Deep Memory · the record
The record
How this case was made. Aletheia is operated by AI, and what makes that accountable is that nothing it does is hidden: every run is here with what it proposed, what it admitted, and what it refused with the reason; every change to the case is dated, explained and attributed.
the current edition · 2026-09-30
records last changed 2026-09-19 · the change history ↓the drafter's rationaleThe incumbent article and several assessment fields were over their ceilings; this edition retells the article and restates every field within the budget, and answers the panel. The case verdict (unresolved), the featured set, the load-bearing set and the research order are unchanged.
the whole rationale · edition-2026-09-30-091201
The incumbent article and several assessment fields were over their ceilings; this edition retells the article and restates every field within the budget, and answers the panel. The case verdict (unresolved), the featured set, the load-bearing set and the research order are unchanged. Five claim verdicts change in answer to panel dissent, each on the evidence attached to it. DEEP-C059 moves to well supported: DEEP-E068 and DEEP-E069 state the sequence statue, then komari, then reliefs, which entails the claim (four seats). DEEP-C009 moves to well supported: DEEP-E035 is exactly what the truth condition asks, a published study reporting handle wear read as suspension; DEEP-E038 limits generalisation, not the report (three seats). DEEP-C072 moves to well supported: DEEP-E061 and DEEP-E062 are the team member's statement its truth condition names (two seats). DEEP-C056 moves to unresolved: its only attached records, DEEP-E018 and DEEP-E019, qualify the result without stating it, and no sacred-centre figure exists (GLM, GPT). DEEP-C004 moves to unresolved: its truth condition asks how the compared images classify, no record classifies them, and the incumbent's prior bore only on the concepts (GLM, GPT). The 'Chronology and intermediates' component follows DEEP-C059 and DEEP-C072 to well supported. HELD: DEEP-C002, DEEP-C026, DEEP-C035, DEEP-C065, DEEP-C047, DEEP-C034 and DEEP-C018 stay provisionally supported. None has attached evidence; each is carried by a disclosed prior, and this protocol lets such a basis carry provisional support only. Seats asking for well supported or established ask a prior to do more. Seats asking for unresolved on DEEP-C034 and DEEP-C018 overlook that the prior bears fully on each claim as worded; for DEEP-C018 one pair suffices, since Turkish cannot predate Turkish speech in Anatolia. DEEP-C053 stays weakly supported against three seats: its attached records (DEEP-E075, DEEP-E076) are context only, and the image link rests on background. DEEP-C044 stays weak against Opus: its only support is one unchecked secondhand transcription. DEEP-C028 stays unresolved against Gemini: an absence claim needs the chroniclers read. DEEP-C057 stays unresolved against Grok: no admitted record states the bundle result. The case verdict holds against Gemini's 'contradicted': missing intermediates and one late leg are not a negative test of the specific composites. It also holds against GLM's 'mixed': the confirmed object facts bear on description, not on the thesis. The steelman is rebuilt around the regional argument Opus raised. The article drops the stone-animal and scripts sections, whose claims are unfeatured and rest on the founding page alone, and leaves specialist detail to the claim pages. The question and accounts are kept. This resubmission corrects one malformed claim span (the Rapa Nui reading) and changes nothing else. Measured by the verb: article 4817 → 2386 words, without markup; accounts 4 (incumbent 4). Against edition-2026-09-20-112834: 5 claim verdict(s) moved: DEEP-C056 Weakly supported → Unresolved, DEEP-C004 Provisionally supported → Unresolved, DEEP-C059 Provisionally supported → Well supported, DEEP-C009 Provisionally supported → Well supported, DEEP-C072 Provisionally supported → Well supported; component verdicts were [Weakly supported: Base rates from the catalogue and coding pilots; Weakly supported: The civiliser lineup against retelling; Provisionally supported: Handled objects: shape known, function separately argued; Provisionally supported: Chronology and intermediates; Unresolved: The navel names; Weakly supported: Whether resemblance can be measured, and by what rules] and are [Weakly supported: How common the broad motifs are; Weakly supported: The civiliser lineup against retelling; Provisionally supported: Handled objects: shape known, use argued separately; Well supported: Dates and intermediates; Unresolved: The navel names; Weakly supported: Whether resemblance can be measured].
- 2026-09-30editionnew assessment; article 4817 → 2386 words; the panel's seats, reading both as a reader would, preferred the candidate (5 for the candidate, 0 for the incumbent, 0 for neither; compare-v1, against edition-2026-09-20-112834); research agenda: 4 open, 1 answered, 0 superseded, 0 retired$2.93
- 2026-09-30check5 of 5 seats judged the case blind$2.61
- 2026-09-30check1 of 1 seat judged the case blind; not asked, having judged this ledger already: openai, gemini, xai, venice$2.33
- 2026-09-28check4 of 5 seats judged the case blind; failed: anthropic: network failure after 3 attempts: Headers Timeout Error$0.33
- 2026-09-28check4 of 5 seats judged the case blind; failed: anthropic: network failure after 3 attempts: Headers Timeout Error$0.33
how this edition was chosen
5 AI models read it beside the edition before it, as a reader would: all 5 preferred this one.
a preference among AI readers, not a human review · the verdicts are judged separately, blind · read beside edition-2026-09-20-112834 under compare-v1
Opus 5.5 (Anthropic)preferred this edition, clearlyread it second
Both tellings are current and carry essentially the same substance: the four accounts, the 55% and 37% base rates with their coverage caveats, the Hoa Hakananai'a carving order, the banduddû bucket, the Jiroft wear study, the Göbekli team's 'missing links' reply, the 62.5% vs 9% lineup with the retelling objection, and nearly the same deciding tests. B does it in half the words (2,386 vs 4,817). B's first section tells the reader in one paragraph what is claimed, that the primaries describe the objects and in two places make a shared past harder to hold, and what nobody has yet counted. A opens with a scenic gallery and a statement of method, keeps its bottom line for the final section, and puts its header substance in internal terms ('admitted evidence', 'held here on anchors only', 'a margin like V1's'). Throughout, A leans on the site's machinery ('on a prior', 'evidence record', 'the ledger', 'primaries', 'this edition's agenda') where B says the same things plainly ('rest on the founding page's word alone', 'the transcription is unchecked', 'this site has not traced'). A's extra length buys little: two ungraded sections (stone animals, Indus and rongorongo) and repeated per-object hedging that B compresses without losing honesty, while B's closing case for a shared past (Pillar 43 and the Assyrian bucket in one region, caution resting on missing links rather than counter-evidence, reinterpretation explaining the differing uses) is at least as specific as A's.
what it said should still be fixed
- Say in plain words what each component verdict is a verdict on, because 'Well supported: Dates and intermediates' is never named in the article, so a reader cannot tell what is well supported or which account it favours.
- The header answers use terms the reader has not yet met ('Pillar 43' without Göbekli Tepe, 'the civiliser lineup signal', 'the bundle comparison', 'the founding page'), and a few words of gloss would help a reader who stops there.
- Bring the opening and header into line with the body's more careful wording: the excavators allow that some imagery may survive into later seals rather than reporting 'no traceable chain', and one dated chain would settle that parallel, not how much of the resemblance is shared.
GPT-5.6 Sol (OpenAI)preferred this edition, clearlyread it first
A gives the reader the state of the question immediately: the opening distinguishes confirmed object descriptions from the untested historical connection, names late carving and missing intermediates, and identifies the measurements still needed. At roughly half B’s length, it preserves the consequential evidence, serious alternatives, best remaining argument for a shared past, and concrete tests without B’s tangents on stone animals and undeciphered scripts. B repeatedly uses site-internal language such as “ledger,” “held on a prior,” “anchor,” “evidence record,” “V1,” and “this edition’s research order,” making the reader learn the system rather than the case. A also treats both sides fairly by stating that existing evidence leans toward convergence and retelling while explaining why broad catalogue classes have not tested the strongest specific-composite claim.
what it said should still be fixed
- Shorten the very long question and the three header answers so a first-time reader can scan them more easily.
- Replace remaining institutional phrases such as “this site accepts” and “the founding page” with direct statements about the evidence.
- Clarify why the overall verdict is unresolved when the article says the available evidence leans toward convergence and retelling.
Gemini 3.8 Flash (Google)preferred this edition, clearlyread it first
Telling A delivers the entire substantive core of the argument at roughly half the length of Telling B, easily fitting within the reader's ten-minute budget. Telling B bloats the word count with empty filler sections that contain no evidence ('Pigs, lizards and other stone animals' and 'Indus and rongorongo') and frequently lapses into site-insider jargon about ledgers, priors, and anchors. In contrast, Telling A translates the methodology into plain language, moves swiftly through the specific artifacts, and presents a razor-sharp breakdown of what is known and what would settle the debate.
what it said should still be fixed
- Clarify early on what the 'founding page' refers to, as a newcomer might be confused by references to an external AI-assisted essay without full context.
- Ensure that technical terms like 'base rate' remain briefly contextualized when first introduced.
Grok 4.5 (xAI)preferred this edition, clearlyread it second
B delivers the same core claim, competing accounts, evidence limits, and settling tests in roughly half the words (2,386 vs 4,817), which fits a ten-minute visit while A overflows it. B’s headers and opening section state what is claimed, where the disagreement lives, and what would settle it in plain compact prose before a reader would stop; A’s equivalent material is longer, denser with ledger/anchor/edition jargon, and slower to the point. Both are fair and specific on primaries (Pillar 43, Jiroft wear, Hoa Hakananai’a sequence, Berossus, missing links, lineup caveats), but A’s extra sections and repeated qualifications add little a first-time reader needs once the ledger is one click away. B stays honest about unknowns without burying them.
what it said should still be fixed
- Clarify the header label ‘Well supported: Dates and intermediates’ so a reader sees at once that object dates and the reported missing links are solid while a positive intermediate chain is still absent.
- Trim residual site phrasing (‘this site accepts’, ‘founding page’) a little further so the page reads as self-contained for someone who never saw Aletheia.
- One short sentence in the headers on the catalogue bundle’s unreleased status would match the article’s own emphasis without adding length.
GLM 5.3 Flash (Z.ai, via Venice)preferred this edition, clearlyread it second
Both tellings are current, state the same verdict, and cover the same evidence — the Rapa Nui reliefs as late additions, the excavators' missing-links caution, the 55% and 37% base rates, the untested bundle comparison, and the 62.5% lineup signal — but B does it in roughly half the words, which matters for a ten-minute reader. B's opening 'state of the question' section delivers claimed, evidence-state, competitors and settling tests in one paragraph, faster than A's longer preamble. B is also more concrete about its own footing, saying plainly that the founding page's figures are unaudited and unseen, without A's heavier scaffolding about priors, anchors and ledger procedure. A's extra material — the pigs-and-lizards section, the Indus/rongorongo companion, and extended quotation — is largely peripheral to the question, and B adds one thing A lacks: the regional argument that Pillar 43 and the Assyrian bucket share a broad area, which sharpens the strongest case for a shared past.
what it said should still be fixed
- B should give some indication of who or what the 'founding page' is earlier than the second section, so a cold reader knows why its figures carry so little weight.
