draw-the-source-figure-panel-for-panel

# Every result the source drew, you draw A reproduction is read by putting the source's figure beside yours and asking whether they show the same thing. That is what a reader does with a reproduction, and it is the only form in which "we got the same answer" is checkable. Your study design is not the index anyone reads you under: a result you settled early, or discharged with a headline number, still needs its panel. ## The failure this prevents A run reproduced a packing theory to 110 of 111 published closed-form values — more of the paper reproduced than either comparator managed — and scored 18.7 where a plain agent that did less scored 53.4. Every figure slot in its plan was bound to one of its own preregistered hypotheses, and the paper's *first* result was discharged as the headline number "110 of 111". That result is a lattice-path construction carrying two magic-number series. Both were on disk. Neither was drawn. The word "magic" appears once in the entire report, in a bibliography entry, and the requirement scored 20 against 55 and 48 for the two runs that drew it. The mechanism is not how many panels exist — that run had spare slots, and other runs in the same batch shipped fourteen panels and still lost the comparison. The mechanism is **which result owns a panel**. An agreement count and a CDF are a better summary and a worse exhibit than the panel they replace. A second shape: a scan panel carried two series where the published panel carries three, because ...