# Two repertoires, side by side — 23 September 2026 - Added [the two walls](https://ejhong.github.io/birdmen/corpus.html): 72 openly licensed photographs of carved surfaces, 39 from Göbekli Tepe, Karahan Tepe and Nevalı Çori and 33 from Orongo and the wider Rapa Nui, each with what is legible on it. Filtering highlights and never removes, because the stones that carry nothing are the population the question is asked against. - Stated the selection rule before looking: files were listed from fixed Wikimedia Commons categories, then kept if one carved surface is the subject and legible enough to say what is and is not visible. Undecorated pillars were kept deliberately. Every survivor was then opened at full size; nineteen were dropped at that stage, including a modern illustration of a bull, two annotated composites and eleven views of rock in which no carving could be read. All exclusions are published with their counts. - Recorded a result the collection could not previously see. A bird and a rounded form share one surface in 3 of the 39 Anatolian photographs — and those three are one object, Pillar 43. On the Rapanui side the same combination appears in 3 photographs across 2 objects, inside a repertoire where birds are the dominant subject (13 of 33 photographs, against 8 of 39). The comparison is not between two bird-centred repertoires; it is between one pillar and a tradition. - Set the two local explanations beside each other. Rapa Nui has a documented reason to carve a birdman holding a rounded form: the tangata manu competition at Orongo, and the first manutara egg brought back from Motu Nui. Neolithic Anatolia has no documented meaning for the disc on Pillar 43, and its carved repertoire is overwhelmingly animal. - Drew the interval once. Every registered date on a single scale from 10,000 BCE to 2000 CE, with the unresolved birdman-carving date listed rather than plotted. - Feature states record what is legible in one photograph. "Not legible in this view" is not a finding that an object lacks the feature, and no count on the page is a frequency in either tradition: photographic coverage is not corpus coverage. # Better photographs, one type scale, one site map — 22 September 2026 - Replaced four weak evidence images with openly licensed photographs of the same objects, keeping each previous image as a second view. The Cahokia birdman tablet is now a photograph of the object rather than a modern illustration. The Lingjiatan pig gains a museum-display photograph whose caption independently repeats the 88 kg weight and tomb 07M23; an excavation report is still the check that would settle them. The Gate of the Sun and the front of Hoa Hakananai'a replace a rights-unresolved supplied screenshot and a dim gallery view. Credits, rights, source URLs, hashes and transformations are recorded; the tablet crop is the only local change to any file. - Rebuilt the type scale. The most-used text on several pages was 9-pixel monospace; small type now starts at 12 px, and explanatory sentences moved from the monospace into the reading serif. The monospace keeps labels, counts, dates, credits and controls. No wording changed. - Fixed colour contrast throughout. Muted grey, the accent copper and the labels on dark panels all now pass WCAG AA for their size. One lede on the homepage's dark section had been rendering near-black on near-black at 1.25:1. - Gave every page the same header and the same site map. All twenty-four pages, in three different page shells, now carry one navigation with the current section marked, and one footer listing the whole site by section. - Fixed horizontal overflow on phones. A grid minimum blew out the two-column study layout, wide data tables now scroll inside themselves, and long credit lines wrap. # The cluster index — 22 September 2026 - Added a culture-pair cluster index, derived entirely from the maintained register: 23 pairs, 19 distinct comparisons, 18 of them with no route established in this review. Each pair has its own page setting the two traditions side by side, line by line, with what corresponds, what differs, the transmission check, the dated episodes on each side and the ordinary explanations already on the table. - Ordered the index by how many comparisons stay open *after* a source check, never by evidence strength. A subcomparison is counted inside its parent line, because it is not separate evidence. Göbekli Tepe and Rapa Nui lead with three separate lines of comparison; that is a property of what has been collected here, not a measurement of the world or a finding of contact. - Published a sparse culture-by-culture table: 23 of 210 possible pairings carry a comparison. Its numbers count lines of comparison, and the ranked index below it carries the same information as text. - Replaced the catalogue's "Cultures" view with the same cluster ranking, keeping every existing `?view=cultures` link working. The collection's landing page now opens the three densest clusters directly. - Gave every catalogue photograph the framed-plate treatment used elsewhere on the site — a hairline frame and a light mount instead of a beige field — across the comparison cards, motif families, dossiers, homepage pair and intake