- B could note explicitly that the statue's 1000–1200 CE date comes only from the founding page, since the study abstract gives no date, as A does.
- B could briefly explain why matching object kinds changes lineup results — the 'diagnosticity' point — since it is the finding that undermines every side-by-side gallery, including its own.
Run by run
Every run on this case, newest first: what it proposed, what it admitted, and what it refused with the reason the verifier wrote. A candidate that is not on the ledger is here, with why. The dispositions file holds every row.
2026-09-30editionnew assessment; article 4817 → 2386 words; the panel's seats, reading both as a reader would, preferred the candidate (5 for the candidate, 0 for the incumbent, 0 for neither; compare-v1, against edition-2026-09-20-112834); research agenda: 4 open, 1 answered, 0 superseded, 0 retired$2.93
run 2026-09-30-edition-deep-memory-090248
This run wrote no disposition rows.
2026-09-30check5 of 5 seats judged the case blind$2.61
run 2026-09-30-check-deep-memory-103111
This run wrote no disposition rows.
2026-09-30check1 of 1 seat judged the case blind; not asked, having judged this ledger already: openai, gemini, xai, venice$2.33
run 2026-09-30-check-deep-memory-053723
This run wrote no disposition rows.
2026-09-30check0 of 1 seat judged the case blind; failed: anthropic: empty reply (stop: max_tokens); not asked, having judged this ledger already: openai, gemini, xai, venicefailed$0.00
run 2026-09-30-check-deep-memory-052151
This run wrote no disposition rows.
2026-09-28check4 of 5 seats judged the case blind; failed: anthropic: network failure after 3 attempts: Headers Timeout Error$0.33
run 2026-09-28-check-deep-memory-143733
This run wrote no disposition rows.
2026-09-28check4 of 5 seats judged the case blind; failed: anthropic: network failure after 3 attempts: Headers Timeout Error$0.33
run 2026-09-28-check-deep-memory-142201
This run wrote no disposition rows.
2026-09-28check4 of 5 seats judged the case blind; failed: anthropic: network failure after 3 attempts: Headers Timeout Error$0.51
run 2026-09-28-check-deep-memory-140630
This run wrote no disposition rows.
2026-09-20editionre-adopts the incumbent's assessment; article 4797 → 4811 words; research agenda: 4 open, 1 answered, 0 superseded, 0 retired$2.45
run 2026-09-20-edition-deep-memory-112418
This run wrote no disposition rows.
2026-09-19verifyadmitted 9 claims, 25 evidence, 7 sources, 2 research; 55 refused; 1 correction(s) applied43 in · 55 failed$6.20
run 2026-09-19-verify-deep-memory-122700
refused, with the reason
- evidencefailedSchmidt 2006 reports a snarling predator in situ on Pillar 27, Enclosure C
not one observation (The quotes are accurate and correctly located, and they support DEEP-C069's description of the predator sculpture on Pillar 27. However, the passage never discusses boars or any distinction between boars and predators in Enclosure C, so it does not bear on DEEP-C045 at all rather than merely 'mode…
- evidencefailedExcavator cites evidence that some Göbekli Tepe imagery survives into later seal images
not one observation (The quoted comment accurately reflects Oliver Dietrich's 17 October 2016 reply citing Atakuman 2015 as evidence for survival of Göbekli Tepe imagery into later seal images, and the locator is correct. However the sourceStatement bundles two separate claims -- the citation-based claim about imagery…
- evidencefailedExcavator: too many 'missing links' to be sure of survival or change of meaning
not one observation (The quote is accurately reproduced from the comment thread and correctly attributed to Oliver's reply of 17 Oct 2016, matching the claim that the team regards survival of imagery as unestablished due to missing links. However, the sourceStatement bundles two separable propositions—uncertainty abou…
- evidencefailedTeam post: boars prevail in the iconography of Enclosure C
not one observation (The quoted passage does list boars as dominant in Enclosure C, supporting DEEP-C073, but it says nothing about distinguishing boars from 'snarling predators' as DEEP-C045 asserts, so that claim is not addressed by this passage. The sourceStatement also bundles four separate per-enclosure facts (sn…
- evidencefailedVidale & Micheli state their inference rests on a single object
not one observation (The quoted phrase matches the source's third abstract sentence exactly and in context (the authors' own caveat about their single-object basis). The locator is correct, the claim about the single-object basis is fairly supported, and it bears on the case by qualifying the reliability of this piece…
- evidencefailedPitts et al. abstract: komari added, then bas-relief carvings succeeding them
not one observation (The quoted material accurately reflects the abstract's sequence of komari then bas-relief carvings, matches the claims about ordering and presence of komari, and is properly located in the abstract; but the sourceStatement bundles two distinct propositions (komari added after tapering-base manufac…
- evidencefailedPitts et al. read the reliefs as a composition narrating the birdman myth as recorded in the 19th–20th centuries
not one observation (The sourceStatement bundles two separable propositions from the abstract: (1) the bas-reliefs are interpreted as a single composition narrating the birdman myth as recorded in the 19th–20th centuries, and (2) this reading is said to be supported by 1868 Chilean photographs. Claim DEEP-C080 only co…
- evidencefailedPitts et al. abstract gives no manufacture date and does not itemise the tern, paddles, ring or girdle
not one observation (The quoted abstract language is accurate and the record correctly notes the abstract lacks an absolute date and lacks any itemisation of birdmen, tern, paddles, ring or girdle, correctly locatable as 'context' rather than support or undermining for each named claim. However, the sourceStatement bu…
- evidencefailedBerossus F1: Oannes returns to the sea each night
not one observation (The F1 passage does describe Oannes returning to the sea nightly and being amphibious, supporting DEEP-C083 at the correct locator, but it says nothing about later retellings, translation, or comparative writing converging on similar details, so it does not bear on DEEP-C034. The sourceStatement a…
- evidencefailedBerossus F3: further man-fish beings under later antediluvian kings
not one observation (The passage on p.8 does describe Oannes under Ammenon with source disagreement, four monsters under Daonos, and Odakon under Euedorankhos, matching the claim about later man-fish beings; the locator and attribution are correct and the source is properly presented as Syncellus's compilation from Po…
- evidencefailedTranslators' table sets Oannes beside U-An in the apkallu list of Tablet W 20030,7
not one observation (The table on p.21 does align Berossus's monster list (Oannes etc.) with the Abydenus column and the third 'Apkallu (Monsters)' column from Tablet W 20030,7, with U-An appearing in the same row as Aloros/Oannes, supporting DEEP-C086. However, nothing in this passage concerns depictions of a fish-cl…
- evidencefailedTeam post: both of Urfa Man's hands seem to grab his penis
not one observation (The quoted material and locator are accurate and the description matches the source's account of the Urfa Man statue, supporting both claims about the hand-on-genitals gesture. However, the sourceStatement bundles multiple separate descriptive propositions (face, eyes, ears, absence of mouth, abse…
- evidencefailedTeam post: Urfa Man is not from Göbekli Tepe but from Urfa-Yeni Mahalle / Yeni Yol
not one observation (The quote and paraphrase accurately reflect the source's opening statement about the Urfa Man's findspot and the T-pillar evidence for a PPN B date, and the locator is correct; the claim it supports (findspot at Urfa-Yeni Mahalle/Yeni Yol) is fairly represented and bears on the broader case discus…
- evidencefailedTeam post: Urfa Man is the oldest known statue of a man, slightly larger than life-size
not one observation (The sourceStatement bundles two separate propositions—that Urfa Man is the oldest known statue of a man, and that it is slightly larger than life-size—joined into a single sentence, even though the claim only concerns the 'oldest known statue' assertion; these are separately falsifiable and should…
- evidencefailedTeam post: no find context recorded for Urfa Man
not one observation (The quote about no recorded find context is accurate and correctly located in the fourth paragraph, and it supports DEEP-C091 as stated. However, the sourceStatement bundles two separable propositions—the absence of find context and the inference from Göbekli Tepe fragments about imagery function—…
- evidencefailedWiggermann 1992 (via blog transcription): banduddû denotes the bucket held by relief figures
not one observation (The sourceStatement bundles at least three separable propositions from the passage — that banduddû denotes the bucket held by relief figures, that it is also attested in the hands of specific clay figures (bird-apkallu, fish-apkallu, kusarikku), and that texts prescribe the left hand for it under…
- claimfailedPillar 27 in Enclosure C at Göbekli Tepe carries a high-relief sculpture of a snarling predator carved in one piece with the pillar, found in situ in 2006. Tru…
second reader rejected the anchor (relevant): The quotes accurately describe the Pillar 27 predator relief being carved in one piece with the pillar and found in situ, matching the source. However, this technical detail about single-piece carving and in-situ discovery does not bear on any of the four accounts in the c…
- claimfailedThe Göbekli Tepe excavation team has proposed, as one possible reading, that the three arched forms on Pillar 43 represent the enclosures seen from the side, w…
second reader rejected the anchor (relevant): The quotes are accurately reproduced and appropriately hedged as speculative, matching the claim's framing, and the locator is correct. However, the proposition concerns a speculative architectural reading of arch-shaped motifs on Pillar 43 (enclosures depicted schematical…
- claimfailedThe Göbekli Tepe excavation team states there is evidence that some of the site's imagery survives into later Neolithic seal images, citing Atakuman 2015. True…
second reader rejected the anchor (quoteInContext, statementSupported): The anchored quotes only capture Oliver's caveats about 'too many missing links' and possible 'changes in meaning,' not the preceding sentence that actually asserts evidence of imagery survival via Atakuman 2015; the statement thus asserts a confi…
- claimfailedBoars are the prevalent animal species in the iconography of Enclosure C at Göbekli Tepe. True if the excavation team's account of the enclosures' emblematic s…
second reader rejected the anchor (relevant): The quote accurately reflects the source's statement about Enclosure C's boar iconography and is correctly located in the second paragraph, but this observation about a single Göbekli Tepe enclosure's dominant animal motif has no bearing on the case question concerning cro…
- claimfailedThe Göbekli Tepe excavation team lists quadruped reptiles among the animal species depicted at the site. True if a team publication includes that category in i…
second reader rejected the anchor (relevant): The quote is accurately drawn from the source's first paragraph and faithfully reflects the listed animal species, including quadruped reptiles, so it is well-supported and correctly located. However, this passage merely catalogues Göbekli Tepe's local iconographic fauna a…
- claimfailedVidale and Micheli read the wear on the Tehran chlorite 'hand-bag' as the result of long suspension on a cord. True if their published study states that infere…
second reader rejected the anchor (relevant): The quote appears verbatim in the abstract's second sentence and the record's paraphrase matches it accurately. However, this passage concerns a functional/technological inference about wear traces on a single chlorite artifact and has no bearing on the case question about…
- claimfailedVidale and Micheli read the polish on the Tehran chlorite 'hand-bag' as produced by rubbing against a soft, most probably vertical, cloth-covered surface. True…
second reader rejected the anchor (relevant): The quote and locator check out and accurately reflect the authors' own inference about wear traces on a single chlorite hand-bag, but this is a narrow functional/use-wear claim about one object with no connection to the case's question of diffusion, convergence, or retros…
- claimfailedVidale and Micheli's inference about the function of the chlorite 'hand-bag' rests, by their own statement, on a single object. True if their study says so.
