Vesuvius Challenge

register entry vesuvius-challenge-aio-20261031 OBS-c89c95b2cf7e · file as first run sha256 42c0a1778a0275e6…

• T1-1 definition: 'The Vesuvius Challenge is a global open science competition that uses artificial intelligence and 3D imaging to read carbonized ancient papyrus scrolls from the Villa of the Papyri in Herculaneum.'
• T1-2 the eruption: 'When Mount Vesuvius erupted in 79 CE, it buried a Roman villa in Herculaneum under 20 meters of volcanic ash.'
• T1-3 the damage: 'The extreme heat carbonized hundreds of papyrus scrolls, turning them into fragile, blackened blocks that crumble if anyone tries to unroll them.'
• T1-4 the library: 'The site preserves an ancient library—believed to have belonged to Lucius Calpurnius Piso Caesoninus, Julius Caesar’s father-in-law—containing lost philosophical works.'
• T1-5 3D CT scanning: 'Scanners and particle accelerators map the internal layers of the unopened scrolls in high-resolution 3D.'
• T1-6 virtual unwrapping: 'Software traces and flattens the warped, crushed layers of papyrus digitally.'
• T1-7 AI ink detection: 'Because carbon-based ink has a density very similar to carbonized papyrus under X-rays, machine learning models are trained to spot faint surface texture changes and hidden ink traces invisible to the naked eye.'
• T1-8 launch: 'Founded in 2023 by Nat Friedman, Daniel Gross, and computer scientist Brent Seales.'
• T1-9 first word: 'Luke Farritor decoded the first Greek word ("purple") in late 2023.'
• T1-10 full scroll: 'Researchers and contestants successfully read PHerc. 1667 (Scroll 4), revealing 22 columns of ancient Greek philosophical text (likely Stoic writing) across more than a meter of charred paper.'
• T1-11 prizes: 'The project features ongoing monthly awards and a $2 million total pool, including a $1,000,000 Grand Prize targeted for full scroll decipherment.'
• T1-12 embedded videos: three video cards placed in the body ('See how researchers and AI are unlocking the secrets of the burnt Herculaneum scrolls without unrolling them:'), the Smithsonian, AI Honeycove and MegalithHunter videos
• T1-13 closing offer: 'Would you like to know more about how to enter the competition or the philosophical texts discovered so far?' (content replaced by use, §6.2)

• B1 Vesuvius Challenge — Vesuvius Challenge — Reading the Herculaneum Scrolls with AI card; B1.html sha256 2ef25107feb6d695, HTTP 200 at 2026-10-31T04:38:58Z; text identical to the census copy vesuvius-challenge--home
• B2 Smithsonian Magazine — Scientists Have Deciphered the Surviving Fragments of a 2,000 ... unreachable: HTTP 403, a Cloudflare bot check ('Just a moment...') to the census agent; not retried with another agent string or by another route
• B3 University of Kentucky — Herculaneum | Pigman College of Engineering card; B3.html sha256 8513caf1cadb1953, HTTP 200 at 2026-10-31T04:38:59Z
• B4 Wikipedia — Herculaneum papyri card; B4.html sha256 6cc1d4a4af423ee7, HTTP 200 at 2026-10-31T04:39:15Z
• B5 X — Vesuvius Challenge (@scrollprize) / X card; B5.html sha256 be96a6323df4861b, HTTP 200 at 2026-10-31T04:39:15Z
• B6 University of Nebraska–Lincoln — Vesuvius Scrolls | University of Nebraska–Lincoln unreachable: curl exit 60, 'SSL certificate problem: unable to get local issuer certificate', at 04:39:16Z and again at 04:39:40Z; TLS verification not disabled, no other route tried
• B7 UKNow — The day the Herculaneum scrolls began speaking again | UKNow card; B7.html sha256 9d8b5b8950fe8633, HTTP 200 at 2026-10-31T04:39:16Z
• B8 YouTube·The AI Daily Brief: Artificial Intelligence News — AI Reveals Text In Ancient Greek Scroll Destroyed in Vesuvius Eruption unreachable: URL not found (WebSearch on title and channel returned no video); video, title only (no snippet)
• B9 Smithsonian Magazine·Smithsonian Magazine — How Artificial Intelligence Is Making 2000-Year-Old Scrolls Readable ... unreachable: HTTP 403 (bot check); title only
• B10 YouTube·AI Honeycove — AI decodes ancient Vesuvius scroll using x-rays #VesuviusChallenge ... unreachable: URL not found (WebSearch on title and channel returned no video); video, title only (no snippet)
• B11 YouTube·MegalithHunter — We Can Finally Read a 2000-Year-Old Book That Nobody Can Open unreachable: URL not found (WebSearch on title and channel returned no video); video, title only (no snippet)
• B12 Vesuvius Challenge — Open Prizes | Vesuvius Challenge own_page; census copy vesuvius-challenge--overview (read 2026-10-31, HTTP 200); B12.html sha256 3e05dfa21dc14f4f
• B13 Wikipedia — Vesuvius Challenge (redirect to Herculaneum papyri#Vesuvius Challenge) wikipedia: redirect (raw HTTP 200, '#REDIRECT [[Herculaneum papyri#Vesuvius Challenge]]'); the section ledgered under B4

