# Held-out test · Concept > Checking a reading or model on data that was not used to build it. Source: https://mysteriesdept.com/concepts/CPT0DECW06RF522/held-out-test Publisher: Mysteries Department Licence: CC BY 4.0 (credit Mysteries Department and link the page) --- [Concepts](https://mysteriesdept.com/concepts)Held-out test Concept · method · in 7 mysteries # Held-out test Checking a reading or model on data that was not used to build it. [Glozel: the inscribed tablets of the Fradin field](https://mysteriesdept.com/mysteries/glozel)[Voynich Manuscript](https://mysteriesdept.com/mysteries/voynich-manuscript)[Unread oracle bone characters](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered) A key, a model or a translation fitted to one part of the data can always be made to fit that part. The test is whether it works on text, folios or observations kept aside. The site's rule for claimed decipherments and cipher solutions is that they must read held-out material, and runs are marked by whether they did. ## Where it comes up **[Glozel: the inscribed tablets of the Fradin field](https://mysteriesdept.com/mysteries/glozel)** If the sequences behave like a language, Hitz's Celtic reading can be scored on held-out tablets. [Q: Do the Glozel signs form a writing system that records a language?](https://mysteriesdept.com/mysteries/glozel/questions/QST0DEDTPBXAPB0/do-the-glozel-signs-form-a-writing-system-that-records-a) **[Voynich Manuscript](https://mysteriesdept.com/mysteries/voynich-manuscript)** One of the mechanical procedures proposed for the text - independent slot-grammar sampling, Timm-Schinner self-copying, Rugg's table and grille, a Naibbe-type affix-table cipher of Latin (plain or vowel-less), a… [Q: Does any reading survive held-out pages?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST08JGFZ4R3FQM/does-any-reading-survive-held-out-pages)[Q: Which combination of fifteenth-century cipher practices, if any, reproduces a single section's statistics on its held-out pages?](https://mysteriesdept.com/mysteries/voynich-manuscript/questions/QST0DCAYTD6HZV9/which-combination-of-fifteenth-century-cipher-practices-if)[A published mechanical procedure produced it](https://mysteriesdept.com/mysteries/voynich-manuscript/answers/ANS0D8FZZ4C2TD3/a-published-mechanical-procedure-produced-it)[Held-out context scaling (Kneser-Ney orders 1-8)](https://mysteriesdept.com/methods/MTH08JFXA767X5C/held-out-context-scaling-kneser-ney-orders-1-8)[Generator competition on a held-out fingerprint battery](https://mysteriesdept.com/methods/MTH08JFXAYM4J2X/generator-competition-on-a-held-out-fingerprint-battery)[Topic model (LDA) of pages, k by held-out perplexity and fixed k=10](https://mysteriesdept.com/methods/MTH08JFXD4Y38JF/topic-model-lda-of-pages-k-by-held-out-perplexity-and-fixed) **[Unread oracle bone characters](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered)** OBI-Bench finds the best of 23 large multimodal models deciphering at about the level of untrained people, and EVOBC's evolution-chain baseline scores 16.7 percent top-1 on held-out read graphs. [Q: Can computational methods produce readings of unread graphs that palaeographers accept?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQFMCXW4D/can-computational-methods-produce-readings-of-unread-graphs)[OBI-Bench finds the best of 23 large multimodal models deciphering at about the level of untrained people, and EVOBC's…](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTJ2A3HW3/obi-bench-finds-the-best-of-23-large-multimodal-models)[A 2020 deep-learning system ranked the correct join in its top 10 for 98.39 percent of held-out fragment pairs, and…](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTM293ZBW/a-2020-deep-learning-system-ranked-the-correct-join-in-its) **[Etruscan: the readable language that cannot be fully translated](https://mysteriesdept.com/mysteries/etruscan-language)** [Q: How far can the combinatory method and computational analysis extend the roughly 250 securely understood Etruscan words?](https://mysteriesdept.com/mysteries/etruscan-language/questions/QST0DEDYMXF5X5P/how-far-can-the-combinatory-method-and-computational) **[Herculaneum Papyri and the Vesuvius Challenge](https://mysteriesdept.com/mysteries/herculaneum-papyri)** [For PHerc. 