A method of reading arbitrary symbols is described. Symbols may degrade over time. Shapes of symbols as they degrade are used to update an initial symbol library, which is created based on actual markings, rather than idealized symbols. Marks associate with shapes, which associate to symbols, which in aggregate associate with an object ID. A read mark is compared to all the shapes in the library to determine a most likely shape. A selection set is used to limit shape selection, based on the comparison, to shapes of valid symbols. Comparison methods use probability distributions. Confidence values are used to validate output, generate warnings, and to control updating of the library. Weighted averaging based on confidence values or age of reads may be used at the level of shapes, comparison distributions, or selections.