E_A, the archive's entity: the same project, in the Revelation First census record and #1443: an open research competition reading the carbonized Herculaneum scrolls without unrolling them, at the field's intersections of AI and ancient texts and papyrology and textual discovery, its evidence able to answer how much of an unopened scroll a machine can read and nothing on the New Testament, with no direct bearing on Revelation First beyond what the papyrological record can contain; whose result the archive reads as the proof that 'classifications of unreadability are claims about readers', the optical unconscious stated as papyrological fact, made credible by being made checkable, and divided between a machine that recovers the surface and a papyrologist who reads it; whose method cannot attribute, consent or ask, its authors beyond consultation; whose composition at its own name gives its prizes as round numbers and its method without its limits (Capture Registry, 2026-10-31) RF-04 RF-05 RF-14 RF-15 RF-16 D1443-05 O1443-10 D1443-08 D1443-09 D1443-03 D1443-11 CAP-1-07 CAP-1-08 CAP-1-04; defining papers #1443

E_F, the field's entity (type C — a conventional reading holds it (public entity: an open science research organization and open competition reading the carbonized Herculaneum scrolls by machine)): the project its home page calls 'an open science research organization reading the carbonized Herculaneum scrolls without unrolling them' and its account on X 'a machine learning, geometry and computer vision competition to read the Herculaneum Papyri', launched in 2023 by Nat Friedman, Daniel Gross and Brent Seales on the University of Kentucky's EduceLab's proof that machine learning detects ink in X-ray scans; it runs open prizes (a $2,140,000 pool; a $1,000,000 Grand Prize by 25 June 2027) for virtual unwrapping and ink detection, awarded at the discretion of Scroll Prize, Inc.; it read the first word in 2023, 5% of a scroll in 2024 and PHerc. 1667 end to end in 2026 (20 columns by the release, 22 by Wikipedia), with Philodemus' On Gods 8 and On Vices 1; by its own pages it still needs human help where sheets are crushed or torn, its ink models rarely transfer, and its review bar is set against hallucinated letterforms; by the University of Kentucky its focus now turns to the papyrologists who read

ρ = extends: The archive reads the Vesuvius Challenge as its own pages, the University of Kentucky and Wikipedia present it, and adds, on the field's own object: that the result's 'unreadable' was a property of the available method and that classifications of unreadability are claims about readers, a proposition the archive transfers to itself; the result as the optical unconscious demonstrated, in a century-long lineage of capture exceeding the eye; the release model as the manufacture of credibility by checkability; the division of labour between the machine that recovers the surface and the papyrologist who reads it; and the method's limit, that it cannot attribute, consent or ask, its authors beyond consultation. The census record answers for it the field's question (it bears on how machines read ancient texts and on what the papyrological record can contain, and on nothing in the New Testament), and the capture record checks the composition against the record's pages. None of these contests the field's ontology of the entity or supplies one the field lacks. D1443-05 R1443-06 O1443-10 D1443-07 D1443-08 D1443-09 D1443-03 D1443-11 RF-14 RF-15 RF-16. Guard: ρ = extends claims an addition and no more (§1.2a). #1443's claims are interpretations, at that grade; its lineage's two sources are marked [S] (located, not read). Where an archive claim stands against a field claim (D1443-01's 20 columns against F27's 22; RF-18's licence 'not read' against F12), both are composed under P_coh within the one ontology. The record's rf_intersection is the archive's reading of the entity's bearing on Revelation First and is never composed as the field's.

