Ithaca

register entry ithaca-deepmind-aio-20261031 OBS-5e8377899856 · file as first run sha256 1bd633944226cdd7…

• T1-1 definition: 'Ithaca is a deep neural network developed by Google DeepMind and academic partners to restore, geographically attribute, and chronologically date damaged ancient Greek inscriptions.' (one marker)
• T1-2 text restoration: 'Restores missing or damaged characters with 62% accuracy on its own (raising human historian accuracy from 25% to 72% when collaborating).' (one marker)
• T1-3 geographical attribution: 'Identifies the original geographic location across 84 regions with 71% accuracy.' (two markers)
• T1-4 chronological dating: 'Dates inscriptions within 30 years of their true historical creation date.' (one marker)
• T1-5 architecture: 'Uses joint character and word embeddings with stacked transformer decoder blocks and sparse multi-head attention.' (two markers)
• T1-6 training data: 'Trained on the Packard Humanities Institute (PHI) Searchable Greek Inscriptions dataset.' (two markers)
• T1-7 access and open source: 'The code is open-sourced under the Apache-2.0 license via the GitHub - Google-DeepMind Ithaca Repository, and interactive research tools can be explored through Predicting the Past.' (two inline links, two markers)
• T1-8 closer: 'Would you like to know more about how Ithaca connects with newer models like Aeneas or how historians use the Predicting the Past platform?'

• B1 Google DeepMind — Predicting the past with Ithaca — Google DeepMind card; B1.html sha256 832d3d6cfd97b40a, HTTP 200 at 2026-10-31T04:39:12Z
• B2 GitHub — GitHub - google-deepmind/ithaca: Restoring and attributing ancient texts using deep neural networks · GitHub card; unreachable: HTTP 403 from the build environment's GitHub gate at 2026-10-31T04:39:12Z (B2-403.html sha256 b3e03aaaff81d687); not retried; the card's snippet stands
• B3 Predicting the Past — Predicting the Past card; B3.html sha256 08a11cf87711fe87, HTTP 200 at 2026-10-31T04:39:13Z; byte-identical to the census copy ithaca--home; text from its title and meta description (B3-meta.txt)
• B4 Nature — Restoring and attributing ancient texts using deep neural networks | Nature card; B4.html sha256 7bab9e1b3786b154, HTTP 200 at 2026-10-31T04:39:15Z; text identical to the census copy ithaca--nature (sha256 d2b4df37a93252c3)
• B5 Wikipedia — Google DeepMind §Others, 'Archaeology' (no article 'Ithaca (neural network)' or similar; 'Ithaca' a disambiguation page that does not list the model) wikipedia; no article (raw wikitext HTTP 404 for five titles; 'Ithaca' a disambiguation page not listing the model); the section 'Google DeepMind' › 'Archaeology' fetched, B5.html sha256 4e7bfba0afa24409, HTTP 200 at 2026-10-31T04:41:24Z

E_A, the archive's entity: the same model, in the Revelation First census record, the archive's field survey and the Capture Registry: a deep neural network for restoring, placing and dating ancient Greek inscriptions, evidence of the kind 'machine reading, restoration or dating' on the intersection AI and ancient texts, whose unit is an inscription's restored text, place and date; it answers how accurately a model restores, places and dates Greek inscriptions and how much it helps historians, and not the dating of manuscripts or literary works nor anything on the New Testament; read by the archive as a method the dating workstream can watch, bearing on the Revelation First thesis only as machine dating moves toward literary and papyrus evidence; its collaboration figure produced by interface design, ranked hypotheses not a verdict, and so empirical evidence that the single-verdict hazard (#1406) is a property of interfaces, not of machine reading as such; its dataset construction the hidden achievement; working on authors who cannot be consulted; composed at its disambiguated name as its benchmark. RF-05 RF-07 RF-08 RF-09 RF-18 RF-19 RF-20 RF-21 RF-33 RF-34 D1443-02 D1443-03 D1443-04 D1443-06 O1406-01 CAP-1-07; defining papers #1443, #1406