images. No image bytes, credits or crops changed. # A closer visual investigation — 20 September 2026 - Added a 22-object study with interchangeable photographs, source-informed annotation regions, five explicit feature questions and shareable filters. Uncertain, unseen and unresolved features remain distinct; new AI rankings or independently certified reference labels are not claimed. - Acquired unchanged DAI photographs of Pillars 33, 38 and 56 and twelve public-domain Met photographs. Saved title queries, metadata, selection amendments and exclusions make the museum retrieval replayable. Ordinary variants and incomplete scenes remain visible. The previous Assyrian eagle-headed figure is a named anchor, separate from the new selections. This is not a global rarity sample. - Corrected the moai ring discussion. Pitts et al. (2014) propose an egg role for the older ring in the later birdman scene. Damaged limbs make the earlier categorical held/not-held description too strong. Frozen test material is unchanged. - Added nine dated historical records and a bounded search log, with distinct context, object, model-estimate and documentary date types. The Egyptian object ranges are plotted separately; collection dates are not carving dates. The checked sources do not establish a motif transmission chain to Rapa Nui. - Added six objects and six motif attestations to the main collection, including a separate Late/Ptolemaic Egyptian context for three newly sourced bird figures. The catalogue now holds 43 entities, 51 scoped observations and 56 images across 24 cultural/context groups. The 68 original input files remain untouched. - Linked the study from the homepage, central dossier, local context, relevant motif families and research review. Added maintenance and follow-up instructions. # Motif families, complete intake and smaller, better tests — 20 September 2026 - Tightened the homepage by 59%, research overview by 45%, central case by 16% and submerged assessment by 18%, counting document text including disclosures but excluding scripts/styles. Original quotations, protocols and raw results remain intact. The catalogue gains new records, so its total document text grows. - Made motif families the main entrance: 19 families across cultural traditions, with 45 explicit, scoped observations. Claim tags no longer have to stand in for object-level motif evidence. Date filters use each observation's relevant episode; unknowns and provisional readings remain explicit. Comparisons, cultures, map and timeline remain available, with old shared URLs preserved. - Audited all 68 input files. Added the Gate of the Sun to the comparison register and recorded 22 intake leads, including the staff-holder montage, animal-over-back comparison, Egyptian bird figure, bird-mask views and smaller Fenton comparisons. All 21 staff-montage panels have intake slots; most still lack verified identities. AI planning documents and contextual genetic/seafaring arguments are accounted for without being promoted to independent motif evidence. - Corrected two particularly useful examples: Sydney's Gilgamesh is modern (unveiled in 2000), and two panels in the new serpent montage repeat the same Olmec La Venta Monument 19 under conflicting labels. Egyptian and Māori panel identities remain unresolved. Original files and new published copies are unchanged. - Added a descriptive breakdown of the civiliser signal: 248/397 completed lineups reuse 11 pairs in two directions. Equal weighting of pooled pair rates gives 61.8%; the pair-specific variation is now visible. This is not a new AI run, independence claim or revised significance test. - Downloaded and retained the CC BY 4.0 Geersen/DLR Baltic AUV survey. A reproducible numerical experiment measures resolution loss at six scales. The 115 m averaging retains about 0.3% of a defined local-relief variance. That quantity includes noise and natural relief and is not wall recovery, survey accuracy or an archaeological detection score. Raw ZIP, hashes, method, measurements and figure are published. - Proposed bounded next tests for visual localization, source-based narrative comparison and archaeological detection across independently labelled surveys. Their outstanding corpus/reference requirements are explicit. No new paid model requests or archaeological discoveries are claimed. - Updated AGENTS, schema/acquisition instructions and the research programme. Added checks for scoped motif bundles, carving-date selection, input coverage, duplicated panels, aggregation of repeated lineups and numerical handling of missing survey cells. Browser checks cover new pages and desktop/mobile navigation. --- # A collection of connections, and submerged worlds — 19 September 2026 - Built the first group-based catalogue: 19 comparison threads and 4 optional controls, 35 curated groups, 23 cultural/context clusters, 36 evidence records, 27 image records and 40 sources. These are editorial record counts, not counts of independent objects or verified unexplained historical connections. - Added image-led comparisons, cultural connections, map, timeline and theme browsing, with shared filters and URLs. Dossiers preserve the selected collection view when returning. The