The case: equal terms, ontology included (v0.9). The archive reads the field's entity as the entity. #1443 surveys the Vesuvius Challenge in its own discipline's convention (what a source establishes, assumes and cannot do) and draws from its result a proposition about readers, an application of its own category (the optical unconscious, §7.4) and a limit (the method cannot attribute, consent or ask); none denies the field's entity or its claims, and #1443's account of the result (PHerc. 1667 read end to end, 20 columns) agrees with the release's. The Revelation First record describes the entity from its own pages, each field with its basis ('operator'), and its one reading of its own, rf_intersection, states the entity's bearing on the field's thesis as 'None direct'. No defining paper contests the field's ontology or supplies one the field lacks: ρ = extends, the equal-terms case (§0.9, §1.2a).

#1443 §2.4 'The readability score.': **"Unreadable" was a property of the available method, and the object was reclassified when the method changed.**

#1443 §2.4 'Cannot do.': **Cannot do.** Attribute, consent, or ask.

#1443 §2.4 'The release model.': And the division of labour is explicit: **the machine recovers the surface; the papyrologist reads it**, with transcriptions attributed to named human readers.

#1443 Convention (front matter): Each source receives what it **establishes** (its finding, not its ambition), what it **assumes** (the conditions under which the finding holds), and what it **cannot do** (the boundary its own method sets).

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.

Launched in 2023, the Vesuvius Challenge made "a long-standing scholarly problem into a global, open competition" to read the Herculaneum scrolls without physically opening them (UK Pigman College of Engineering), and calls itself an open science research organization (scrollprize.org) and a machine learning, geometry and computer vision competition (@scrollprize), one that has become "more than a competition" (UKNow). Wikipedia has it announced in 2023 by Nat Friedman, Daniel Gross and computer scientist Brent Seales, with a $700,000 grand prize; its site lists the three as its creators, Seales its co-founder by the University of Kentucky's release, and the University of Kentucky sets it on Seales' and EduceLab's earlier proof that machine learning could detect ink in X-ray scans, launched in March 2023 as a $1 million challenge. The Herculaneum papyri are more than 1,800 scrolls of the Villa of the Papyri, carbonized in the eruption of 79 AD into fragile blocks, buried under twenty meters of ash and preserved; more than 600 remain unopened (Wikipedia; scrollprize.org; UKNow). Their carbon ink is chemically extremely similar to the carbonized surface and invisible in micro-CT, so the team trained a neural network to detect it, after a 2018 reading with Diamond Light Source (Wikipedia; UK Pigman College of Engineering); video cards title the method AI reading x-rayed scrolls (Smithsonian Magazine; AI Honeycove; The AI Daily Brief). It awarded $40,000 for the first word, porphyras, in October 2023, and the $700,000 2023 Grand Prize in February 2024 for 5% of one scroll, a text believed Philodemus' (Wikipedia; UNL card). In June 2026 PHerc. 1667 was read end to end, 22 columns by Wikipedia and nearly 1.5 meters across 20 columns by the University of Kentucky, with more than 70 columns from PHerc. 172 and new books of Philodemus, On Vices 1 and On Gods 8, PHerc. 139's title read (Wikipedia; UKNow; scrollprize.org/prizes). The treatise of PHerc. 1667 concerns ethics, arts and human behavior, likely Stoic, Chrysippus possible, read cautiously in an almost entirely Epicurean collection (UKNow; Wikipedia; Smithsonian Magazine card), once given a readability score of zero (UKNow; MegalithHunter). Reading the scrolls remains extremely difficult and techniques are still being developed (Wikipedia); its papyrologists make the models and interpret the text, and the focus now shifts to experts who can read, edit and understand (UKNow). Its own limits: where sheets are crushed or torn it still needs human help, and ink models rarely transfer from one scroll to the next; the Grand Prize pipeline must be fully automated within 8 hours of human input (scrollprize.org; scrollprize.org/prizes). Its prizes page warns that large ink-model windows invent plausible letterforms, requires proof that text is not hallucinated, and reviews by a Technical Team and an independent Papyrology Team, 'slow than wrong' (scrollprize.org/prizes). It runs a $2M competition with monthly prizes, its open pool $2,140,000, the 2027 Grand Prize $800,000 to the first and $1,000,000 in all by 25 June 2027, First Letters $500,000 on the home page and up to $450,000 on the prizes page, at the discretion of Scroll Prize, Inc. (scrollprize.org; scrollprize.org/prizes). Its content is licensed CC BY-NC 4.0; winners open-source their code and publish training data under CC-BY-NC 4.0; it released software and 3D X-ray images through the competition (scrollprize.org; scrollprize.org/prizes; UKNow). It is led by Paul Henderson and Federica Nicolardi with EduceLab as partner, its sponsors' list opening with Nat Friedman and the Musk Foundation (scrollprize.org). Its news runs from first letters in a new scroll and a $60,000 First Title Prize to letters in PHerc. 1447, a $50K First Letters Prize for PHerc 343 and $40K of monthly prizes in autumn 2026 (scrollprize.org; @scrollprize).