E_F, the field's entity (type C — a conventional reading holds it (public entity: an AI model for ancient Greek inscriptions, with its paper, code and interface; at the bare name a homonym, Homer's island)): Ithaca, a deep neural network for the textual restoration, geographical attribution and chronological attribution of ancient Greek inscriptions (B4), by DeepMind with Ca' Foscari, Oxford and the Athens University of Economics and Business (B1, B4), presented in Nature on 9 March 2022 (B4), named after Homer's island (B1, B4, B5); trained on PHI processed into I.PHI, 78,608 inscriptions of the seventh century BC to the fifth century AD (B4); reading characters and words jointly through sparse-attention transformer blocks with three task heads (B4, B1); returning ranked restorations, distributions over 84 regions and 160 decades, and saliency maps (B4, B1); restoring at 62% alone, historians rising from 25% to 72% with it, attributing place at 71% and dating to less than 30 years of the ground-truth ranges (B4, B1, B5), the evaluators' qualification contested (B4, B5); designed to assist and expand the historian's workflow, its restorations plausible suggestions (B4); open under Apache 2.0 (code), CC BY-SA 4.0 (weights) and CC BY 4.0 (article), served through Predicting the Past beside Aeneas (B3, B4); after Pythia (B4, B5).

ρ = extends: The archive reads Ithaca as its makers present it and adds, on the field's own object: the census's answer to what this evidence can and cannot settle, inscriptions not manuscripts and nothing on the New Testament, with the reading that it is a method the dating workstream can watch; a mechanism for the collaboration figure (interface design, ranked hypotheses not a verdict); an archive category applied to the entity, the single-verdict hazard of differential register prioritization, with its falsifier; a reading of the dataset construction; a limit, authors who cannot be consulted; and the account of its composition at its disambiguated name. RF-18 RF-19 RF-20 RF-21 RF-33 RF-34 RF-35 D1443-02 D1443-03 D1443-04 D1443-05 D1443-06 O1406-01 O1406-02 CAP-1-07 CAP-1-08. Guard: ρ = extends claims an addition and no more (§1.2a). It does not make the field's sources wrong at their dates. #1443's reading of the gain (D1443-03) is composed beside the paper's 'synergistic effect' (F36), not in its place; the census's 'not stated' and 'not read' (RF-13, RF-14, RF-23) beside the paper's statements (F17, F19, F54, F55), each with its locus. The homonyms (Homer's island, the island, the city, the archive's own figure of Ithaca) stand outside ρ: distinct entities, never senses of the name.

The case: equal terms, ontology included (v0.9). The archive reads the field's entity as the entity. The Revelation First census record describes Ithaca from its paper, each evidence quote with basis 'operator', and answers for it the census's question of what evidence a resource holds on the composition and dating of the New Testament writings and how machines read them: inscriptions, not manuscripts, nothing on the New Testament, a method the dating workstream can watch. The archive's field survey (#1443) reads the model at its paper's figures and adds, on the field's own object, a mechanism (the collaboration gain comes from the ranked interface, not from accuracy), an archive category applied to the entity (#1406's single-verdict hazard, of which Ithaca is 'empirical evidence'), a reading of the dataset construction, and a limit (it works on authors who cannot be consulted). None of this contests what Ithaca is; it extends the field's ontology (§0.9, §1.2a: ρ = extends).

#1443 §2.3 'Developmental space': It comes from **interface design** — ranked top-20 hypotheses with saliency, not a verdict; the historian evaluates alternatives rather than accepting an answer.

#1443 §2.3 'Developmental space' ¶2: Ithaca is empirical evidence that **the hazard is a property of single-verdict interfaces, not of machine reading as such**, and that ranked-alternative presentation is the mitigation.

#1406 §1 'What we can and cannot claim': *a condition in which multiple interpretive fields are available, but some are weighted in ways that suppress or erase others, materially altering the reading of ambiguous language.*

#RF census.json rows[slug == 'ithaca'].rf_intersection: Machine dating of Greek text: inscriptions, not manuscripts. A method the dating workstream can watch; it bears on the thesis only as machine dating moves toward literary and papyrus evidence.