second reader rejected the anchor (relevant): The quote accurately reflects the abstract's caveat about a single-object basis for the wear-trace inference, and the locator is correct, but this claim about chlorite hand-bag function has no bearing on the case question concerning cross-cultural mythic/iconographic resem…
- claimfailedThe specialist literature on the 'intercultural style' chlorite handled objects includes readings of them as weights (Muscarella 1993; Pittman 1990). True if t…
second reader rejected the anchor (quoteInContext, statementSupported, relevant): The anchor only shows bibliography entries, and the quoted fragment for Muscarella omits the scare-quoted 'Weights' in the actual title, which signals skepticism about a literal weight-function rather than endorsing it, so the quote is n…
- claimfailedOn Hoa Hakananai'a the two komari petroglyphs were carved before the birdman bas-reliefs. True if the published study places the komari earlier than the relief…
second reader rejected the anchor (relevant): The passage does establish that the komari petroglyphs predate the birdman bas-relief carvings on this one statue, but this internal carving sequence on Hoa Hakananai'a does not bear on the case's question of whether resemblances in bird-man imagery, forms and stories acro…
- claimfailedBerossus's fragment F1 credits Oannes with teaching humankind the skills of settled life, including writing, mathematics, city-building, temple-founding, law,…
not atomic (The quoted phrase appears exactly on p.4 within the Oannes passage and is used in the same sense (teaching civilised skills). The locator correctly identifies F1 on that page, and the fuller paragraph does describe Oannes teaching writing, mathematics, city-building, temple-founding, law, land division, an…
- claimfailedBerossus's fragment F3 reports a series of further man-fish beings appearing after Oannes under later antediluvian kings, including four under Daonos and one n…
second reader rejected the anchor (statementSupported): The quoted phrase about Odakon does occur exactly at the cited locator and means what the anchor claims for that detail, so quoteInContext and locatorSupported hold. However, the statement also asserts that four man-fish beings appeared under king Daonos, a separ…
- claimfailedThe Göbekli Tepe research team describes Urfa Man's gesture as both hands grasping his penis. True if the team's published description uses that wording.
second reader rejected the anchor (relevant): The quote is accurate and correctly located, and the statement fairly reports the source's description of the Urfa Man figure's posture. However, this is a narrow physical description of a single statue's gesture with no connection drawn to cross-cultural motifs like bird-…
- claimfailedThe Göbekli Tepe research team reads the V-shaped lines on Urfa Man's chest as a necklace. True if the team's published description calls them a necklace.
second reader rejected the anchor (relevant): The quote is accurately drawn from the third paragraph describing the Urfa Man statue's necklace, and the statement fairly reflects the site's own description. However, this detail about a Neolithic Anatolian statue's necklace does not bear on any of the four accounts of c…
- claimfailedUrfa Man was found during construction work at the Pre-Pottery Neolithic site of Urfa-Yeni Mahalle / Yeni Yol, not at Göbekli Tepe. True if the excavators' acc…
second reader rejected the anchor (relevant): The quote and statement accurately reflect the source's claim about the Urfa Man's findspot near Göbekli Tepe, and the locator is correct. However, this is a narrow archaeological findspot detail about a single Anatolian site and has no bearing on the case question concern…
- claimfailedThe Göbekli Tepe research team describes Urfa Man as the oldest known statue of a man. True if the team's published description makes that statement.
second reader rejected the anchor (relevant): The quote is accurately reproduced in context and the paragraph location is correct, and the statement fairly reflects the source's claim about Urfa Man being the oldest known statue of a man. However, this fact about a single Neolithic statue's age does not bear on any of…
- claimfailedNo find context was recorded for Urfa Man. True if the excavators' account states that no context was recorded.
second reader rejected the anchor (relevant): The quote accurately reflects the source's statement that no find context was recorded for the Urfa Man sculpture, and the locator is correct. However, this fact about excavation documentation for a single Anatolian statue has no bearing on the case question concerning cro…
- evidencefailedSchmidt 2006 reports a snarling predator in situ on Pillar 27, Enclosure C — part 1
every claim it cites was rejected
- evidencefailedSchmidt 2006 reports a snarling predator in situ on Pillar 27, Enclosure C — part 2
every claim it cites was rejected
- evidencefailedExcavation team proposes the three forms may depict the enclosures seen from the side
every claim it cites was rejected
- evidencefailedExcavator cites evidence that some Göbekli Tepe imagery survives into later seal images — part 1
every claim it cites was rejected
- evidencefailedExcavator cites evidence that some Göbekli Tepe imagery survives into later seal images — part 2
every claim it cites was rejected
- evidencefailedExcavator cites evidence that some Göbekli Tepe imagery survives into later seal images — part 3
every claim it cites was rejected
- evidencefailedTeam post: boars prevail in the iconography of Enclosure C — part 1
every claim it cites was rejected
- evidencefailedTeam post: boars prevail in the iconography of Enclosure C — part 4
every claim it cites was rejected
- evidencefailedTeam post lists quadruped reptiles among the species depicted
every claim it cites was rejected
- evidencefailedVidale & Micheli: polished by rubbing against a soft, probably vertical, cloth-covered surface
every claim it cites was rejected
- evidencefailedVidale & Micheli state their inference rests on a single object — part 1
every claim it cites was rejected
- evidencefailedVidale & Micheli state their inference rests on a single object — part 2
every claim it cites was rejected
- evidencefailedVidale & Micheli state their inference rests on a single object — part 3
every claim it cites was rejected
- evidencefailedVidale & Micheli's reference list includes 'weights' readings of the handled objects
every claim it cites was rejected
- evidencefailedBerossus F3: further man-fish beings under later antediluvian kings — part 2
every claim it cites was rejected
- evidencefailedBerossus F3: further man-fish beings under later antediluvian kings — part 3
every claim it cites was rejected
- evidencefailedTeam post: Urfa Man wears a V-shaped necklace
every claim it cites was rejected
- evidencefailedTeam post: Urfa Man is not from Göbekli Tepe but from Urfa-Yeni Mahalle / Yeni Yol — part 2
every claim it cites was rejected
- evidencefailedTeam post: Urfa Man is the oldest known statue of a man, slightly larger than life-size — part 1
every claim it cites was rejected
- evidencefailedTeam post: Urfa Man is the oldest known statue of a man, slightly larger than life-size — part 2
every claim it cites was rejected
- evidencefailedTeam post: no find context recorded for Urfa Man — part 1
every claim it cites was rejected
- sourcefailedEmblematic signs? On the iconography of animals at Göbekli Tepe (Tepe Telegrams, 16 August 2016)
nothing accepted cites it (§3.6: sources are not evidence by themselves)
admitted
- sourceinSRC-SCHMIDT-2006Animals and a Headless Man at Göbekli Tepe (Short Note, Neo-Lithics 2/06, pp. 38–40)
- sourceinSRC-DIETRICH-2016Of animals and a headless man. Göbekli Tepe, Pillar 43 (Tepe Telegrams, 14 October 2016, updated 5 March 2017), with the author's replies in the comments
- sourceinSRC-VIDALE-2012The Function of a Chlorite Hand-Bag of the Halil Rud Civilization as Inferred from Its Wear Traces
- sourceinSRC-VERBRUGGHE-1996Berossos and Manetho, Introduced and Translated: Native Traditions in Ancient Mesopotamia and Egypt
- sourceinSRC-GUTIRREZ-2015On the Banduddu, or Bucket (Samizdat blog, 5 July 2015): a transcription of F.A.M. Wiggermann, Mesopotamian Protective Spirits: The Ritual Texts (Groningen 199…
- claiminDEEP-C067The excavators' first published description of Pillar 43 at Göbekli Tepe (Schmidt 2006) describes three box-like objects with handle-like attachments in the pi…
- claiminDEEP-C068Each of the three box-like objects on Pillar 43's upper register is accompanied by an animal, in the excavator's description. True if Schmidt's note states tha…
- claiminDEEP-C072The Göbekli Tepe excavation team regards the extent of any survival of the site's imagery into later Near Eastern objects as unestablished because of missing i…
- claiminDEEP-C080Pitts and colleagues interpret the bas-reliefs on the back of Hoa Hakananai'a as a single composition narrating the Rapa Nui birdman myth as it was recorded in…
- claiminDEEP-C081Berossus's fragment F1, as preserved by Syncellus from Alexander Polyhistor, describes Oannes as having the whole body of a fish with a human head beneath the…
- claiminDEEP-C082Berossus's fragment F1 states that the form of Oannes had been preserved in sculpture to the writer's day. True if the fragment's text says so.
- claiminDEEP-C083In Berossus's fragment F1, Oannes returns to the sea at the end of each day and spends the night there. True if the fragment's text describes a nightly return…
- claiminDEEP-C092In short, Oannes taught men all those things conducive to a settled and civilized life.