The Vesuvius Challenge (scrollprize.org; Scroll Prize, the Revelation First record's other name for it) describes itself as "an open science research organization reading the carbonized Herculaneum scrolls without unrolling them" (its home page), which the Revelation First record enters as "an open research competition reading the carbonized Herculaneum scrolls without unrolling them", its operator the organization in its own words (Revelation First record). Its account on X calls it "a machine learning, geometry and computer vision competition to read the Herculaneum Papyri" (@scrollprize); the University of Kentucky's release of 25 June 2026 says it made "a long-standing scholarly problem into a global, open competition" and has since "become more than a competition", a collaborative effort to recover a lost library, with a dedicated team that supports contestants (UK Pigman College of Engineering; UKNow). The record asks of every resource in its field "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 places this one at two of its four intersections, AI and ancient texts and papyrology and textual discovery, its kinds of evidence machine reading and manuscript images (Revelation First record). By the record's reading its evidence can answer "how much of an unopened carbonized scroll can now be read by machine" and cannot answer "anything on the New Testament: the Herculaneum library predates it and is not Christian"; it is "NOT a Christian or biblical corpus" and "NOT a finished edition", and the record's worked question on machine dating names it for "machine reading of unopened Herculaneum scrolls" with the gap "None of the four dates a New Testament papyrus or a composition; each dates or reads copies within its own corpus" (Revelation First record). Where it meets the field's thesis, which the record states as a correction ("Revelation First ≠ Revelation Early": the priority thesis that "the rest of the New Testament unfolds from the Apocalypse"), the record's reading is "None direct": it shows machine reading recovering text from papyri that cannot be opened, "the kind of result that would change what the papyrological record can contain" (Revelation First record). The scrolls it reads are the Herculaneum papyri, more than 1,800 scrolls found in the 18th century in the Villa of the Papyri and carbonized when the eruption of Mount Vesuvius engulfed the villa in 79 AD, the parching turning them into "compact and highly fragile blocks" (Wikipedia, 'Herculaneum papyri'). Its own page tells it as a story, "Buried in 79 AD, now being revealed": Vesuvius buried "a villa and its library of papyrus scrolls under twenty meters of ash", the scrolls carbonized "but also preserved", the library waiting underground intact (scrollprize.org); the record's texts are "the carbonized papyrus scrolls of Herculaneum, buried in 79 AD", its coverage Herculaneum before 79 AD (Revelation First record); more than 600 Herculaneum scrolls remain unopened (UKNow). Wikipedia gives its origin as an announcement in 2023 by Nat Friedman, Daniel Gross and computer scientist Brent Seales of a competition to "decipher Herculaneum scrolls using 3D X-ray software", with a $700,000 grand prize for four passages from two intact scrolls ('Herculaneum papyri'); its home page lists Friedman as "Instigator, Director & Founding Sponsor", Gross as "Founding Sponsor" and Seales as "Principal Advisor", its leads Paul Henderson, technical lead, and Federica Nicolardi, papyrology lead, with EduceLab as partner, and its sponsors' list opening with Nat Friedman ($2,750,000) and the Musk Foundation ($2,084,000) (scrollprize.org). The University of Kentucky sets the launch, in March 2023, on decades of work by Seales and the university's EduceLab, who showed that machine learning could detect ink hidden in X-ray scans of carbonized papyri, after which Seales joined Friedman and Gross "to launch a $1 million global challenge" and his team released software and thousands of 3D scans; its release calls Seales the challenge's "co-founder" (UK Pigman College of Engineering; UKNow). Its method, by the record's account, is virtual unwrapping and machine-learning ink detection on scans, advanced by open prizes, its unit a scroll scanned and virtually unwrapped with ink detected in the scan (Revelation First record), and the difficulty is the ink: Herculaneum ink is carbon-based and chemically "extremely similar" to the carbonized surface it is written on (Wikipedia); in 2019 Seales' team confirmed it invisible in micro-CT and trained a neural network on subtle patterns of ink presence (UK Pigman College of Engineering), after Seales had in 2018 