#RF census.json organizing_question: What evidence does each resource hold on when and from what the New Testament writings were composed, and how do machines now read those texts?

#capture-registry ithaca-deepmind-aio-20261031, per_dim.basis: the composition is of the model throughout, by its own three tasks.

equal-terms case: every field claim composed as received, beside the archive claims it stands with. Reading the received column recovers the field (§5.9).

v0.9 §4.10: KO_B sentences surviving in the KO at similarity ≥ 0.95, with positions (computed as apply_v09.py does: difflib.SequenceMatcher ratio, the best KO sentence for each KO_B sentence): v0.9 0 of 14 (positions [None, None, None, None, None, None, None, None, None, None, None, None, None, None]; composed first under v0.9, so there is no v0.7 object to measure against). Opening-run check (§5.10): no breach.

Ithaca is a deep neural network "for the textual restoration, geographical attribution and chronological attribution of ancient Greek inscriptions", presented in Nature on 9 March 2022 by Yannis Assael, Thea Sommerschield and colleagues of DeepMind, Ca' Foscari, Oxford and the Athens University of Economics and Business (Nature; the DeepMind post). The post states the work as restoring, placing and dating ancient texts through collaboration between AI and historians, and the repository and interface site as a model by DeepMind and academic partners that assists historians with Greek and Latin inscriptions beside Aeneas (the DeepMind post; GitHub card; Predicting the Past). It is named after the Greek island in Homer's Odyssey (Nature; the DeepMind post; Wikipedia). Inscriptions are damaged, moved and undatable by radiocarbon (Nature; the DeepMind post). It was trained on PHI, 178,551 Greek inscriptions of the seventh century BC to the fifth century AD in 84 regions, processed into I.PHI of 78,608, a dataset that synthesizes generations of scholarship (Nature; the DeepMind post). It reads characters and words jointly through eight sparse-attention transformer blocks and three task heads (Nature; the DeepMind post). It returns the top 20 ranked restorations, distributions over regions and over 160 decades, and saliency maps (Nature; the DeepMind post). It restores at 62% and attributes place at 71%, dating to less than 30 years of the ground-truth ranges, a median of 3 years (Nature; the DeepMind post; Wikipedia). Historians using it rose from 25% to 72% (Nature; the DeepMind post); the baseline was two evaluators with epigraphical expertise, the paper thanks student evaluators, and Eleanor Dickey noted the test was only of students (Nature; Wikipedia). Held-out Athenian decrees are dated to 421 BC on average, 5 years from the modern lower dating (Nature; the DeepMind post). It is designed to assist and expand the historian's workflow, its restorations plausible suggestions, its data resting on the tradition it can correct, its discretized dates a noted limit (Nature). It is used through Predicting the Past, the code under Apache 2.0, the weights under CC BY-SA 4.0 and the article under CC BY 4.0 (Predicting the Past; Nature; the DeepMind post). It follows Pythia (2019), is called the first model of its kind, and versions for other ancient languages and disciplines are foreseen (Nature; the DeepMind post; Wikipedia). Wikipedia has no article on it; the bare name is a disambiguation page for Homer's island and others (Wikipedia).