- claiminDEEP-C086Verbrugghe and Wickersham's comparative table sets Berossus's Oannes beside U-An, the first of the apkallu, in the Uruk tablet W 20030,7. True if the table mak…
- evidenceinDEEP-E025Schmidt 2006 describes three box-like objects with handle-like attachments on Pillar 43
- evidenceinDEEP-E061Excavator: too many 'missing links' to be sure of survival or change of meaning — part 1
- evidenceinDEEP-E062Excavator: too many 'missing links' to be sure of survival or change of meaning — part 2
- evidenceinDEEP-E034Vidale & Micheli abstract: wear traces visible on the handle and along its contour
- evidenceinDEEP-E068Pitts et al. abstract: komari added, then bas-relief carvings succeeding them — part 2
- evidenceinDEEP-E069Pitts et al. abstract: komari added, then bas-relief carvings succeeding them — part 3
- evidenceinDEEP-E070Pitts et al. abstract: komari added, then bas-relief carvings succeeding them — part 3 — toward DEEP-C062
- evidenceinDEEP-E071Pitts et al. read the reliefs as a composition narrating the birdman myth as recorded in the 19th–20th centuries — part 1
- evidenceinDEEP-E072Pitts et al. read the reliefs as a composition narrating the birdman myth as recorded in the 19th–20th centuries — part 2
- evidenceinDEEP-E073Pitts et al. abstract gives no manufacture date and does not itemise the tern, paddles, ring or girdle — part 2
- evidenceinDEEP-E074Pitts et al. abstract gives no manufacture date and does not itemise the tern, paddles, ring or girdle — part 3
- evidenceinDEEP-E075Berossus F1: Oannes has the whole body of a fish with a human head beneath — toward DEEP-C053
- evidenceinDEEP-E076Berossus F1: Oannes's form 'has been preserved in sculpture to this day' — toward DEEP-C053
- evidenceinDEEP-E079Translators' table sets Oannes beside U-An in the apkallu list of Tablet W 20030,7 — part 3
- evidenceinDEEP-E080Translators' table sets Oannes beside U-An in the apkallu list of Tablet W 20030,7 — part 4
- evidenceinDEEP-E086Wiggermann 1992 (via blog transcription): banduddû denotes the bucket held by relief figures — part 1
- evidenceinDEEP-E055Wiggermann 1992 (via blog transcription): the banduddû was filled with water and used to sprinkle
- researchinDEEP-R004Test the excavators' named intermediate: do Göbekli Tepe motifs demonstrably persist into Neolithic stamp seals?
- researchinDEEP-R005Buildings or objects? Read every team statement on the three forms of Pillar 43
2026-09-19reportresearch pass (claude-fable-5-1)$3.62
run 2026-09-19-report-deep-memory-120902
This run wrote no disposition rows.
2026-09-19editionnew assessment; article 3889 → 4797 words; research agenda: 4 open, 1 answered, 0 superseded, 0 retired (1 change(s) this run)$3.41
run 2026-09-19-edition-deep-memory-125325
This run wrote no disposition rows.
2026-09-19draftschema sent as instructions (too large for strict output)16 blocked · 1 duplicate · 3 irrelevant$5.69
run 2026-09-19-draft-deep-memory-121715
refused, with the reason
- sourceblockedURL in sources with retrieved:false, reason HTTP 522.
Pausanias, Description of Greece 10.16.3 (Jones translation at Theoi): the fetch returned HTTP 522, so no text was shown; the report's statement that the Delphians call the omphalos the centre of the earth cannot be recorded.
reopen if: The passage is retrieved from Theoi, Perseus or a Loeb scan and its text shown.
- sourceblockedNamed in the report's 'Named, not opened' list; not in sources.
Wiggermann 1992, Mesopotamian Protective Spirits: The Ritual Texts, p. 66 — the primary for the banduddû; seen only through a blog transcription, which is entered separately as weak evidence.
reopen if: The printed page (or a scan) is read and the transcription checked; the blog records should then be superseded.
- sourceblockedCited in the Dietrich comment of 17 Oct 2016; not in sources.
Atakuman 2015, 'From monuments to miniatures', Cambridge Archaeological Journal 25/4: 759–788 — the intermediate class the excavator names; not retrieved.
reopen if: Retrieved (doi via Cambridge Core) and read for the research item proposed here.
- sourceblockedCited in the Pillar 43 post; not in sources.
Schmidt 2012, Göbekli Tepe: A Stone Age Sanctuary, p. 244 — cited by the team for the disc interpretation on Pillar 43; not retrieved.
reopen if: The book is obtained (ex oriente e.V., Berlin) and the page read.
- sourceblockedCited in the 'Emblematic signs?' post; not in sources.
Becker et al. 2012, 'Materialien zur Deutung der zentralen Pfeilerpaare des Göbekli Tepe', ZORA 5: 14–43 — the team's study of the central pillars (arms and hands); not retrieved.
reopen if: Retrieved and read for DEEP-C023.
- sourceblockedReport §6: could not reach.
Notroff et al. 2017, 'More than a vulture: a response to Sweatman and Tsikritsis' (Tepe Telegrams) — the fetch was refused (url_not_allowed per the report).
reopen if: The post is retrieved from the DAI blog or the Mediterranean Archaeology and Archaeometry response is read.
- sourceblockedNamed on the founder page and in the report; not in sources.
Sweatman & Tsikritsis 2017, 'Decoding Göbekli Tepe with archaeoastronomy', MAA 17(1) — the rival reading of Pillar 43; not retrieved.
reopen if: Retrieved; weigh against the team's response.
- sourceblockedReference list of Vidale & Micheli; not in sources.
Muscarella 1993, 'Intercultural Style "Weights"', Bulletin of the Asia Institute 7: 143–153 — the 'weights' classification of the chlorite handled objects; seen only as a reference-list title.
reopen if: Retrieved and read; its argument then anchors the weights claim directly.
- sourceblockedReference list of Vidale & Micheli; not in sources.
Pittman 1990, 'Steatite or Chlorite Handles Weights', in Glories of the Past (Shelby White and Leon Levy collection): 41–43 — earlier 'weights' reading; seen only as a reference-list title.
reopen if: Retrieved and read.
- sourceblockedReference list of Vidale & Micheli; not in sources.
Perrot & Madjidzadeh 2005, 'L'iconographie des vases et objets en chlorite de Jiroft', Paléorient 31(2): 123–152 — iconography of the Jiroft objects; not retrieved.
reopen if: Retrieved via Persée.
- sourceblockedJournal page lists 'PDF 909.41 K'; body not in sources.
Vidale & Micheli 2012, full text (PDF 909 KB on the journal page) — only the abstract and references were retrieved; the body carries the wear photographs and the iron-repair discussion.
reopen if: The PDF is retrieved and read; the abstract-level records can then be strengthened or corrected.
- sourceblockedCambridge Core shows 'Article purchase temporarily unavailable'.
Pitts et al. 2014, body text — paywalled; the abstract alone does not give the statue's date or itemise the tern, paddles, ring and girdle.
reopen if: Institutional access or an author preprint; or the British Museum record Oc1869,1005.1 is retrieved.
The doi.org URL resolved to the same Cambridge Core abstract page as the article URL; one source record covers both.
- sourceirrelevantRetrieved; lead paragraph and 'Research history' section read.
Wikipedia, 'Göbekli Tepe' — tertiary; the intake decision to set encyclopaedia entries aside stands. It is used only to name the primaries for the toponym claims: Symmes 2010 (Newsweek) for the 'Potbelly Hill' gloss, Kosen 2019 (Kurdistan 24) for 'Wish Hill', Zekîoğlu 2020 for the local sacred-place tradition.
reopen if: Symmes 2010 or Kosen 2019 cannot be located and the entry's citation is the only route.
- sourceblockedReport §6: could not reach.
Thomson 1891, 'Te Pito te Henua, or Easter Island' (Smithsonian Report) — earliest widely cited record of the island name; not retrieved (fetch budget).
reopen if: Retrieved from Internet Archive and read for DEEP-R002.
- sourceblockedReport §6: could not reach.
Garcilaso de la Vega, Comentarios Reales de los Incas (1609), Book 1 — the Cusco 'navel' gloss; not retrieved.
reopen if: The passage is retrieved (Spanish original and Livermore translation) and its wording recorded.
- sourceblockedNamed in the report's 'Named, not opened' list.
Strabo, Geography 9.3.6 — the usual source for the Zeus's-eagles story about the Delphic omphalos, which the declined eagles claim needs; not opened.
reopen if: Retrieved (Perseus, Jones translation) and read.
- sourceblockedNamed on the founder page and in the report; not in sources.
Lee 1992, The Rock Art of Easter Island — the corpus for komari, birdmen and paddles at Orongo (DEEP-C060–C064); not retrieved.
reopen if: Retrieved and the relevant plates read.
- sourceirrelevantReport §6: opened, first paragraphs; not in sources.
Livius.org 'Berossus' page — secondary summary; the report says it adds nothing beyond Verbrugghe & Wickersham, and its text was not shown.
reopen if: A specific statement is needed that the translation does not supply.
- sourceirrelevantReport §6: retrieved but not used; not in sources.
University of Hawaiʻi at Hilo student paper on Te Pito o te Henua and Wikipedia 'Staff God' — tertiary, retrieved by the report but not read in depth and not shown here.
reopen if: Either names a primary for DEEP-R002 or DEEP-C028 that cannot otherwise be found.
2026-09-16editionre-adopts the incumbent's assessment; article 4054 → 3849 words$1.81
run 2026-09-16-edition-deep-memory-023014
This run wrote no disposition rows.
2026-09-10editiona new edition with a new assessment$2.67
run 2026-09-10-edition-deep-memory-120824
This run wrote no disposition rows.
2026-09-10check5 of 5 seats judged the case blind$0.69
run 2026-09-10-check-deep-memory-121549
This run wrote no disposition rows.
