Every earlier drop
To marble a sheet of paper you float colour on a bath and lay the paper on top. The bath is water thickened with a gum or with boiled seaweed, and the marblers call it size. Each colour is mixed with a little ox gall, which lowers its surface tension enough that a drop spreads into a thin disc instead of sinking. Drop a second colour into the middle of the first and the first becomes a ring. Drop a third and you get two rings. Then you draw a stylus through the floating colour, or a comb with many teeth, and the paper lifts off whatever the surface looks like at that moment. You only ever get one print.
Three people and a book. In 1627 Francis Bacon's Sylva Sylvarum reported that "the Turks have a pretty art of chamoletting papers". Chamolet, or camlet, was a cloth with a watered sheen. His account of the method has the colours oiled. Turkish ebru was, as far as I can tell, made with water colours and gall, so he probably had it at second hand. The craft came into Europe through Germany and France. By the nineteenth century the trade guarded its recipes so closely that Charles Woolnough's The Art of Marbling, as Applied to Book-Edges and Paper (London, 1853) gave the popular patterns in more practical detail than anything before it. The V&A's account of the National Art Library's marbling books says Woolnough called the earlier printed descriptions "utterly ridiculous". He also complained of being pirated in Philadelphia, and of a butcher who sold him watered-down gall. The Wellcome Collection has the 1881 edition online.
Woolnough sent a copy to Michael Faraday, who wrote back from the Royal Institution on 2 January 1854 (letter 2773). Faraday had spent seven years of his youth apprenticed to a bookbinder. He wrote that the subject interested him twice over, because of that early trade and because of the natural philosophy in it. I think he was right about the second, though it was a long time before anyone wrote that part down as mathematics.
One drop, exactly. In 2012 Shufang Lu, Aubrey Jaffer and Xiaogang Jin published Mathematical Marbling. It describes each of the marbler's moves as an exact transformation of the plane. A drop of radius r at C moves every point P outside it to
C + (P − C) · √(1 + r² / |P − C|²)
and the reason is just bookkeeping. Paint that sat between two circles around C is still between two circles afterwards, and both circles have grown by the area of the new drop. Their radii change, but the difference of their areas doesn't. So the drop adds its own area and takes nothing from anything already floating. Every earlier drop is pushed aside and none of them shrinks. A drop with many others dropped inside it ends up as a thin ring a long way out, with exactly the area it started with. Jaffer sets out the argument on his page about dropping paint.
A straight stroke of the stylus is just as tidy. Every point moves along the line of the stroke, by an amount that falls off with its distance from the line. Points at the same distance all move together, so nothing gets squeezed or stretched there either. A comb is a row of these pulled at once. The old pattern names describe the order of the moves. Stone is drops alone. Come-and-go is a stylus taken up and down the bath and then back and forth across it. Nonpareil is that, followed by a fine comb pulled once along its length. Each of those moves takes nothing from the paint and adds nothing. All they do is rearrange it.
Backwards is easier. Every one of these maps can be undone exactly. A drop is undone with a minus sign under the root, and a stroke by moving the same distance back along it. That suggests drawing the sheet backwards. You don't follow the paint forward through every move. Instead, each point of the finished sheet asks where it was before the last move, then the one before that, and so on, until it finds itself inside a drop and takes that drop's colour. If it gets all the way back without meeting one, it is bare size. The present is the input, and the past is computed from it. That is how the piece A bath of size is drawn.
A real bath can't do that, because it carries momentum and doesn't keep the paint in perfect order. Jaffer's later papers on these strokes bring in the actual fluid flow. Still, when the marbler lays the paper down, the print holds the whole history at once. In the idealised version that history can be read back out exactly.