core evidence remains readable without JavaScript. - Kept the pillar–moai–boulder inquiry central while expanding the Anatolia/Rapa Nui repertoire. Alternate photographs live within an object's record. Narrow subcomparisons are linked to their parents; repeated views do not inflate the culture-connection count. Dated episodes remain separate from culture lifetimes. - Added traceable Met photographs of an Assyrian protective figure and Indus seal, and a historical kavakava photograph. Guest authors Lee Paqui, Hugh Newman, Ryan Seven and Richard Cassaro are attributed separately from Graham Hancock. Caption leads, unresolved rights and missing architectural photographs remain visibly incomplete. No new paid visual ranking was run. - Added the submerged-landscapes strand: six places/projects, six data entry points and 18 sources. Actual retained NOAA relief data drive world, North Sea, Sunda and Gujarat/Khambhat depth experiments. Numbered project locations are approximate; the gold depth interval has no ancient date assigned. - Distinguished ongoing SUBNORDICA work, published archaeological discoveries, landscape models and disputed Khambhat structure claims. Listed concrete data access and satellite opportunities. The North Sea is the initial research priority; a defensible regional reconstruction around 12,000 BP remains future work, not a result of this demonstrator. - Added root `AGENTS.md`, a catalogue acquisition guide and contribution form, and underwater programme/data documentation. The homepage now opens both strands; previous essays, experiments and bookmarks remain accessible. - Extended checks for group membership, source and image integrity, dates, filtering, nested comparisons, original/derived grid hashes, geographic orientation, map behaviour, navigation, mobile layouts and static fallbacks. This is the first working edition of the wider resource. A systematic worldwide search, larger local repertoires, independent specialist review and dated regional palaeolandscapes remain substantial next stages. --- # The pillar, the moai and the bird — 19 September 2026 This edition centres the general question of ancient resemblance on the actual objects. It does not claim to solve their historical relationship. - New photographic homepage and central dossier, with alternate views, zoom, optional location guides, linked scrolling, a geographic context map and a chronology that separates the moai from later carving episodes. - Identified the supplied bird-and-round-object rock as the British Museum's Orongo birdman boulder, Oc1920,0506.1. Added Routledge's 1919 figure 112 as an earlier photographic witness. Dark lines on the modern reference remain unexplained. - Retained a wider atlas and moved Fenton's full illustrated thread into a separate essay. Hancock and Fenton have comparable, sourced introductions on the homepage. Guest authors on Hancock's site are not attributed to him. - Added a comparison of visible features and alternative explanations. The central ring and the small modern sampling plug on the moai are distinguished. Recent evidence of Polynesian–American contact is included without implying an Anatolian connection. No generated imagery is used as evidence. - Audited all five studies. Removed the claim that the old image pilots answered the resemblance question, corrected the use of historically loaded sub-scores as an independent resemblance test, qualified blinding, and removed causal claims drawn from geographical dispersion alone. Original statistics and responses remain. - Ran a frozen recognition diagnostic: 20 planned requests, 19 completed and one connection failure. Of 94 scorable feature answers, 73 matched provisional source-informed references, four were wrong, 12 uncertain and five missing. It did not pass the perception screen. References and model locations have not been independently reviewed; no historical ranking is permitted. - Published every request record, input/reference register, prompt/schema/settings freeze and result. Errors are preserved. The four older responses selected for the audit demonstrate failures; they are not a measured failure rate. - Updated offline and browser checks for source integrity, missing/uncertain answers, critical perception failures, accessible image controls, mobile layouts, old bookmarks and pages readable without JavaScript. Fixed the old browser test that waited for a lazy-loaded image outside the scrolled-to region. New edition source files are in `research/pillar-moai/v2/`; new response files are in `data/pillar-moai/v2/`. Old data and original protocol/change logs are untouched. The earlier layout notes below describe earlier editions, not the present navigation. --- # Editorial and research revisions — 5 September 2026 All five investigations remain equally accessible. This is not a deletion, a ranking of their importance, or a claim that their observations were false. - The first layout revision separated the study collection and illustrated narrative. The follow-up below combines them again, with the studies first. - The original fonts, palette and framed-photograph treatment are retained. - Six user-supplied reference images are reproduced unchanged with a provenance