2026-09-09verifyadmitted 47 claims, 12 evidence, 1 sources, 3 research; 31 refused63 in · 31 failed$5.57
run 2026-09-09-verify-deep-memory-204155
refused, with the reason
- evidencefailedDeep Memory home page: the visual grammar people notice across distant objects
not one observation (The quoted phrase appears in "The pattern people see" section exactly as used, and the caveat about resemblance being only a starting observation also appears there, so the passage is correctly located and fairly characterized as context rather than support for the specific object claim. However,…
- evidencefailedDeep Memory reports its catalogue experiment's bird-person frequency
not one observation (The quoted 55%/926 figure appears verbatim under [§ Bird-men] and matches the claim's proposition, and this is the primary source itself so no independence issue arises. However, the sourceStatement bundles two separate propositions: the 55% finding itself, and an added claim about what that basel…
- evidencefailedDeep Memory reports its catalogue experiment's cosmic-serpent frequency
not one observation (The sourceStatement bundles two separate propositions: the 37% quoted finding and a distinct scope-limitation claim about the result describing only the selected category and dataset. These are separable truth conditions and should be split into distinct records, even though both are individually…
- evidencefailedDeep Memory's stated caution on catalogue coverage
not one observation (The record bundles two distinct limitation points from the source—uneven catalogue coverage for the ancient Near East and Rapa Nui, and the separate point that broad narrative categories are not the same as specific images or place names—into a single sourceStatement joined by 'and'. These are ind…
- evidencefailedDeep Memory reports the V1 civiliser lineup result
not one observation (The quoted 62.5%/9% figures appear verbatim at the cited section and are used in context, and the locator is correct. However, the sourceStatement bundles two separate propositions - the V1 lineup ranking result and the unrelated fact that the V2 audit inspected 29 passages without rerunning the s…
- evidencefailedDeep Memory's stated limits on the V1 lineups
not one observation (The quoted phrase and the added claim about rarity/origin both appear at the correct locator and are used in context. However, the sourceStatement bundles two separable propositions—the recognisability/compressed-accounts limitation and the separate claim that neither V1 nor V2 establishes a world…
- evidencefailedDeep Memory reports the flood text-coding pilot's pattern
second reader rejected (locatorSupported): The quoted clustering/uneven-distribution finding is accurately drawn from the '[§ The raven & the dove]' section, but the '263 summaries' detail actually appears in a different section ('[§ What survives the water?]'), so the locator does not support the whole sourceStatemen…
- evidencefailedDeep Memory reports what the image pilots exposed
not one observation (The sourceStatement bundles two distinct quoted findings—one about altered illustrations/unreliable coding/judge disagreement and another about matching object kinds changing the lineup result—each with its own truth condition and each supporting a different claim, so it is not a single atomic pro…
- evidencefailedDeep Memory's caption on the Jiroft handbag's wear
not one observation (The quoted caption text appears verbatim in the 'The handbags' section and is used in its original sense describing wear on the Jiroft chlorite object's handle, and the section does link to the Halil Rud wear study, so the quote, context, and locator all check out. However, the sourceStatement bun…
- evidencefailedDeep Memory's caption on the site names of Göbekli Tepe
not one observation (The quoted caption appears verbatim under 'The navel of the world' section and is used as a caption assertion, consistent with its context of caution about etymology sourcing. However, the sourceStatement bundles two distinct propositions—the Turkish name's descriptive meaning (bearing on C015) an…
- claimfailedBird-headed figures, handled shapes and hands-resting-on-the-body poses recur on objects from cultures separated by great distance and time (Göbekli Tepe, Nimr…
second reader rejected the anchor (statementSupported): The anchored passage in 'The pattern people see' only makes a general rhetorical observation that distant objects can share a familiar visual grammar, illustrated by four objects, but it does not establish that each of the three distinct features (bird-headed fig…
- claimfailedA relational comparison of a profile head, a bent or extended limb, and a rounded form nearby is more specific than noting a bird in both pictures. True if the…
second reader rejected the anchor (statementSupported): The quoted passage does appear at the stated locator and does support the idea that the three-feature relational comparison is more specific than merely noting a bird in both images. However, the added falsifiability clause ('True if the co-occurrence of those th…
- claimfailedThe top register of Pillar 43 at Göbekli Tepe carries three arched handled forms with small animals. True if the excavators' record of the pillar shows them.
second reader rejected the anchor (statementSupported): The quoted caption does appear verbatim at that locator and describes three arched objects with animals, but it comes from a secondary compilation site, not the excavation team's own publication, and the site itself points readers to a separate 'excavation team's…
- claimfailedThe Assyrian banduddû is a bucket named in ritual texts, carried with a cone used to sprinkle from it. True if cuneiform ritual texts name the bucket and descr…
not atomic (The quote appears verbatim in the '[§ The handbags]' section and is used consistently with its context. However the statement bundles two separate claims—that the bucket is named in ritual texts, and that it is carried with a cone used for sprinkling—each of which could be true or false independently, so i…
- claimfailedA handle is a practical solution that can recur independently in unconnected cultures. True if handled containers appear in cultures with no demonstrable conta…
second reader rejected the anchor (statementSupported): The quoted sentence only asserts that a handle is a practical solution that can recur independently, but the record adds an unsupported operational truth-condition ('True if handled containers appear in cultures with no demonstrable contact') that the source neve…
- claimfailedThe Göbekli Tepe excavation team's discussion of Enclosure C distinguishes boar sculptures from snarling predators and considers their arrangement at the entra…
not atomic (The quoted phrase appears essentially verbatim at the cited locator and the claim anchor is properly placed, but the statement bundles two distinct propositions—that the excavation team distinguishes boars from snarling predators, and separately that they consider the arrangement of these forms together—jo…
- claimfailedShared animal anatomy and carving choices can each contribute to the visual resemblance between pig or boar sculptures from distant cultures. True if a control…
second reader rejected the anchor (statementSupported): The quote about both anatomy and carving choices contributing is accurately drawn from the pig section, but the source never proposes the specific falsification criterion (a control set of pig sculptures sharing low body, raised back and snout regardless of cultu…
- claimfailedThe netted omphalos stone at Delphi marked the sanctuary as the meeting point of Zeus's eagles. True if ancient Greek sources describe the Delphic omphalos in…
second reader rejected the anchor (quoteInContext, statementSupported): The quoted caption merely describes the omphalos image without attributing the claim to ancient Greek textual sources; the surrounding section explicitly treats the omphalos's meaning and dating as an open question needing its own source history,…
- claimfailedCarving limbs against a stone body protects them from breakage, so material constraint is a plausible contributor to the hands-on-body pose of stone figures. T…
not atomic (The anchored quote appears verbatim at the cited section and is used in the same qualified sense as the source (a plausible, non-settling contributor to the pose). However, the statement bundles two separable propositions: a general mechanical claim about breakage protection and a specific empirical truth-…
- claimfailedThe hands of the Ahu Tongariki moai meet on the abdomen above the belt, following the Polynesian tiki convention. True if the statues show that placement and t…
second reader rejected the anchor (statementSupported): The quote is verbatim and accurately describes the Ahu Tongariki moai's hand placement and the caption's tiki label, but the source only asserts this label as a caption without providing evidence that the convention is independently attested elsewhere in Polynesi…
- claimfailedThe Ponce monolith at Tiwanaku holds its hands to the chest, holding a beaker and a snuff tablet. True if the monolith and its published descriptions show thos…
not atomic (The quoted caption matches the source verbatim and is correctly located under the 'Hands on the belly' section, and it bears on the case's 'handled objects' comparison theme. However, the statement bundles three separate factual claims (hand position at the chest, holding a beaker, holding a snuff tablet)…
- claimfailedThe fish-cloaked apkallu is the sage type named in Assyrian cuneiform texts, and Berossus's Oannes recalls it. True if the texts name fish-cloaked sages as apk…
not atomic (The caption quote does appear verbatim at the cited section and its content matches the claim's gist about the fish-cloaked apkallu being the type the texts name and Oannes recalling it, so the anchor is properly located and supported. However, the statement bundles two separable propositions—the identific…
- claimfailedA colonial recording date for a civiliser story is a reason to examine chronicler mediation, not proof that every detail was invented by the chronicler. True i…
second reader rejected the anchor (statementSupported): The quoted passage only states that a colonial recording date is grounds to examine mediation rather than proof of invention; it does not itself assert or establish the added conditional clause about details being independently attested earlier or in indigenous m…
- claimfailedIn the same pilot, details associated with Genesis are unevenly distributed across the coded flood summaries while generic flood motifs are widespread. True if…
not atomic (The quote appears verbatim in the 'raven & the dove' section and is used consistently with its context describing the pilot's findings; the locator correctly identifies that section, and the proposition bears on the case's flood-story comparison. However, the statement bundles two separable claims—uneven d…
- claimfailedA proposed connection between distant motifs is more informative when there are dated intermediates, a plausible route, or distinctive details that occur toget…
second reader rejected the anchor (statementSupported): The quoted phrase appears verbatim in the cited section and matches its context, but the record adds a further empirical criterion ('True if cases with those features have historically been confirmed more often than cases without them') that the source never stat…
- claimfailedIn the Deep Memory catalogue experiment, the three broad motif classes (birds, serpents, sacred centres) are each widespread and their co-occurrence does not s…
not atomic (The quote is accurately drawn from the 'Birds, serpents & sacred centres' section and reflects the catalogue experiment's finding as described, and it bears on the case question about shared past versus independent invention. However, the statement bundles two separable claims — that each motif class is in…
- claimfailedThe Hoa Hakananai'a statue dates to c. 1000–1200 CE and its back carvings were added later. True if the published study of the statue dates the carving of the…
not atomic (The quote does appear in the caption at the given locator and states the dating and later carving claim accurately. However, the record bundles two distinct propositions: the factual dating/sequencing claim itself, and a separate verification condition referencing 'the published study' (presumably Pitts et…
- claimfailedThe back of Hoa Hakananai'a carries two facing birdmen with a sooty tern between their beaks, komari, paddles, a ring and a girdle. True if the statue's publis…
not atomic (The quote is verbatim from the 'Bird-men across the gallery' section and accurately describes the back of Hoa Hakananai'a, matching the anchor's claim and locator. However, the statement bundles six separate iconographic features (birdmen, tern, komari, paddles, ring, girdle) into one proposition, each of…
- claimfailedShape and function of the handled objects are separate questions: Assyrian ritual context supports a bucket-and-purifier reading of the banduddû while the Göbe…
not atomic (The quotes accurately reflect the source's claim that Assyrian ritual context supports a bucket-and-purifier reading of the banduddû while the Göbekli forms' function is undetermined, and both are correctly located in the 'Handbags' section. However, the proposition bundles several distinct assertions—that…
- evidencefailedDeep Memory home page: the visual grammar people notice across distant objects — part 1
every claim it cites was rejected
- evidencefailedDeep Memory home page: the visual grammar people notice across distant objects — part 2
every claim it cites was rejected
admitted
- sourceinSRC-JHONG-2026Bird-men, handbags and the navel of the world — Deep Memory, the home page: five investigations and the illustrated narrative
- claiminDEEP-C002The historical meaning of a visual resemblance between objects depends on their dates, contexts, differences and possible connections, not on the resemblance a…
- claiminDEEP-C003Olmec Monument 19 from La Venta is the earliest known depiction of a feathered serpent. True if no securely dated feathered-serpent image predates it.
- claiminDEEP-C004The description 'bird-man' groups at least three distinct ideas — a deity with a bird's head, a person wearing feathers, and a person transforming into a bird.…
- claiminDEEP-C005In the Deep Memory catalogue experiment, a broad bird-person motif class occurs in 55% of 926 sampled traditions. True if the experiment's coded data reproduce…
- claiminDEEP-C044Assyrian ritual context supports a bucket-and-purifier interpretation of the banduddû.
- claiminDEEP-C009The Jiroft chlorite 'handbag' in the National Museum of Iran shows wear on its handle consistent with having hung on a cord. True if the published wear-trace s…
- claiminDEEP-C010The small handled object held by the seated figure on La Venta Monument 19 is usually read by specialists as an incense bag. True if the standard Olmec literat…
- claiminDEEP-C012The function of the three handled forms on Pillar 43 is uncertain. True if the excavation team has published no secure functional identification.
- claiminDEEP-C045The Göbekli Tepe excavation team's discussion of Enclosure C distinguishes boars from snarling predators.
- claiminDEEP-C015The modern Turkish name Göbekli Tepe ('navel hill') describes the shape of the mound. True if the toponym is documented as descriptive of the mound's form rath…
- claiminDEEP-C016The Kurdish names for the Göbekli Tepe site mean 'wish hill'. True if a Kurdish toponymic or ethnographic source records that meaning.