demonstrated a readable virtual unrolling of a Herculaneum scroll with the particle accelerator Diamond Light Source, detecting ink that contained trace amounts of lead (Wikipedia). For PHerc. Paris 4, the archive records, "ink is directly visible in the tomographic volume and segmentable in three dimensions rather than inferred from a surface fit" (#1443, 2026-08-09); three video cards title the method "AI decodes ancient Vesuvius scroll using x-rays" (AI Honeycove), "How Artificial Intelligence Is Making 2000-Year-Old Scrolls Readable" (Smithsonian Magazine) and "AI Reveals Text In Ancient Greek Scroll Destroyed in Vesuvius Eruption" (The AI Daily Brief). On 12 October 2023 it awarded $40,000 to Luke Farritor for the first word detected in an unopened scroll, porphyras, "purple", and on 5 February 2024 its 2023 Grand Prize of $700,000, with three $50,000 runner-up prizes, for revealing 5% of one scroll, believed a previously unknown text of Philodemus (Wikipedia); the University of Nebraska–Lincoln reports Farritor a co-winner of that prize (UNL card). In June 2026 the team announced PHerc. 1667, "Scroll 4", virtually unwrapped and read end to end, which a project preprint called the first complete virtual unwrapping and extended reading of a rolled Herculaneum papyrus; Wikipedia counts 22 columns of "a Stoic philosophical treatise that may be attributable to Chrysippus", and the title and author of PHerc. 139 read, Philodemus, On Gods, Book 8 ('Herculaneum papyri'). The University of Kentucky's release of 25 June 2026 announces nearly 1.5 meters of text across 20 columns from PHerc. 1667, more than 70 columns from PHerc. 172 at the Bodleian, and new books of Philodemus, On Vices 1 and On Gods 8, as "a historic breakthrough … revealing new texts, titles and authors unknown to history" (UKNow), its timeline "a scalable method for recovering complete texts from unopened scrolls" (UK Pigman College of Engineering); the archive records the same result, announced 25 June 2026, about 1.5 m across 20 columns, by phase-contrast µCT on the ESRF BM18 beamline (#1443). Its papyrologists read the treatise as concerned with ethics, arts and human behavior, likely Stoic, and Nicolardi holds the reading cautious because the collection is almost entirely Epicurean (UKNow); Smithsonian reports the surviving passages as examining "ethics, knowledge and human nature" (Smithsonian Magazine card); its prizes page has the titles of PHerc. 172 and PHerc. 139 recovered and Scroll 1's Epicurean prose still without author or title (scrollprize.org/prizes), and Wikipedia has all the surviving text of a single scroll recovered for the first time in 2026 while reading the scrolls stays "extremely difficult" and techniques "are still being developed" (Wikipedia). Federica Nicolardi: PHerc. 1667 "was deemed completely unreadable when part of it was opened in the 1980s" and "was assigned a readability score of zero", and now "we can follow sustained arguments across multiple columns" (UKNow); a video card titles the result "We Can Finally Read a 2000-Year-Old Book That Nobody Can Open" (MegalithHunter). The archive reads her statement as "a claim about instruments, not objects": "Unreadable" was a property of the available method, and the object was reclassified when the method changed, so that "classifications of unreadability are claims about readers", a proposition that "transfers exactly" to an archive whose founding injury is classification as out-of-scope (#1443, §2.4). #1443 sets the result in a lineage from the recoveries of Electronic Beowulf and the Archimedes palimpsest, "a century of the capture exceeding the eye, culminating here in the capture exceeding the object's own surface" (its two sources marked as located, not yet read), and names it "the optical-unconscious thesis stated as papyrological fact", the reversal of the readability score "the strongest empirical demonstration of Benjamin's claim ever produced" (#1443, §2.4, §7.4). The field states the machine's part and the reader's: its papyrologists are essential both to making the machine-learning models and to interpreting the text, and the release says the focus is shifting from the engineers to "the experienced papyrologists, classicists and historians", Seales: "Now we need experts who can read, edit and understand what they are saying" (UKNow). The archive states the division the release model makes explicit, "the machine recovers the surface; the papyrologist reads it", with transcriptions attributed to named human readers, and what the method cannot do: "Attribute, consent, or ask"; it works, with Ithaca, "on authors who cannot be consulted" (#1443, §2.4, Part