Abstract The Excerpta Constantiniana, the scarce remains of the Byzantine Emperor Constantine VII’s (sole r. 945–959) ambitious project to systematize historical works in Greek, has been long understood as a rich repository of fragments from historical texts that no longer survive in their entirety. Since the late sixteenth century, this notion encouraged antiquarian philologists to assemble fragments from otherwise lost works. The concept of reassembling fragments into lost wholes relied on the analogy of solid objects which are fragmented and can be re-unified, implying the uniformity of a single fragment. The case of the dialogue between Polybius and the weeping Scipio Aemilanus is particularly illustrative. It was first excavated as a solid fragment from Appian and became a Polybian fragment for centuries. The discovery of two versions, a legible one from Diodorus and the illegible from Polybius encouraged the reconstruction of the Polybian version based on the alternative versions, especially Appian’s. The new reading of the Polybian version in Vat. gr. 73, facilitated by multispectral photography, demonstrates that such reconstructions were at some points incorrect.