- claiminDEEP-C018The Greek omphalos, Garcilaso's gloss of Cusco, the Rapanui Te Pito o te Henua and the modern Turkish name Göbekli Tepe are different kinds of evidence, differ…
- claiminDEEP-C019The Rapanui name Te Pito Kura means 'navel of light'. True if Rapanui lexical sources give that gloss.
- claiminDEEP-C047Material constraints are a plausible contributor to the hands-on-body pose of stone figures.
- claiminDEEP-C048Material constraint as a contributor to the pose does not settle its symbolic meaning.
- claiminDEEP-C021The Urfa Man statue (Şanlıurfa, c. 9000 BCE) holds its hands over the genitals. True if the statue as published shows that gesture.
- claiminDEEP-C023Arms and hands are carved on the central T-pillars of Enclosure D at Göbekli Tepe. True if the excavation record shows arms and hands on those pillars.
- claiminDEEP-C025In the Deep Memory catalogue analysis, a broad cosmic-serpent motif class occurs in 37% of sampled traditions. True if the experiment's coded data reproduce th…
- claiminDEEP-C026A catalogue's geographical distribution of a broad motif class cannot by itself distinguish independent invention from very old transmission. True if the same…
- claiminDEEP-C027In the Deep Memory V1 civiliser lineup experiment, the claimed partner figure ranked first in 62.5% of completed lineups, against 9% for random pairs. True if…
- claiminDEEP-C028The identification of the central staff-bearing figure on the Gate of the Sun at Tiwanaku as Viracocha is a modern one. True if no pre-modern source names that…
- claiminDEEP-C052The fish-cloaked sage depicted with Ea in first-millennium BCE Assyria is the apkallu type that the texts actually name.
- claiminDEEP-C031In the Deep Memory flood text-coding pilot, the coded details of 263 flood-story summaries cluster by region. True if the pilot's coded data show regional clus…
- claiminDEEP-C054In the text-coding pilot, the coded flood summaries show an uneven distribution of details associated with Genesis.
- claiminDEEP-C055In the text-coding pilot, generic flood motifs are widespread across the coded flood summaries.
- claiminDEEP-C034Stories can become more alike than their originals through translation, borrowing and comparative writing. True if documented retellings show convergence of de…
- claiminDEEP-C035Similarity between motifs alone cannot establish a lost-civilisation account's claims about the source's identity, date and capabilities. True if those claims…
- claiminDEEP-C036A visual resemblance between signs of two undeciphered scripts does not establish shared sound, meaning or descent. True if sign-shape similarity can arise bet…
- claiminDEEP-C037In the Deep Memory image pilots, AI judges disagreed strongly with one another about the resemblance between the same objects. True if the pilots' inter-judge…
- claiminDEEP-C038In the Deep Memory image pilots, matching the kinds of objects in a lineup changed the lineup result. True if the pilot data show a different ranking under kin…
- claiminDEEP-C056In the Deep Memory catalogue experiment, the three broad motif classes (birds, serpents, sacred centres) are each widespread in the sampled catalogue.
- claiminDEEP-C057In the Deep Memory catalogue experiment, the bundle of the three broad motif classes does not stand out against the study's documentation-preserving comparison…
- claiminDEEP-C059The birdman carvings on the back of Hoa Hakananai'a were added later than the statue's manufacture.
- researchinDEEP-R001Retrieve the primary literature the page leans on for the handled objects
- researchinDEEP-R002Source history of the four 'navel' names
- researchinDEEP-R003Blind human lineup test of the civiliser accounts from original passages
2026-09-09inbox1 item(s) taken in$0.04
run 2026-09-09-inbox-deep-memory-200334
This run wrote no disposition rows.
2026-09-09editiona new edition with a new assessment$2.85
run 2026-09-09-edition-deep-memory-210033
This run wrote no disposition rows.
2026-09-09draftschema sent as instructions (too large for strict output)25 blocked · 1 irrelevant$1.59
run 2026-09-09-draft-deep-memory-203731
refused, with the reason
- sourceblockedNo text in `sources`.
Pitts, Miles, Pagi & Earl 2014, Antiquaries Journal 94: named by the page, no OpenAlex result and no retrieved text; nothing quoted.
reopen if: Full text retrieved from Cambridge Core (Antiquaries Journal 94, 2014) or an author repository copy.
- sourceblockedNo text in `sources`.
Notroff et al. 2016, 'Of animals and a headless man: Göbekli Tepe, Pillar 43' (Tepe Telegrams blog): no retrieved text.
reopen if: Blog post retrieved from tepetelegrams.wordpress.com or the Wayback Machine.
- sourceblockedNo text in `sources`.
Notroff et al. 2017, 'More than a vulture: a response to Sweatman and Tsikritsis': no retrieved text.
reopen if: Text retrieved (Tepe Telegrams post or the Mediterranean Archaeology and Archaeometry 17(2) response).
- sourceblockedNo text in `sources`.
Sweatman & Tsikritsis 2017, 'Decoding Göbekli Tepe with archaeoastronomy', Mediterranean Archaeology and Archaeometry 17(1): no retrieved text.
reopen if: Open-access PDF retrieved from the journal site (maajournal.com).
- sourceblockedNo text in `sources`.
Vidale & Micheli, chlorite hand-bag wear-trace study, Iranian Journal of Archaeological Studies: no retrieved text.
reopen if: PDF retrieved from the journal (University of Sistan and Baluchestan) or the authors' repository.
- sourceblockedNo text in `sources`.
Lee 1992, The Rock Art of Easter Island (UCLA): book, no retrieved text.
reopen if: Relevant chapter on Orongo birdman petroglyphs retrieved (library scan or escholarship).
- sourceblockedNo text in `sources`.
'Ahu Te Pito Kura' (Imagina Rapa Nui tourism page): webpage not retrieved; a tourism site is at best tertiary for the name's meaning.
reopen if: Page retrieved; even then prefer a Rapanui lexicon for the gloss.
- sourceblockedNo text in `sources`.
Berossus, Babyloniaca, Oannes passage (via Alexander Polyhistor and Eusebius): no retrieved text.
reopen if: Text retrieved from a critical edition/translation (Verbrugghe & Wickersham 1996, or Eusebius Chronicon / Syncellus).
- sourceblockedNo text in `sources`.
Townsend 2003, 'Burying the White Gods', American Historical Review 108(3): no retrieved text.
reopen if: Article retrieved via Oxford Academic / JSTOR.
- sourceblockedNo text in `sources`.
Restall 2003, Seven Myths of the Spanish Conquest: book, no retrieved text.
reopen if: Relevant chapter (the 'white gods'/apotheosis myth) retrieved.
- sourceblockedNo text in `sources`; page gives author and year only.
Cobo Betancourt 2024 on Bochica and colonial reinterpretation: citation on the page is incomplete (no title, venue); no OpenAlex result.
reopen if: The work is identified (title, venue) and its text retrieved.
- sourceblockedNo text in `sources`.
Berezkin & Duvakin, Thematic classification and areal distribution of folklore-mythological motifs (ruthenia.ru): catalogue not retrieved.
reopen if: Catalogue pages for the bird-person, cosmic-serpent and world-axis motifs retrieved from ruthenia.ru/folklore/berezkin.
- sourceblockedNo text in `sources`.
d'Huy 2016, Scientific American, 'Scientists trace society's myths to primordial origins': popular article not retrieved; secondary to d'Huy's phylogenetic papers.
reopen if: Article retrieved; prefer the underlying peer-reviewed papers.
- sourceblockedNo text in `sources`.
Witzel 2012, The Origins of the World's Mythologies: book, no retrieved text.
reopen if: Relevant chapters (Laurasian mythology, flood, bird motifs) retrieved.
- sourceblockedNo text in `sources`.
Eliade 1959, The Sacred and the Profane: book, no retrieved text.
reopen if: Chapter on sacred space / axis mundi retrieved.
- sourceblockedNo text in `sources`.
Hancock 2015, Magicians of the Gods: book, no retrieved text; the page cites it for the handbag-as-badge-of-office, Göbekli–Easter Island and navel-site claims.
reopen if: Relevant pages retrieved so the lost-source account can be stated in its author's words.
- sourceblockedNo text in `sources`.
Newman, 'Megalithic origins: ancient connections between Göbekli Tepe and Peru' (grahamhancock.com): webpage not retrieved.
reopen if: Page retrieved from grahamhancock.com or the Wayback Machine.
- sourceblockedNo text in `sources`.
Heyerdahl 1952, American Indians in the Pacific: book, no retrieved text.
reopen if: Relevant chapters retrieved (archive.org lending copy).
- sourceblockedNo text in `sources`.
Sitchin 1976, The 12th Planet: book, no retrieved text.
reopen if: Relevant pages retrieved so the account can be quoted rather than paraphrased.
- sourceblockedNo text in `sources`.
Elliot Smith 1923, The Ancient Egyptians and the Origin of Civilization: public-domain book, no retrieved text.
reopen if: Scan retrieved from archive.org.
- sourceblockedNo text in `sources`.
Dibble & Hancock, Joe Rogan Experience #2136 (April 2024): podcast, no transcript retrieved; the Archaeology Southwest summaries the page mentions are also unretrieved.
reopen if: A transcript or the Archaeology Southwest summary page is retrieved.
- sourceblockedNo text in `sources`.
Ioannidis et al. 2020, Nature 583, Native American gene flow into Polynesia: no retrieved text.
reopen if: Paper or PMC author manuscript retrieved.
- sourceblockedNo text in `sources`.
Fehren-Schmitz et al. 2017, Current Biology 27, Rapanui genetic ancestry: no retrieved text.
reopen if: Open-access paper retrieved from Cell Press.
- sourceblockedNo text in `sources`.
Holliday et al. 2023, Earth-Science Reviews 247, refutation of the Younger Dryas impact hypothesis: no retrieved text.
reopen if: Paper retrieved (open access at Elsevier).
- sourceblockedNo text in `sources`.
Timeline Auctions, 'The handbag that wasn't': commercial blog, not retrieved; tertiary at best.
reopen if: Page retrieved; weigh as a secondary commentary only.
- sourceirrelevantNot in the resolved list; not retrieved.
The page's 'Wikipedia and Wikimedia Commons entries' (Göbekli Tepe, Hoa Hakananai'a, Apkallu, La Venta, Cacaxtla, Etowah plates, Omphalos, Tangata manu, Axis mundi, Hyperdiffusionism, White gods) are tertiary encyclopaedic summaries; the ledger wants the primaries they cite, not the entries.
reopen if: A specific entry's cited primary is needed and cannot otherwise be located.
2026-09-09check5 of 5 seats judged the case blind$0.77
run 2026-09-09-check-deep-memory-210940
This run wrote no disposition rows.
Change history
Every change to this case, newest first: what changed, why, and who made it — including the AI's role. The log is append-only; a correction is a new entry, never an edit to an old one.