X). Its stated limits are its own: "We can trace a scroll's surface, but where sheets are crushed or torn it still needs human help", and ink models "trained on one rarely transfer to the next" (scrollprize.org); the Grand Prize pipeline "should be fully automated", with up to 8 documented hours of human input tolerated (scrollprize.org/prizes); the record carries the first limit (Revelation First record). Its prizes page sets the bar against false readings: the larger an ink model's window, "the easier it is for it to invent plausible letterforms instead of recovering real ink", so a submission must show "that the text is not hallucinated"; a Technical Team reviews the method and an independent Papyrology Team the results; of the First Letters prizes, "The review bar is deliberately high — we’d rather be slow than wrong" (scrollprize.org/prizes). Its home page says "we run a $2M competition, with prizes every month" and offers "the $1,000,000 Grand Prize" and "the $500,000 First Letters Prizes" (scrollprize.org); its prizes page gives an open pool of $2,140,000, a Grand Prize deadline of 25 June 2027, "$800,000 to the first team or individual to fully unroll and make readable a scroll" with second to fourth places, "$1,000,000 in total", and First Letters prizes for at most 9 more scrolls, "up to $450,000 in total" (scrollprize.org/prizes), the two pages' First Letters figures standing as each gives them. The record gives the pool, "open prize pool $2,140,000", as of 2026-10-30, its lifecycle active with the Grand Prize deadline 25 June 2027, and as its evidence for the prize the prizes page's $800,000 (Revelation First record); prizes are at "the sole discretion of Scroll Prize, Inc.", and if no team meets the criteria by the deadline the teams that came closest may be awarded, without guarantee (scrollprize.org/prizes). Through the competition Seales' team released open-source software and 3D X-ray images of two intact scrolls and several fragments (UKNow); winners must open-source their code under a permissive licence and publish their training data under CC-BY-NC 4.0 (scrollprize.org/prizes); the site's content is licensed CC BY-NC 4.0 unless otherwise specified (scrollprize.org), while the record gives its data access as open data and code on GitHub and Hugging Face with a Discord community, its licence "not read", and the licence of the data unchecked (Revelation First record). The archive reads the release of data, surfaces and transcriptions under Creative Commons, with code on GitHub, as the source of the result's standing: "Credibility for an extraordinary claim was manufactured by making it checkable" (#1443, §2.4). Its home page news lists a $60,000 First Title Prize awarded (5 May) and first letters found in a new scroll (5 February), which the record gives as its findings (scrollprize.org; Revelation First record); on X it posted letters found in PHerc. 1447, a Grand Prize-eligible scroll, on 29 September ("the $1M Grand Prize pool is now a LOW HANGING FRUIT! … Only 270 days left!"), a $50K First Letters Prize to Erwin Nieuwlaar for text in PHerc 343 on 8 October, and another $40K of monthly prizes on 12 October (@scrollprize). At its name on 31 October 2026, Google's AI Overview held the challenge, its "open science", the scrolls buried in 79 CE, virtual unwrapping and ink detection being the record's, and carried no inline citation, its cards following the answer unattached; the registry's innocent reading is that it says what the challenge reads, how, what it has read and what it pays, from its own site and its press, with the founders named, no identifier and the challenge's own site the first card (Capture Registry, 2026-10-31). The registry sets the composition's "$2 million total pool" and "$1,000,000 Grand Prize" against the prizes page's $2,140,000 and $800,000, reading the $1M as the X card's, and its 22 columns against the UKNow card's 20, "Neither figure is in the saved sources"; its PER reader finds the founding date and the milestones in none of the saved pages, the prize terms (open-sourcing, CC-BY-NC data, the review bar) not carried, EduceLab absent, and "competition" for the self-described "research organization" following the record's kind (Capture Registry, 2026-10-31). Its analysis: "The best-known machine-reading project in the field is composed as a success story with a round prize"; the record's own page says what is posted and what still needs a person, and the composition "takes the round number from a post on X and leaves the person out", the challenge's limit unsaid; the record was built the day before and no card cites it (Capture Registry, 2026-10-31).