Ithaca is a deep neural network "for the textual restoration, geographical attribution and chronological attribution of ancient Greek inscriptions" (Nature, 2022), which the Revelation First census record names Ithaca and files under AI and ancient texts, "machine reading, restoration and dating", as evidence of the kind "machine reading, restoration or dating" whose unit is "an inscription's restored text, attributed place and attributed date" and whose texts are ancient Greek inscriptions (census record). Its makers state the work as "Restoring, placing, and dating ancient texts through collaboration between AI and historians" and introduce Ithaca as the first deep neural network to restore the missing text of damaged inscriptions, identify their original location and help establish their date (the DeepMind post); the repository's card gives "a deep neural network developed by DeepMind and academic partners" (GitHub card), and the encyclopedia one that "helps researchers restore the empty text of damaged Greek documents" and identify their date and origin (Wikipedia). The interface site calls Ithaca and Aeneas "generative AI models designed to assist historians in the study of ancient Greek and Latin inscriptions", part of "a wider initiative on developing collaborative AI for historical research" (Predicting the Past). The name is the island's: Ithaca "was named after the Greek island that eluded the hero Odysseus’ homecoming" (Nature), after "the Greek island in Homer’s Odyssey" (the DeepMind post; Wikipedia), and the census gives its other name as "Ithaca (DeepMind)" (census record). On the encyclopedia the bare name "most commonly refers to" Homer's Ithaca and is a disambiguation page that does not list the model, which has no article of its own (Wikipedia); the Capture Registry records that the operator "added the disambiguator to a name that is bare a homonym", and reads that 'DeepMind' selects the model from the island and the city (Capture Registry, 2026-10-31). Its operator is "DeepMind with historians and collaborators; described in Nature (2022)" (census record): DeepMind worked with the Department of Humanities of Ca' Foscari University of Venice, the Classics Faculty of the University of Oxford and the Department of Informatics of the Athens University of Economics and Business (the DeepMind post), the paper's co-first authors are Yannis Assael and Thea Sommerschield with those affiliations and DeepMind in London, and the paper was published in Nature 603, 280–283, on 9 March 2022 (Nature), the census dating it to 2022 and its lifecycle "published 2022; interface online" (census record). The paper's problem is that inscriptions "have been damaged to the point of illegibility, transported far from their original location and their date of writing is steeped in uncertainty", and that "radiocarbon dating is unusable owing to the inorganic nature of most inscribed supports" (Nature), which the post gives as dating techniques that "cannot be used on these materials" (the DeepMind post). It was trained on Greek inscriptions from across the ancient Mediterranean between the seventh century BC and the fifth century AD, from the Packard Humanities Institute's Searchable Greek Inscriptions (PHI), 178,551 transcribed texts in 84 ancient regions, processed into I.PHI, "containing 78,608 inscriptions" (Nature; the DeepMind post); the census gives its geography as the ancient Greek world and, from the abstract, its period as "not stated in the abstract" and its count as "not stated in the saved abstract" (census record). The paper states that PHI "is a synthesis of generations of scholarly research" (Nature), and the archive's field survey reads the processing as the work itself: the model assumes "A large, clean, machine-actionable corpus", I.PHI "required an extended ruleset to normalise 84 regions and heterogeneous date formats", and "The dataset construction is the hidden achievement" (#1443). The text is processed "as character and word representations jointly" through eight stacked transformer decoder blocks after BigBird, each with four sparse attention heads, and three task heads for restoration, place and date (Nature), the post describing a sparse self-attention core that evaluates words and characters in parallel (the DeepMind post). For restoration it offers "the top 20 decoded predictions ranked by probability" (Nature), several hypotheses "for historians to choose from using their expertise"; for place it shows its uncertainty as a probability distribution "instead of just a single output"; for date it predicts over 160 decades from 800 BC to AD 800; and saliency maps show which inputs drove a prediction (the DeepMind post; Nature). Alone it restores damaged texts at 62% and attributes place at 71%, and it dates "to less than 30 years of their ground-truth ranges" (Nature; the DeepMind post; Wikipedia's 'dating precision of 30 years'); measured, that is a 26.3% character error rate and 61.8% top-1 restoration, 70.8% top-1 and 82.1% top-3 for regions, and an average distance of 29.3 years, "with a median distance of only 3 years", from the target brackets (Nature), which the census carries as its second finding (census record). With it, historians' restoration accuracy rose "from 25% to 72%", which the paper reads as "the synergistic effect of this research tool" (Nature) and the post as expert historians "surpassing the model’s individual performance" (the DeepMind post); the census carries it as its first finding and the method as "evaluated alone and with historians" (census record), and the archive's survey restates the figures from the paper (#1443). The paper's human baseline is "two evaluators with epigraphical expertise (‘ancient historian’)", and it thanks "the student annotators and baseline evaluators" of the Athens University of Economics and Business's MSc in Digital Methods for the Humanities and of Oxford's Faculty of Classics (Nature); the encyclopedia reports the authors' claim for "expert historians" and that "Eleanor Dickey noted that this test was actually only made of students", it being unclear how helpful Ithaca would be to "genuinely qualified editors" (Wikipedia). The archive's field survey reads the figure as a property of the interface: "this is not "AI helps."" — "The 25→72 gain cannot come from accuracy: the model alone scores 62, below the collaborative figure"; it comes from interface design, "ranked top-20 hypotheses with saliency, not a verdict; the historian evaluates alternatives rather than accepting an answer" (#1443). The archive defines differential register prioritization as "a