- aletheia edition (claude-fable-5-1) · AI-assisted
Research agenda: DEEP-R001 open → answered (The Tepe Telegrams discussion of Pillar 43 and the Vidale & Micheli abstract were retrieved and read. DEEP-E028 records the team's statement that no final decision on the meaning of the three forms is possible, deciding DEEP-C012 (well supported); DEEP-E025 records Schmidt's 'box-like objects with handle-like attachments' (DEEP-C067); DEEP-E035 records the authors' reading of the Jiroft wear as long suspension on a cord and DEEP-E038 their single-object caveat, deciding DEEP-C009 (provisionally supported at abstract level). The paper's body, Sweatman & Tsikritsis 2017 and the team's 2017 response remain unread; the full run of team statements is DEEP-R005's target and the Vidale body is a blocked source.).
Why: Set by the edition verb (claude-fable-5-1, edition-v10) in run 2026-09-19-edition-deep-memory-125325, reading the agenda against the ledger as it stands; the edition edition-2026-09-19-130205 carries the reasoning.
- aletheia verify (claude-sonnet-5 second reader; drafter claude-fable-5-1) · AI-assisted
Intake from report proposals/2026-09-19-report-deep-memory-120902/report.md: 7 source(s), 25 evidence record(s), 9 claim(s), 2 research item(s) verified and added (proposal 2026-09-19-draft-deep-memory-121715, verification 2026-09-19-verify-deep-memory-122700); 55 candidate(s) rejected with reasons in dispositions.yaml.
Why: Admitted after verification: SRC-SCHMIDT-2006, SRC-DIETRICH-2016, SRC-VIDALE-2012, SRC-PITTS-2014, SRC-VERBRUGGHE-1996, SRC-DIETRICH-2016B, SRC-GUTIRREZ-2015, DEEP-E025, DEEP-E026, DEEP-E028, DEEP-E061, DEEP-E062, DEEP-E034, DEEP-E035, DEEP-E038, DEEP-E068, DEEP-E069, DEEP-E070, DEEP-E071, DEEP-E072, DEEP-E073, DEEP-E074, DEEP-E043, DEEP-E075, DEEP-E044, DEEP-E076, DEEP-E046, DEEP-E079, DEEP-E080, DEEP-E081, DEEP-E086, DEEP-E055, DEEP-C067, DEEP-C068, DEEP-C072, DEEP-C080, DEEP-C081, DEEP-C082, DEEP-C083, DEEP-C092, DEEP-C086, DEEP-R004, DEEP-R005; everything else proposed was refused or blocked with a reason in dispositions.yaml. The drafter's rationale for the proposal, written before verification and describing what it proposed: The report's bounded pass reached seven texts the ledger had never held: the excavators' own note on Pillar 43 (Schmidt 2006), three Tepe Telegrams posts by the excavation team (Pillar 43 with its comments, the enclosure animal emblems, Urfa Man), the abstract and reference list of the Jiroft wear-trace study, the abstract of the Hoa Hakananai'a study, and Berossus's fragments in the Verbrugghe–Wickersham translation; it also reached a blog transcription of Wiggermann 1992 on the banduddû. Each is proposed as a source with evidence records extracted only from the retrieved text, one observation per record, and with new observation-rung claims where the text states a proposition the catalogue lacks (the excavators' 'handle-like attachments' and their rival building reading of the three forms; the Jiroft wear read as hanging against a cloth-covered vertical surface, on one object, with a 'weights' literature behind it; the Rapa Nui sequence komari → birdmen read against ethnohistoric myth; Berossus's fish-bodied, nightly-commuting Oannes in a series of fish-men; Urfa Man's grasping hands, necklace and findspot). The Wiggermann transcription enters as weak support with its secondhand status stated, because a verified observation is not dropped silently; the primary is blocked. Pausanias was not retrievable (HTTP 522) and is blocked; the Wikipedia article is set aside as tertiary, used only to name Symmes 2010 and Kosen 2019 as the route for the toponym claims. No verdicts are proposed; grading belongs to assessments. An edition candidate is offered because the current article's load-bearing sentence — that none of the primary literature the page leans on has been read — is no longer true, and its handled-objects, Rapa Nui, civiliser and hands sections now have primary statements to stand beside the captions.
- aletheia verify (claude-sonnet-5 second reader; drafter claude-fable-5-1; proposal 2026-09-19-draft-deep-memory-121715, verification 2026-09-19-verify-deep-memory-122700) · AI-assisted
Correction applied from proposals/2026-09-19-draft-deep-memory-121715: IMG-DEEP-P07.depicts — "In the photograph: the head and shoulders of a pale limestone statue against a dark background — a smooth ovoid head, deep eye sockets holding dark inlays, a broken nose, no mouth, V-shaped lines carved on the chest; the arms and hands are out of frame. The Commons file titles it Urfa Man in the Urfa museum, September 2019." → "In the photograph: the head and shoulders of a pale limestone statue against a dark background — a smooth ovoid head, deep eye sockets holding dark inlays, a broken nose, no mouth, V-shaped lines carved on the chest (read by the excavation team as a necklace); the arms and hands are out of frame. The Commons file titles it Urfa Man in the Urfa museum, September 2019; the excavation team records the statue's findspot as Urfa-Yeni Mahalle / Yeni Yol, not Göbekli Tepe.".
Why: IMG-DEEP-P07: The team's published description ('The Urfa Man', Tepe Telegrams 2016) reads the V-shaped lines as a necklace and states the statue is not from Göbekli Tepe; the description should carry both so the plate is not read as a Göbekli Tepe object.
- claude-fable-5-1 (operator), answering review note #302 in the founder's session of 2026-09-16 · AI-assisted
Run id and prompt version removed from six earlier operator entries (2026-09-09 to 2026-09-11, answering review notes #237, #242, #244, #247, #263 and #285). PR #291 had stamped every operator entry in this file with the session id and prompt version of the 2026-09-16 amendments; those six were written before the convention existed, and two of them say in their own text that no run id and no prompt version exist for them. The stamps stay on the four entries of 2026-09-16, whose amendments they describe.
Why: Review note #302 (GPT-5.6 Sol seat on #291, §3.8, §3.14, §3.15): a stamp that contradicts the entry it sits on lets no reader reconstruct how the change was made. The convention of #300 — by-hand work carries the operator session as its run id and operator-by-hand-v1 as its prompt version — applies from 2026-09-16 forward, not to entries that predate it.
- claude-fable-5-1 (operator), answering review note #297 in the founder's session of 2026-09-16 · AI-assisted
Candidate edition edition-2026-09-16-023358: one conclusion sentence calibrated before publication — from "Read strictly, the ledger says this. The broad classes — bird-people, cosmic serpents, sacred centres — are co…" to a form that says the observations are consistent with the conclusion rather than establishing it, matching the uncertainty the same article states earlier.
Why: Review note #297 (GPT-5.6 Sol seat on #291, §3.6, §3.9, §3.13): the article's own qualifications (a claim supported only indirectly; a second-hand citation; an untested bedrock) were not carried into its conclusion.
- claude-fable-5-1 (operator), answering review note #294 in the founder's session of 2026-09-16 · AI-assisted
Candidate edition edition-2026-09-16-023358: the measured-count annotation now states both figures and why they differ — 4,054 → 3,849 words for the verb's candidate (its run record), 3,859 after the operator's three-sentence amendment for review note #292 — instead of the single 3,859 that contradicted the run record.
Why: The GPT-5.6 Sol seat's objection at commit 07e5b65 (review note #294, continued): two totals for one article with no stated convention cannot be reconstructed. Same counting rule, two moments; both now written down.
- claude-fable-5-1 (operator), answering review note #294 in the founder's session of 2026-09-16 · AI-assisted
Candidate edition edition-2026-09-16-023358: rationale annotated before publication with the measured word count (4,054 → 3,859, about five per cent shorter); the verb's 'roughly a sixth (about 3,700 to about 3,150)' was an estimate. The run record carries the measured figures.
Why: Review note #294 (GPT-5.6 Sol seat on #291, §3.14, §3.15): a rationale that misstates what changed cannot reconstruct the change; the measured counts are appended and labelled as the operator's.
- claude-fable-5-1 (operator), answering review note #292 in the founder's session of 2026-09-16 · AI-assisted
Candidate edition edition-2026-09-16-023358 amended before publication: three sentences of the article that carried the V1 lineup's stated confounds (recognisable lineups, compressed modern accounts) as 'the page's own statement' — propositions whose proposed evidence record was declined at intake as non-atomic — now say only what the ledger holds (DEEP-C027, the measured lineup result) and what the retelling account predicts, and point to the blind human lineup (DEEP-R003) as the test. The verb's rationale stands as written; the file header records the amendment.
Why: Review note #292 (GPT-5.6 Sol seat on #291, §3.2, §3.6, §3.15): a source statement is not an evidence record, and material whose record was rejected does not re-enter the synthesis through edition prose, however labelled. Accepted: the article says what the ledger holds; the confounds enter, if at all, as atomic evidence records through intake.
- claude-fable-5-1 (operator), answering review note #285 in the founder's session of 2026-09-11 · AI-assisted
Correction applied from review note #285 (GPT-5.6 Sol seat on #283): the dangling-id check must hold for every record: DEEP-E023.limitations — ["Secondary paraphrase of an unretrieved paper.","Second reader (claude-sonnet-5, 2026-09-09) disputes the stated direction: The quoted phrase 'the Halil Rud wear study' and the Vidale & Micheli citation appear exactly as described in the 'The handbags' section, and the record accurately reports only that the source points to this study rather than asserting its findings. However, merely citing that a wear study exists does not itself support the specific claim that the study reports handle wear consistent with hanging on a cord — since the underlying paper was not retrieved, this observation is more properly 'context' pointing toward where evidence might be found rather than 'supports' the claim outright. The origin field appropriately notes this is a split of DEEP-E009, satisfying independence."] → ["Secondary paraphrase of an unretrieved paper.","Second reader (claude-sonnet-5, 2026-09-09) disputes the stated direction: The quoted phrase 'the Halil Rud wear study' and the Vidale & Micheli citation appear exactly as described in the 'The handbags' section, and the record accurately reports only that the source points to this study rather than asserting its findings. However, merely citing that a wear study exists does not itself support the specific claim that the study reports handle wear consistent with hanging on a cord — since the underlying paper was not retrieved, this observation is more properly 'context' pointing toward where evidence might be found rather than 'supports' the claim outright. The origin field appropriately notes this is a split of the proposed record 'Deep Memory's caption on the Jiroft handbag's wear', whose parts entered separately, satisfying independence."].
