The traditional Muslim account is specific. Revelations were memorised and written on whatever came to hand during Muhammad's lifetime; after many memorisers were killed at the battle of Yamama, the caliph Abu Bakr had Zayd ibn Thabit gather the material; and under the third caliph Uthman, around 650 CE, a standard consonantal text was fixed, copies were sent to the main garrison cities, and divergent copies were ordered destroyed.
Early Arabic script wrote only a skeleton, the rasm, without short vowels and often without the dots distinguishing consonants, so a single written text still supported several oral renderings. These are the qiraat. Ibn Mujahid, at Baghdad in the tenth century, fixed seven as canonical; Ibn al-Jazari, in the fifteenth, extended the accepted set to ten, each transmitted through two named narrators. In practice one dominates: the reading of Hafs an Asim, standardised in the Cairo edition printed under King Fuad in 1924 after seventeen years of work by scholars of al-Azhar, is now the text of most printed Qurans worldwide. Warsh an Nafi remains in use across North and West Africa.
Two manuscripts anchor the early evidence. The Sanaa palimpsest, found in the Great Mosque in 1972, carries a washed-off lower text that differs from the Uthmanic text in word order and wording, the only substantial surviving witness to a non-Uthmanic tradition. The Birmingham folios, radiocarbon dated to 568-645 CE at 95.4 per cent confidence, are often reported as proving a date within Muhammad's lifetime. They do not. Radiocarbon dates the death of the animal that supplied the parchment, not the writing on it, and the range is wide; the folios agree with the standard text, and the honest conclusion is that they are very early, not that they are datable to a particular decade.