reconstruct-grains

1. Read the scope resource and identify whether the input is an image or raw EBSD. 2. Preserve source bytes. Ask for explicit crop/calibration when labels or scale bars are present. 3. Use microstructure.segment_image for visible boundaries; use microstructure.segment_ebsd for measured orientations. Never infer orientations from IPF screenshot colors. 4. Compare global, adaptive and ridge methods with explicit bright/dark boundary polarity. Use guided markers for reviewed corrections; preserve pore/unknown masks. Do not claim method agreement or a processed reference mask is independent grain ground truth. 5. Show the candidate map and its unknown/phase/size diagnostics. Request review or correction. 6. Use microstructure.initialize_fields with recorded interface width. Inspect label recovery, diffuse area changes and unresolved grains. Never hide small-grain loss or replace the map with a synthetic tessellation. 7. Simulation requires accepted reconstruction, supported single-phase data and sourced or explicitly hypothetical gamma/mobility. Report dimensionless time when uncalibrated. 8. Use microstructure.trajectory to show the complete sampled evolution: stable grain IDs, diffuse interfaces, time, grain areas, disappearance and energy. Call it a hypothetical scenario when mobility/energy are not calibrated. Unknown/pore masks and mixed phases must not be evolved by silently filling or relabeling them. 9. Distinguish a successful computation, numerical verification and experimental validation.