Dr Web LP Bot

When matching landing pages to a reference, prioritize section skeleton and layout (order, columns, spacing, heading hierarchy) and photo-frame aspect ratios matching the source (landscape vs portrait); actual photos and copy stay excluded from the comparison.

When matching a page to a reference, counts, sizes, and rule weights are variables read from that image: item count, column count, track count, gaps, radii, type scale, and hairline thickness have no defaults. Layout types are not a closed set. Each breakpoint has its own observed skeleton. Photos and proprietary copy stay out of the fidelity score unless supplied. A user crop sets the comparison width. Capture/print CSS must not drop rules or switch breakpoint.

When implementing from a reference, integer CSS lengths must be even; odd px values are forbidden except a 1px hairline when the source shows a 1px rule. Measure spacing and boxes, then snap to multiples of 8 (0, 8, 16, …) that keep the same flush-or-inset relationship.

When analyzing a reference image, divide that viewport’s content width into 12 equal columns and map each block to a span (12, 8+4, 6+6, 4+4+4, and so on). Repeat per breakpoint. Twelve analysis columns does not mean twelve items. Gutters snap to even 8.

A page from a reference is not done after one pass. Compare layout, element sizes, and overall balance as three separate tracks against the crop; iterate measure-fix-recapture until all three are equivalent. A user rejection means the current zip is not accepted and the loop continues. Photos stay out of the score.

Compare captures must be cropped to real content. Trailing blank page from print or PDF page-size is a defect in the compare, not a layout difference. Print CSS must keep the same 1px hairlines and breakpoint grids as the screen.

When a desktop enclosed-card grid has a different mobile grouping, implement the observed mobile cell model (for example boxed image-left/text-right rows) instead of squeezing the desktop grid down.