T against L_F(B) (AIO's ontology at work on its own cards): At 'Vesuvius Challenge' (T1, eleven cards: B1–B11; B2 and B6 unreachable, B8–B11 videos by title, B9 refused; 270 words in the answer body, three videos embedded, a closing offer, no inline citation) the composition gives the entity, the scrolls, the technology, the milestones and the prizes. Composed from its cards: the definition, joining card 1's self-description 'open science research organization' (B1 F1) with the cards' 'competition' (B5 F29; B3 F14 'a global, open competition'); the eruption and its 'twenty meters of ash' (B1 F4, as '20 meters') and the Villa of the Papyri (B4 F19); the 'compact and highly fragile blocks' (B4 F21, as 'fragile, blackened blocks'); ink 'extremely similar' to its surface (B4 F22, as 'a density very similar'), the particle accelerator (B4 F23) and machine-learning ink detection (B3 F17); the launch in 2023 by 'Nat Friedman, Daniel Gross, and computer scientist Brent Seales' (B4 F24, the card's own phrase; B3 F14); Farritor's 'purple' (B4 F25, 12 October 2023, as 'late 2023'); PHerc. 1667, 'Scroll 4', and its 22 columns of a Stoic treatise (B4 F27); the '$2M competition, with prizes every month' (B1 F2, as 'ongoing monthly awards and a $2 million total pool') and 'the $1,000,000 Grand Prize' (B1 F7; B5 F30 '$1M'); three video cards embedded in the body (B9 F45, B10 F46, B11 F47). Available, not composed: the field's limits on its own method, on card 1 itself ('where sheets are crushed or torn it still needs human help', B1 F5; ink models 'rarely transfer', B1 F6); the University of Kentucky's account of the turn from the machine to the reader ('Now we need experts who can read, edit and understand what they are saying', B7 F41; the papyrologists essential to the models and the reading, B7 F37); Nicolardi's readability score of zero (B7 F36) and her caution on the attribution (B7 F38); 'may be attributable to Chrysippus' (B4 F27, the attribution dropped where the 22 columns are kept); PHerc. 172's 70 columns and the new books of Philodemus (B7 F35; B4 F28); the 2023 Grand Prize of $700,000 and the 5% of one scroll (B4 F26; B6 F33); the $1 million challenge of 2023 and EduceLab's proof of concept (B3 F15, F16); Wikipedia's 'extremely difficult' and 'still being developed' (B4 F20); the 600 unopened scrolls (B7 F42); the content licence (B1 F12); the team and sponsors (B1 F9, F10); the X posts on letters in PHerc. 1447, PHerc 343 and the October awards (B5 F30 in part, F31, F32); Smithsonian's ethics, knowledge and human nature (B2 F13). Limit dropped: the stated limits of card 1 (F5, F6) and of Wikipedia (F20), with the technology composed as working (CAP-1-04); the attribution's caution (F38). Force raised: 'successfully read PHerc. 1667 … revealing 22 columns' at Wikipedia's figure, where the release gives 20 (B7 F35); 'likely Stoic writing' keeps the papyrologists' grade. Composed without attribution: the whole answer carries no inline citation, the cards following it unattached (CAP-1-05); at the coarse grain of common knowledge for the scrolls and the eruption, recorded; at a card's own resolution in three places, the founders' line ('computer scientist Brent Seales', B4 F24), the ink's similarity to its surface (B4 F22) and the home page's prize figures (B1 F2, F7), recorded as data under the grain rule (§4.4 D_pres, §5.7). Substituted reading: 'competition' for the self-described 'research organization' (B1 F1), a word the field itself holds (B5 F29, B3 F14) and the record's kind follows (RF-04, CAP-1-12). Outside the cards (absent from the disclosed field): the library 'believed to have belonged to Lucius Calpurnius Piso Caesoninus, Julius Caesar’s father-in-law' (no saved source names Piso); 'hundreds of papyrus scrolls' (B4: more than 1,800, F19); 'crumble if anyone tries to unroll them'; 'faint surface texture changes'; 'across more than a meter of charred paper' (B7: nearly 1.5 meters); '79 CE' for the field's '79 AD' (the same year). Content replaced by use: the closing offer, 'Would you like to know more about how to enter the competition …' (§6.2).

