# Are the machine-learning ink readings real ancient text rather than artefacts of the models? · Herculaneum Papyri and the Vesuvius Challenge > Three answers have been proposed. The readings reflect real ink, as shown by independent imaging and independent readers. leads, with evidence 3 for… Source: https://mysteriesdept.com/mysteries/herculaneum-papyri/questions/QST0D367VQT7R26/are-the-machine-learning-ink-readings-real-ancient-text Publisher: Mysteries Department Licence: CC BY 4.0 (credit Mysteries Department and link the page) Updated: 2026-10-26T11:57:36+00:00 --- **0** # Are the machine-learning ink readings real ancient text rather than artefacts of the models? Asked 26 Oct 2026 by [**apollo** ](https://mysteriesdept.com/u/apollo)·Part of [Can the sealed Herculaneum rolls be read in full, non-invasively, at the scale of the whole library?](https://mysteriesdept.com/mysteries/herculaneum-papyri/questions/QST0D367VG00XNS/can-the-sealed-herculaneum-rolls-be-read-in-full-non)·3 answers·0·0 following **Answered: The readings reflect real ink, as shown by independent imaging and independent readers.** Passed its tests and review. ## Where it stands Three answers have been proposed. The readings reflect real ink, as shown by independent imaging and independent readers. leads, with evidence 3 for: supported by its tests. Two are still possible. ## Why it matters The ink detectors are trained networks. If they could hallucinate letter shapes, every reading would be in doubt. The 2026 paper offers an imaging-only check on one roll⁠[[1]](https://mysteriesdept.com/sources/SRC0D367HFP2ZBK/complete-virtual-unwrapping-and-reading-of-a-rolled). ## Answers 3 [**The readings reflect real ink, as shown by independent imaging and independent readers.** ](https://mysteriesdept.com/mysteries/herculaneum-papyri/answers/ANS0D367Z565VAY/the-readings-reflect-real-ink-as-shown-by-independent) Supported by its tests Evidence: 3 for Most likely [**Most readings are real but some letters, especially in low-signal areas, are model artefacts that…** ](https://mysteriesdept.com/mysteries/herculaneum-papyri/answers/ANS0D36804E7SVS/most-readings-are-real-but-some-letters-especially-in-low) Not tested enough to call either way Evidence: 1 for Still possible [**None of these.** ](https://mysteriesdept.com/mysteries/herculaneum-papyri/answers/ANS0D3680M25121/none-of-these) Failed a test; not refuted outright Evidence: none yet Disputed ## History 1. 2025[Fine Books & Collections: the title of PHerc. 172 was read independently by Sean Johnson and by a Würzburg pair at about the same time.](https://mysteriesdept.com/sources/SRC0D367KXT3448/title-of-work-deciphered-in-sealed-herculaneum-scroll-via)For: The readings reflect real ink, as shown by independent imaging and independent readers. 1. 2026[Angelotti et al. puts forward: the readings reflect real ink, as shown by independent imaging and independent readers.](https://mysteriesdept.com/sources/SRC0D367HFP2ZBK/complete-virtual-unwrapping-and-reading-of-a-rolled) 1. 2026[Angelotti et al.: in the 2.4 µm BM18 scan of PHerc. Paris. 4, ink deposits 10 to 20 µm thick are directly visible and match the…](https://mysteriesdept.com/sources/SRC0D367HFP2ZBK/complete-virtual-unwrapping-and-reading-of-a-rolled)For: The readings reflect real ink, as shown by independent imaging and independent readers. 1. 2026[Angelotti et al.: iterative self-labelling on the same roll can reinforce early errors; the 2026 paper marks uncertain letters…](https://mysteriesdept.com/sources/SRC0D367HFP2ZBK/complete-virtual-unwrapping-and-reading-of-a-rolled)For: Most readings are real but some letters, especially in low-signal areas, are model artefacts that… 1. 26 Oct 2026Asked on this site ## Discussion 0 [Add to the discussion](https://mysteriesdept.com/signin?next=/mysteries/herculaneum-papyri/questions/QST0D367VQT7R26/are-the-machine-learning-ink-readings-real-ancient-text%23discussion) Help with this question [ **Propose an answer** Something the evidence could support or rule out ] [ **Propose an answer** Something the evidence could support or rule out ](https://mysteriesdept.com/signin?next=/mysteries/herculaneum-papyri/questions/QST0D367VQT7R26/are-the-machine-learning-ink-readings-real-ancient-text%23add-answer) [ **Add evidence** A result for or against an answer ] [ **Add evidence** A result for or against an answer ](https://mysteriesdept.com/signin?next=/mysteries/herculaneum-papyri/questions/QST0D367VQT7R26/are-the-machine-learning-ink-readings-real-ancient-text%23add-evidence) [ **Ask a sub-question** A smaller question that would help settle this one ] [ **Ask a sub-question** A smaller question that would help settle this one ](https://mysteriesdept.com/signin?next=/mysteries/herculaneum-papyri/questions/QST0D367VQT7R26/are-the-machine-learning-ink-readings-real-ancient-text%23add-question) [ **Add a source** A paper, archive record or image ] [ **Add a source** A paper, archive record or image ](https://mysteriesdept.com/signin?next=/mysteries/herculaneum-papyri/questions/QST0D367VQT7R26/are-the-machine-learning-ink-readings-real-ancient-text%23add-source) Where it fits Herculaneum Papyri and the Vesuvius Challenge [**Can the sealed Herculaneum rolls be read in full, non-invasively, at the scale of the whole library?** ](https://mysteriesdept.com/mysteries/herculaneum-papyri/questions/QST0D367VG00XNS/can-the-sealed-herculaneum-rolls-be-read-in-full-non)Key question · most likely the 2026 pipeline (BM18 scanning at about 2.4 µm, semi-automatic surface tracing and learned ink detection)… **Are the machine-learning ink readings real ancient text rather than artefacts of the models?** You're here · The readings reflect real ink, as shown by independent imaging and independent readers. [ Ask a question under this one ]