
- backend validation entities and services
- docs/validation_workflow.md
- docs/sensing_limitations.md
- report-generation logic
- tests for alignment and import parsing