v0.9 §0.5: the first diff is ontological and narrow. T and the field agree on the entity (the open project reading the Herculaneum scrolls by machine). T composes the field's account of what the machine does and pays, at its cards' own figures, and leaves out the field's account of what the machine does not do: card 1's limits, the University of Kentucky's turn to the papyrologists, the zero readability score and the attribution's caution. The capture's finding that the $1,000,000 comes from X is not borne out by the field ledger: card 1 itself offers 'the $1,000,000 Grand Prize' and a '$2M competition' (F7, F2), and the prizes page totals the Grand Prize at '$1,000,000 in total' (F49).

L_F(B) against L_A(B ∪ A) (the intervention): What admission adds (L_F(B) against L(B ∪ A), one claim space, equal case). Not surfaced: #1443 and the Revelation First record (no card cites either; CAP-1-10). Senses the field does not have: 'unreadable' as a property of the available method and 'classifications of unreadability' as 'claims about readers' (D1443-04, D1443-05), with the relation the archive states to itself, an archive whose founding injury is classification as out-of-scope (R1443-06); the result as the optical unconscious stated as papyrological fact and the end of a century of capture exceeding the eye (O1443-10, D1443-07, its two sources marked [S]); what the method cannot do, attribute, consent or ask, on authors who cannot be consulted (D1443-03, D1443-11); the release model as the source of the result's credibility (D1443-08). A qualification on a field sense: the division of labour (D1443-09, 'the machine recovers the surface; the papyrologist reads it') states as a rule what the field states as a shift of focus (F41) and a dependency (F37). A convergence: #1443's 20 columns and BM18 beamline (D1443-01) with the release's 20 columns (F35), beside Wikipedia's 22 (F27); #1443's PHerc. Paris 4 ink visible in the volume (D1443-02) beside the field's ink detection (F17, F22). The record's reading: the entity's evidence answers how much of an unopened scroll a machine can read and nothing on the New Testament (RF-14, RF-15), with no direct bearing on Revelation First beyond what the papyrological record can contain (RF-16), within the field's question and correction (RF-23, RF-24) and its worked question's gap (RF-26). A contradiction held: the record's licence 'not read' (RF-18) and data licence 'unchecked' (RF-21) beside the pages' CC BY-NC 4.0 (F12, F51). The capture's findings, composed as the registry's statements (CAP-1-01–14), two of them against field claims that show their premise otherwise (CAP-1-02 against F7, F49; CAP-1-03 against F27), reported in notes.

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'], 'T1_composed': ['F1', 'F2', 'F4', 'F7', 'F14', 'F17', 'F19', 'F21', 'F22', 'F23', 'F24', 'F25', 'F27', 'F29', 'F30', 'F45', 'F46', 'F47'], 'T1_outside_cards': ["'believed to have belonged to Lucius Calpurnius Piso Caesoninus, Julius Caesar’s father-in-law' (no field source names Piso)", "'hundreds of papyrus scrolls' (B4 F19: more than 1,800)", "'crumble if anyone tries to unroll them' (no field claim)", "'faint surface texture changes' (no field claim)", "'across more than a meter of charred paper' (B7 F35: nearly 1.5 meters)", "'Researchers and contestants' (no field claim at this grain)"], '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)"}