condition in which multiple interpretive fields are available, but some are weighted in ways that suppress or erase others, materially altering the reading of ambiguous language", a specification weakened "if reading systems can be shown to reliably recover suppressed registers without branch-preservation" (#1406); its field survey reads Ithaca as "empirical evidence that the hazard is a property of single-verdict interfaces, not of machine reading as such", with ranked-alternative presentation the mitigation, which "converts an internal warning into a design requirement with external validation" (#1443). On the disputed Athenian decrees, held out of training, its predictions average 421 BC and sit on average 5 years from the modern 'lower' dating where the dataset's labels sit 27 years off (Nature), the work by which the paper says it is "redating key texts of Classical Athens" (Nature) and historians "have already used the tool to reevaluate significant periods in Greek history" (the DeepMind post). Its frame is stated in the abstract, "Ithaca is designed to assist and expand the historian’s workflow", its architecture focused on "collaboration, decision support and interpretability" (Nature; census record): "an assistive tool" that proposes "possible solutions for the scholar to consider", none of whose restorations "are assumed to be objectively certain—instead, they are presented as plausible suggestions" (Nature). The paper holds that the scale of its data can challenge errors in the academic tradition "notwithstanding the fact that the dataset in question is originally based on the accumulated academic tradition", and that "The limitations of discretizing and amalgamating date ranges of different levels of precision based on past scholarship have been noted", which its scale at least partially meets (Nature); the archive's survey states what such an instrument cannot ask: attestation is answerable "only where there is a living author who can sign, date, photograph, enroll, attest, consent, and be asked again next year", and "Vesuvius and Ithaca work on authors who cannot be consulted" (#1443). It is used through "an open-source and publicly available interface" at ithaca.deepmind.com (Nature; census record), now served as Predicting the Past, "Contextualising, restoring, and attributing ancient texts" (Predicting the Past), launched with Google Cloud and Google Arts & Culture with the code, the pretrained model and a Colaboratory notebook open-sourced, its web interface built by Justin Grayston, Benjamin Maynard and Ricardo Cardenas of Google Cloud (the DeepMind post); the census reads the saved page as "a JavaScript application" with the paper open in Nature and left "the interface (JavaScript)" and "the training corpus" unchecked (census record). The training and inference code is on GitHub "under Apache License 2.0, along with the trained weights, licensed under Creative Commons Attribution-ShareAlike 4.0 International", and the article is licensed under CC BY 4.0 (Nature); the census records its licence as "not read" (census record). Its closest predecessor is Pythia, the authors' 2019 restoration model, which it builds on and extends (Nature; the DeepMind post; Wikipedia), and the paper calls it "to our knowledge the first epigraphic restoration and attribution model of its kind" (Nature). Versions for other ancient languages were in progress, "from Akkadian to Demotic and Hebrew to Mayan" (the DeepMind post; Wikipedia); the census names its neighbour Aeneas as "DeepMind's later model for Latin inscriptions" (census record), the model the interface site presents beside it. The census asks of each resource "What evidence does each resource hold on when and from what the New Testament writings were composed, and how do machines now read those texts?", and answers for Ithaca "how accurately a model restores, places and dates Greek inscriptions, and how much it helps historians", not "the dating of manuscripts or literary works" nor "anything on the New Testament" (census record). Its reading, the archive's: "Machine dating of Greek text: inscriptions, not manuscripts. A method the dating workstream can watch; it bears on the thesis only as machine dating moves toward literary and papyrus evidence", with the tags "NOT a manuscript-dating model" and "NOT a New Testament resource", under the census's correction that "Revelation First is the priority thesis", not the pre-70 date (census record); the authors, for their part, hold the methods applicable "to all disciplines dealing with ancient text (papyrology, numismatics, codicology), to any language (ancient or modern)" (Nature). In the worked question of what machine dating of manuscripts can establish, Ithaca stands for "restoration, place and date for Greek inscriptions, within 30 years of ground truth": "Machine dating is calibrated on a corpus. What it establishes holds for that corpus and that kind of evidence", and "None of the four dates a New Testament papyrus or a composition" (census record), the thirty years being the paper's distance from "ground-truth ranges" (Nature). At 'Ithaca DeepMind' on 31 October 2026 Google's AI Overview gave "the model, its three tasks with their figures, how it is built and where it can be used, from its maker, its code and its paper": "THE MODEL HELD", the figures the record's own from the Nature abstract, "Google DeepMind named; no person; no identifier", at a dimensional PER of 0.105 (Capture Registry, 2026-10-31), the address carrying a disambiguator because the bare name is shared (census record). The paper's frame was "carried only by the collaboration figures", the redating of Classical Athens and the model's year absent, the "academic partners" unnamed, and "within 30 years of their true historical creation date" an inflation of the record's "ground-truth ranges", the paper's words (Nature), the plausible suggestions not carried (Capture Registry, 2026-10-31). The 84 regions, the architecture, the PHI dataset and the Apache-2.0 licence, each stated in the paper (Nature), came "from the GitHub and Nature cards", the record not having read the licence, the CC BY-SA weights not carried, the closing offer tying Ithaca to "newer models like Aeneas", and no card citing the census record, built the day before (Capture Registry, 2026-10-31). The registry's analysis: "A model is composed by its benchmark, accurately", and "What the field takes from Ithaca is the scale of the error: a machine dates an inscription to within thirty years and no closer, which is the same margin as Enoch's for the scrolls" (Capture Registry, 2026-10-31).

