steveandClaude Opus 4.8
782f2394b2
Three compounding defects in the measurement layer (#47), measured against the live Home garden rather than estimated. Entry was quantized to whole centimeters even though every measurement column is REAL and the server only range-checks. A 12in grid stored 30cm — an 11.811in grid, short by 0.189in per cell, and because the error is systematic it accumulates: 4.61in of drift across a 24ft garden. cmFromFtIn / cmFromMeters / cmFromSpacing now convert exactly, rounding only at 1e-6 cm to clear IEEE-754 dust (12 x 2.54 is 30.479999999999997 in binary). Display-side rounding is untouched, so cm quantization still can't show through in a field. The garden grid was entered at spacing scale: "Grid size (in)" sat directly under "Width (ft)", two adjacent numeric fields at two different scales distinguished only by a suffix. Home ended up on a 3cm grid, almost certainly "1" typed meaning one foot. The garden grid is a layout concern, so it moves to dimension scale and reads "Garden grid (ft)"; the bed grid in the object inspector is a plant-spacing concern and deliberately stays at cm/in. A sub-10cm garden grid now draws a soft warning rather than being silently accepted. Snap could stay active while the grid stopped drawing: the canvas faded the grid out below 6px per cell, so Home ran with snapToGrid on, no grid drawn, and no snapping you could feel. Replaced the fade with coarsening — visibleGridStepCm picks the smallest multiple of the true grid (2x, 5x, 10x...) whose cells are legible, so every line drawn is still a real grid line and something aligned to the grid is always visible. When the drawn step isn't the true grid the canvas says so; a degenerate grid it can't draw at all says that instead of nothing. Closes #47 Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