# Mechanism position validation · planned protocol

**Status: planned.** No physical mechanism readings or approved parameter revision have been supplied. The blank source sheet is `engineering/data/mechanism-validation.v1.csv`; the matching public download is a template, not a result.

## Coordinate and measurement contract

- Use the revision's CAD XY plane, millimeters, and the same pivot datum `O=(0,0)` as the [ideal model](/notebook/mechanism-explorer/). Positive angle is counterclockwise from +X; the slider's positive travel is +X. Record the angle actually measured, not only a nominal stop label.
- Before collecting readings, identify the physical feature used as the zero reference and how it maps to model X=0. Record that feature in `zero_reference` for every row. `measured_slider_position_mm` is the instrument reading along +X. Add the signed `zero_offset_to_model_mm` to obtain the observed X coordinate in the model frame. Record the offset and how it was established in a dated test record. Do not tune the offset to reduce model error after seeing the results.
- Record the actual revision, specimen, unique trial, date, measurement method, angle, reading, zero information, optional instrument uncertainty, notes, and a relative `evidence_ref` under `engineering/data/evidence/`. Keep original readings and anomalies, including failed or missing attempts, in the test record. Do not create rows for angles that were not measured.
- Generate the ideal position from the referenced revision's shared parameter JSON with the right-hand branch of the solver. The displayed signed error is `observed X − predicted X`; positive means the observed slider lies farther in +X. MAE is the mean absolute error of valid paired rows only. Neither prediction nor error is a raw CSV column.
- The plotted ideal curve may be sampled for display. Measurement points are drawn only for actual rows; no line connects or interpolates missing observations. Preserve distinct revisions and specimen identities in the table.

## Publication gate

Keep `engineering/data/validation-publication.json` at `pending` until real files, a dated review record, reviewed revision parameters, and matching public source downloads exist. A `measured` switch requires the reviewer and review date, the review record source and public copy, exact raw CSV copies, and for the mechanism a revision map with parameter and design-record copies. The build validator checks structure and file correspondence; a reviewer must still establish authenticity, physical datum, instrument setup, uncertainty, and suitability for publication. Do not infer a favorable design decision from a small or incomplete set.