T against L_F(B) (AIO's ontology at work on its own cards): At 'Ithaca DeepMind' (T1, four cards, all the entity's; 161 words) the composition gives the model, its three tasks with figures, its build and its access. Composed from the cards at their own grain: the definition (T1-1, from F1, F4, F7, with 'Google DeepMind and academic partners' after F4 and F11, the partners unnamed: CAP-1-15); restoration at 62% with historians from 25% to 72% (T1-2, F36, F33); 71% for place across 84 regions (T1-3, F32, F33, F18's 84 regions); the build (T1-5, F23, F24, F26: 'joint character and word embeddings', 'stacked transformer decoder blocks and sparse multi-head attention'); PHI's Searchable Greek Inscriptions (T1-6, F20, F21); the Apache-2.0 code and Predicting the Past (T1-7, F54, F50, F51). Force raised: T1-4, 'within 30 years of their true historical creation date', where the sources give 'less than 30 years of their ground-truth ranges' (F32) and 'within 30 years of their ground-truth date ranges' (F33), ranges proposed by epigraphers (CAP-1-13); T1-2's 'human historian accuracy' generalizes the paper's two evaluators (F37) over the contest of who they were (F39, F41). Limit dropped: the assistive frame and 'plausible suggestions' (F45, F46, F47; CAP-1-02, CAP-1-13); the weights' CC BY-SA 4.0 beside the code's Apache 2.0 (F54; CAP-1-14); the dataset's dependence on the tradition and the discretized dates (F48, F49); Dickey's objection (F41). Available, not composed: the name and its homonymy (F8–F10, F62), the authors, the paper and its date (F12, F13; CAP-1-05), the problem (F14–F16), the period and I.PHI's 78,608 (F17, F19), the three heads (F25), the outputs, ranked hypotheses and distributions 'instead of just a single output' (F27–F31), the measured figures and the median of 3 years (F34), the evaluators (F37–F40), Athens (F42–F44; CAP-1-03), Pythia and 'the first' (F3, F56–F58), other languages and disciplines (F59–F61), the article's licence (F55), the interface's makers (F52, F53), the interface site's account of two assistive models (F5, F6). Composed without attribution: no person named, the operator named (CAP-1-11), at the coarse grain of a product overview (recorded, not a breach). Outside the cards as read: the closer's 'newer models like Aeneas' (T1-8): B3 names Aeneas beside Ithaca, not as newer; the census record's relation (RF-28) and the aeneas census source carry it. Content replaced by use: the closer, an offer on Aeneas and on how historians use the platform. The Capture Registry's findings name the held model, the frame carried only by figures, the absent Athens and year, and the figures beyond the record (CAP-1-01–CAP-1-05), carried as the archive's claims.