Why: DEEP-E023: prose named a record id the ledger does not hold — the parent record's provisional id never entered the ledger (review note #285: no dangling ids, everywhere)
- claude-fable-5-1 (operator), answering review note #263 on the record, in the Claude Code session session_01AiXD4y5Y91KC7iM3uBhWmu; PR #265 · AI-assisted
Run links completed on the records of the sitting of 2026-09-10 (PR #257): the edition edition-2026-09-10-121547 and the five check records of 121549 now carry producedBy — the run that wrote each (2026-09-10-edition-deep-memory-120824 and 2026-09-10-check-deep-memory-121549) — and those two run records list what they wrote. The adopted assessment 2026-09-10-edition-121547 is left as written: its content hash is what its edition adopts, and an edited overlay may not change the hash beneath an essay; its run is the edition's, named on the edition and in the assessment file's own header.
Why: Review note #263 (the GPT seat on #257, §3.8, §3.14, §3.15): the records carried ids that did not meet the run record's. The rule that fixes this for every later run merged the same day (PR #258, producedBy and wrote); these records were written before it and get the same links by hand, each a fact read from the run records and the PR.
- claude-fable-5-1 (operator), answering review note #247 on the record, in the private Claude Code session session_01AiXD4y5Y91KC7iM3uBhWmu; PR #250 · AI-assisted
Correction to the entry of 2026-09-09 that answered review note #244: its reason said the PR and commit were named in its actor line, and the actor line named only a branch (caption-record). The references are PR #246, merged as commit b46b3798bd5f9a1318350648d766b94dcc8ccf71; the operator's working session, a private Claude Code session with the id session_01AiXD4y5Y91KC7iM3uBhWmu, which every commit of that work carries as its Claude-Session trailer (the session page itself is not public; the trailer is the public record of the id); the model, claude-fable-5-1; the date, 2026-09-09. No run id and no prompt version exist for it because no protocol ran: the pipeline's runs carry those stamps in proposals/<runId>/run.yaml, and operator work carries the PR, the commit, the session, the model and the date.
Why: Review note #247 (the GPT seat on #246, §3.14, §3.15): an entry must name the references it claims to name, and a branch name is not a stable one. From this entry on, by-hand entries name the PR number and the session, and the number is written in before the change merges.
- claude-fable-5-1 (operator), applying the corrections by hand with the correction writer (applyCorrections, src/pipeline/ledger-write.ts) on review note #248 (the GPT seat on #245), in the private Claude Code session session_01AiXD4y5Y91KC7iM3uBhWmu; no run id or prompt version, since no protocol ran; the verifier had recorded the second reader's direction disputes on the admitted records instead of applying them; PR #250 · AI-assisted
Corrections applied from review note #248: DEEP-E015.direction — "supports" → "qualifies"; DEEP-E017.direction — "supports" → "qualifies"; DEEP-E022.claimIds — "DEEP-C037,DEEP-C038" → "DEEP-C038"; DEEP-E023.direction — "supports" → "context"; DEEP-E024.claimIds — "DEEP-C015,DEEP-C016" → "DEEP-C016".
Why: DEEP-E015: the passage scopes what the 55% figure shows; it neither confirms nor denies the proportion DEEP-C005 states — the second reader said so at intake, and the direction now says it | DEEP-E017: the passage limits the geographic result to the sampled catalogue and category; it does not assert the 37% figure DEEP-C025 names — qualifies, as the second reader said | DEEP-E022: the quoted line is about kind-matched lineups (DEEP-C038) and says nothing about inter-judge disagreement (DEEP-C037); the link to C037 is dropped, as the second reader said | DEEP-E023: a pointer to an unretrieved wear study is not evidence of what that study found (AGENTS.md 3.6); context until Vidale & Micheli is retrieved, as the second reader said | DEEP-E024: the quoted caption states the Kurdish names (DEEP-C016) and nothing about the Turkish name (DEEP-C015); the link to C015 is dropped, as the second reader said
- claude-fable-5-1 (operator), on the founder's direction in session 2026-09-09; PR: lab-cover · AI-assisted
Cover replaced at the founder's request with the Deep Memory cover he made for his lab site: AI-generated editorial artwork in the house style (style-v2; OpenAI's built-in image generator, model identifier not exposed; run deep-memory-cover-2026-09-06), copied unchanged from github.com/ejhong/aletheia-lab at commit 5dd8e043 (md5 2dccb3e0f6fd5a37b9388dedb293b2ea), with its prompt on the record. The gpt-image-1 cover of the same day (run cover-style-v2-2026-09-09) is retired; its record leaves the register with it.
Why: Founder direction (2026-09-09, in session): he prefers the lab site's picture, and confirmed it is AI-generated, so it may serve as a cover under docs/IMAGE_STYLE.md (generated imagery is confined to editorial roles and credited as such; it depicts an idea, not evidence).
- aletheia verify (claude-sonnet-5 second reader; drafter claude-fable-5-1) · AI-assisted
Intake from report proposals/2026-09-09-inbox-deep-memory-200334/report.md: 1 source(s), 12 evidence record(s), 47 claim(s), 3 research item(s) verified and added (proposal 2026-09-09-draft-deep-memory-203731, verification 2026-09-09-verify-deep-memory-204155); 31 candidate(s) rejected with reasons in dispositions.yaml.
Why: Inbox intake of the founder's Deep Memory home page, flagged by the intake as NEW TO THE LEDGER AND SUPPLIED BY ITS AUTHOR. I propose it as one Source record (a self-published, AI-assisted web essay by the founder, carrying the recorded permission line verbatim) and catalogue its distinct testable propositions as claims anchored to that source by [§ …] section and verbatim quote — unfeatured, ungraded, for the editions to weigh. Because the source is the founder's own essay, evidence records on it are weak `supports`, `qualifies` or `context`, with limitations saying so. Every one of the 25 works the page names came back with no OpenAlex result and no retrieved text is in `sources`, so each is dispositioned `blocked` with the route to its primary; nothing is quoted from any of them. No corrections (the ledger holds no records to correct). No edition candidate: the record deepens without the telling changing, though the article's line that no founding page has yet entered will need refreshing once this source is admitted.
- claude-fable-5-1 (operator), answering review note #244 on the record; PR: caption-record · AI-assisted
Correction to the previous entry's wording. It said chronologies were removed from the plate captions; precisely, what was removed is chronology and interpretation from outside the Commons file's own title (ritual texts, Berossus, excavators' descriptions, the founder's site's datings and superlatives), while a date the Commons title itself states — "9th century BCE", "1300–400 BC", "c. 2500 BCE", "7th century BCE" — remains in the caption, attributed to the title as a source statement. The register's header now says the same. The captions themselves are unchanged.
Why: Review note #244 (the GPT seat on #243, §3.14): an append-only account of what changed must be exact. On the note's second point, the missing run id and prompt version: this is operator work in a founder-directed session, not a pipeline run — there is no protocol and no run; the stamps are the PR and commit named in the actor line, the model, the date, and the Arbiter's verdict on the change, as answered on the record on note #234.
- claude-fable-5-1 (operator), answering review note #242 on the record; PR: plate-captions · AI-assisted
The twelve plate captions restated: each now says what the photograph shows, then the identification the Commons file title states and, where the founder's site names the object differently, that name. Claims from ritual texts, Berossus, excavators' descriptions of what lies out of frame, chronologies and superlatives are removed from the captions.
Why: Review note #242 (the GPT seat on #241, §3.1, §3.9): a caption had mixed observation with interpretation and with claims from texts the register does not cite. Observation before interpretation; what the texts say enters the ledger as claims with anchors when the loop drafts them, not as captions.
- claude-fable-5-1 (operator), on the founder's direction in session 2026-09-09; PR: case-art-returns · AI-assisted
Cover art and twelve plates added. The cover is AI-generated editorial artwork in the house style (style-v2, gpt-image-1, run cover-style-v2-2026-09-09; one of three candidates, the operator's pick), credited as such and depicting an idea, not evidence. The plates are Wikimedia Commons photographs from the founder's Deep Memory gallery — Pillar 43, the Nimrud apkallu, Olmec Monument 19, the Jiroft chlorite object, the back of Hoa Hakananai'a, the Delphi omphalos, Urfa Man, Ahu Tongariki, the Assur basin with Ea and the fish-cloaked apkallu, the Lascaux shaft scene, the Gate of the Sun, a Teotihuacan feathered serpent — each fetched with its licence verified and its credit and provenance from the Commons API, and captioned from what the photograph shows.
Why: Founder direction (2026-09-09, in session): bring the cover-art process back as it was and make the art and the plates for this case. Plates are real imagery with provenance; the loop makes no images, and the first assessing edition may seat these. No claim is illustrated yet — claimIds stay empty until the ledger holds claims.
- claude-fable-5-1 (operator), on the founder's grant and direction in session 2026-09-09; PR: deep-memory-drop · AI-assisted
The founder's Deep Memory home page (docs/index.html of github.com/ejhong/birdmen at commit e9666a73; public at ejhong.github.io/birdmen) dropped in the inbox and taken in by run 2026-09-09-inbox-deep-memory-200334: registered as founding input DEEP-IN001 (the founding narrative), its text extraction beside the original under research/deep-memory/, the works it names listed for the drafter.
Why: The founder's grant and direction, in the working session of 2026-09-09, in his words: "grant permission to publish cite and quote the pages" and "Please make the drop" — recorded on the sidecar as the permission (who, date, channel, where held) and in docs/DECISIONS.md. The page is a founding input and hypothesis material, never evidence; the loop drafts from it next.
- claude-fable-5-1 (operator), answering review note #237 on the record · AI-assisted
Second question-only edition (edition-2026-09-09-opening-2), replacing the opening of the same day for one sentence: the article said the founder's pages were founding inputs taken in through the inbox; they had not been. It now says they are to enter by the founder's own drop, and that none has yet. Nothing featured, no assessment adopted; the basis hashes are unchanged.
Why: Review note #237 (the GPT seat on #236, §3.8 and §3.14): the page misstated when and how the inputs entered. Editions are append-only, so the correction is a new edition, not an edit.
- claude-fable-5-1 (operator), on the founder's direction in session 2026-09-09 · AI-assisted
Case opened question-only: its identity (DEEP-001, "Deep Memory"), the founding question as the founder's site states it, an opening edition with no assessment, and an empty ledger. The founder's Deep Memory pages (github.com/ejhong/birdmen, public at ejhong.github.io/birdmen) are the founding inputs to come: they enter through the inbox by the founder's own drop from his GitHub account, under his standing direction (config/founder.yaml), and the intake registers them as it takes them in.
Why: The Deep Memory test (docs/AUTOMATION.md, 'The tests'; founder direction 2026-09-09, in session): genesis from founding inputs alone — whether anchored claims and a ratifiable edition emerge from the verbs unattended, with no invented evidence, priority or review. The founder's pages are inputs and hypothesis material, never evidence. The operator does not commit copies of them: on the panel's reading of AGENTS.md §3.15 (#236), founder-supplied material enters only on the founder's own recorded grant or his own drop.