Clay tokens found across the Near East from about 8000 BCE recorded quantities of grain, oil, and animals, each shape standing for a commodity (Schmandt-Besserat). Sealed in clay envelopes and later pressed into their surfaces, the tokens preceded cuneiform by several thousand years. The Inca quipu recorded census and tribute in knotted cords, with knot type, position, and color as the encoding; several hundred survive. Marshallese navigators built stick charts from palm ribs and shells to map swell patterns between atolls, then left them ashore and sailed from memory. Each of these systems was read by hand as much as by eye.

The thermodynamic surface of water, described in equations in 1873, was modeled in clay and cast in plaster the following year so that its curvature could be read by hand rather than computed point by point (Gibbs, 1873; Maxwell, 1874). Through the 1930s, relief models of functions with too many variables to plot on paper were built in statistics and engineering laboratories and consulted the way a table of values would be. These models were used as instruments rather than as illustrations.

Reed basketry can carry weather buoy and hurricane records, with each woven element a reading and the weave pattern the timeline; the same Hurricane Sandy data has also been scored as music (Miebach). Economic and demographic series, including the US prison population, have been cut in wood and steel since the 1990s, so that the rise and fall of a number becomes a profile at room scale (Madsen). Dear Data (2014 to 2015) was 104 postcards exchanged between New York and London, each a week of hand-counted personal data drawn in a legend invented for that week (Lupi and Posavec). Participatory installations in which visitors add threads, tokens, or balloons let a dataset accumulate in public, so that the act of answering is also the act of charting (Domestic Data Streamers). It is human-scale craft applied to counting.

inFORM (2013) is a table of 900 motorized pins that renders 3D data, pushes objects around, and reproduces a remote user's hands (Follmer, Leithinger, and Ishii). Relief (2009) and TRANSFORM (2014) pursued the same idea at other scales, and all three run into the same limit: each pin is an actuator, so resolution is limited by the number of motors. EMOTO (2012) milled the sentiment of Twitter messages about the London Olympics into a physical relief (Stefaner, Hemment, and Studio NAND). This work sits between spatial interfaces and embodied computation, in which the material does part of the computing.

The same 3D bar charts, shown on screen and as printed objects, were read more efficiently in physical form, and the advantage shrank when touching the object was forbidden (Jansen, Dragicevic, and Fekete, 2013). Tactile maps and printed models now give blind readers direct access to spatial data, and open tools turn a spreadsheet into a printable model. The evidence suggests that handling a chart changes how it is understood, though the studies so far are small.