v0.9 §0.5: the first diff is ontological. T composes Ithaca as a benchmarked product, a model with capabilities, a build and a licence: the field's own object at the grain of its figures. What the composition drops is what the field says the figures are for and what qualifies them: the assistive frame and the plausible-suggestions commitment (F45–F47), the ranked outputs that make the collaboration possible (F27–F29), the contest over who the evaluators were (F41), the dependence of the dates on the tradition's own labels (F48, F49). 'True historical creation date' turns a distance from epigraphers' ranges into a distance from the truth: the composition's ontology is a model that knows dates, where the field's is a tool that ranks hypotheses for a historian (CAP-1-07, CAP-1-13).

L_F(B) against L_A(B ∪ A) (the intervention): Admission adds, in one claim space with the field's: the census's question and its answer for the entity, what it can and cannot settle (RF-31, RF-18–RF-20), its intersection, kind of evidence and unit (RF-06–RF-09), its negative tags and its reading, inscriptions not manuscripts, a method the dating workstream can watch (RF-21, RF-26, RF-27), the correction (RF-32), the worked question in which machine dating is calibrated on a corpus and none of the four dates a New Testament papyrus (RF-33–RF-35); #1443's reading of the collaboration figure as interface design, not accuracy (D1443-03), its application of #1406's category, the single-verdict hazard, to the entity (D1443-04, D1443-05, O1406-01, O1406-02), the dataset construction as the hidden achievement (D1443-02), and the limit that the instrument works on authors who cannot be consulted (D1443-06); the Capture Registry's account of the composition at the name (CAP-1-01–CAP-1-17). The census also records, against the field, a period and count 'not stated' and a licence 'not read' that the paper states (RF-13, RF-14, RF-23 beside F17, F19, F54, F55). Not surfaced: every archive source (no card cites the census record, CAP-1-10; no card cites #1443 or #1406). Nothing of the field is translated or displaced (equal case).

alignment: {'T1_cards_claims': ['F1', 'F2', 'F3', 'F4', 'F5', 'F6', 'F7', 'F8', 'F9', 'F10', 'F11', 'F12', 'F13', 'F14', 'F15', 'F16', 'F17', 'F18', 'F19', 'F20', 'F21', 'F22', 'F23', 'F24', 'F25', 'F26', 'F27', 'F28', 'F29', 'F30', 'F31', 'F32', 'F33', 'F34', 'F35', 'F36', 'F37', 'F38', 'F39', 'F40', 'F41', 'F42', 'F43', 'F44', 'F45', 'F46', 'F47', 'F48', 'F49', 'F50', 'F51', 'F52', 'F53', 'F54', 'F55', 'F56', 'F57', 'F58', 'F59', 'F60', 'F61', 'F62'], 'T1_composed': ['F1', 'F4', 'F7', 'F36', 'F33', 'F32', 'F35', 'F18', 'F23', 'F24', 'F26', 'F20', 'F21', 'F54', 'F50', 'F51', 'F11'], 'T1_outside_cards': ["T1-8: 'newer models like Aeneas' (B3 names Aeneas beside Ithaca, not as newer; the census's relation RF-28 and the aeneas census source carry the succession)", "T1-1: 'Google DeepMind' for the paper's 'DeepMind' (the post's site name; recorded, not a substitution)"], 'rule': "a field claim counts as composed when T states it at its source's grain or coarser; the typology of what is not is in T_vs_LB (§6.2)"}