1667 a held-out column was used for validation and text emerged in it as training iterated.](https://mysteriesdept.com/mysteries/herculaneum-papyri/evidence/EVD0D367ZWM726M/for-pherc-1667-a-held-out-column-was-used-for-validation-and) **[Unread cuneiform: half a million tablets and the limits of Sumerian](https://mysteriesdept.com/mysteries/unread-cuneiform)** Neural translation of Akkadian from five ORACC corpora reached BLEU 37.47 (transliteration) and 36.52 (cuneiform) on held-out sentences, against 23.51 and 27.09 for a translation-memory baseline. [Q: Can the broken tablets of Babylonian literature be reassembled by automated matching of fragments?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT3VBZYJY/can-the-broken-tablets-of-babylonian-literature-be)[Q: Can machines read the cuneiform tablets that no scholar has read?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has)[Neural translation of Akkadian from five ORACC corpora reached BLEU 37.47 (transliteration) and 36.52 (cuneiform) on…](https://mysteriesdept.com/mysteries/unread-cuneiform/evidence/EVD0DEDXNRE4WCG/neural-translation-of-akkadian-from-five-oracc-corpora)[Both restoration studies (recurrent networks 2020, masked language models 2021) present their output as suggestions…](https://mysteriesdept.com/mysteries/unread-cuneiform/evidence/EVD0DEDY3FKDT9Y/both-restoration-studies-recurrent-networks-2020-masked)[No blind held-out test of a photograph-to-translation pipeline on unpublished tablets has been published, so the…](https://mysteriesdept.com/mysteries/unread-cuneiform/evidence/EVD0DEDY64BCDSS/no-blind-held-out-test-of-a-photograph-to-translation) ## Methods that use it 7 [Held-out context scaling (Kneser-Ney orders 1-8)](https://mysteriesdept.com/methods/MTH08JFXA767X5C/held-out-context-scaling-kneser-ney-orders-1-8) Voynich Manuscript MTH [Generator competition on a held-out fingerprint battery](https://mysteriesdept.com/methods/MTH08JFXAYM4J2X/generator-competition-on-a-held-out-fingerprint-battery) Voynich Manuscript MTH [Topic model (LDA) of pages, k by held-out perplexity and fixed k=10](https://mysteriesdept.com/methods/MTH08JFXD4Y38JF/topic-model-lda-of-pages-k-by-held-out-perplexity-and-fixed) Voynich Manuscript MTH [Lattice EM decipherment: sign boundaries and key learned together, with a structure-preserving null](https://mysteriesdept.com/methods/MTH0DCF04BX248D/lattice-em-decipherment-sign-boundaries-and-key-learned) Voynich Manuscript MTH [Naibbe-structured key search with held-out check](https://mysteriesdept.com/methods/MTH0DC7NVMV088S/naibbe-structured-key-search-with-held-out-check) MTH [Crack attempt on Herbal A under the fitted period-cipher setting](https://mysteriesdept.com/methods/MTH0DCB86QKWF8P/crack-attempt-on-herbal-a-under-the-fitted-period-cipher) MTH [Fifteenth-century cipher practices as one fitted family, per section](https://mysteriesdept.com/methods/MTH0DCB5D15DB6M/fifteenth-century-cipher-practices-as-one-fitted-family-per) MTH ## Cite it ``` Mysteries Department. (2026). Held-out test [Concept CPT0DECW06RF522]. https://mysteriesdept.com/r/CPT0DECW06RF522 ``` Key **CPT** 0DECW06RF522 Record Concept Field Method Related [Decipherment](https://mysteriesdept.com/concepts/CPT0DECW01ZWFP5/decipherment) · [Null model](https://mysteriesdept.com/concepts/CPT0DECW06JED24/null-model) Version 2 Added by [apollo](https://mysteriesdept.com/u/apollo) **How this page is made.** The definition is the site's, in its own words. “Where it comes up” is read from the answers, methods and evidence that name the concept. Related concepts [Decipherment](https://mysteriesdept.com/concepts/CPT0DECW01ZWFP5/decipherment)[Null model](https://mysteriesdept.com/concepts/CPT0DECW06JED24/null-model) The record Every record is open data. Take this one as JSON, with its links and its history. [JSON](https://mysteriesdept.com/api/v1/records/CPT0DECW06RF522)