register. Unverified photographic credits, object identities and caption-derived measurements are explicitly distinguished from verified observations. - The original civiliser lineups retain their data and results. Their presentation now distinguishes a similarity signal from an inference of transmission. The misleading description of pretrained embeddings as knowledge-free has been corrected, as has the suggestion that a score below positive controls means no more similarity than strangers. - The flood page now states that regional structure is compatible with common ancestry and that post-contact recording is not proof of borrowing. No original statistics or preregistrations were changed. - Every local study has the same opening scope-and-limitations panel. - Civilisers version 2 is an exploratory source audit. It is not an unseen-data preregistered test, a new batch of model lineups, or a global rarity estimate. The full pre-revision site and its wording can be inspected in [commit c02aee5](https://github.com/ejhong/birdmen/tree/c02aee5/docs). Original pipelines, raw responses, result datasets and preregistrations retain their paths. ## 19 September 2026 · controlled follow-up and local context - Published the preceding presentation and recognition audit in commit `cd51bbdb`. - Ran a separately frozen 24-request view/reasoning experiment on the same three focal objects. All calls completed; estimated API usage cost USD 1.25556. - Primary label matches were 3/10 (low/single), 4/10 (low/paired), 7/10 (high/single) and 9/10 (high/paired). Uncertainty remains in the denominators. - Audited every positive Pillar 43 scorpion answer after the run. All three boxes lie in the upper bird area rather than the lower shaft. The new report shows those locations explicitly; categorical scores are preserved and must not be read as validated perception accuracy. Independent review remains incomplete. - Added the illustrated local-context study and nine-object register. New assets include Teomancimit’s licensed Pillar 2 photograph, an archival reproduction of the existing Mata Ngarau photograph, and a credited preview of Factum Foundation’s 3D recording of the Rapa Nui museum’s painted birdman slab. The duplicate rock photograph is explicitly one surface, not another independent example. - Recorded the broader catalogue proposal: claimed anomalous similarities as the main collection, known diffusion/local examples as linked controls, gallery, map/timeline, motif collections and focused graph views. This is a concept, not a claim that the catalogue or worldwide search has been completed. ## Follow-up layout and clarity fixes - One combined main page: all five studies first, then the illustrated narrative. - The bird-rock/Pillar 43 detail is part of Bird-men. Pigs and reptile-like animal carvings follow Handbags. There is no separate "A closer look" feature. - Personal bylines and name references have been removed from site copy. Functional repository and source links are retained. - `narrative.html` now redirects old bookmarks to the corresponding main-page section. - Civilisers V2 is explicitly identified as the new source-audit pilot, not a new run of the original similarity experiment. A stronger comparison has a separately labelled draft protocol; it has not been run and contributes no new results. # Fenton's theory added to the narrative — 9 September 2026 - The illustrated narrative now has a second section, "A theory from the south", which states Bruce R. Fenton's account of a southern origin for the Göbekli Tepe symbols as he states it, in his order, with the fourteen pictures from his X thread of 24–26 February 2025 and short quotations. Each claim carries a note of what the thread does and does not supply. The later sections are renumbered; no section was removed or reworded beyond the pointers below. - Bird-men, Handbags, Hands, Serpents, Civilisers and Floods each gained a short "Fenton's reading" pointer to the relevant claim. Explanations gained a third column, "A southern source", beside transmission and convergence. - A verbatim transcript of the thread, with one passage marked as missing from the supplied screenshots, is kept in `research/fenton/thread.md` and published as `docs/data/fenton-thread.md`. The screenshots and attachments are preserved in `inputs/fenton/`. - The image register now records a section for every supplied picture. Nine attachments are published unchanged; five pictures captured only inside screenshots are published as crops with their boxes recorded. A map of the claimed movements, drawn by this project, is labelled as the thread's claims rather than evidence. - Sources gained a list for the theory and the studies it cites, including the excavation team's 2017 reply. # Bird-man details, picture frames and identifications — 10 September 2026 - Bird-men gained "Bird, or bird-man? Two details on Pillar 43": the V-shaped line at the bird's throat, which Fenton matched to Urfa Man's collar on 9 September 2026, and the forward-bent knee a reader added. The page states what the photographs show, what the excavators' description says, and what would settle it. A detail of the site's own Pillar 43 photograph was added beside the Urfa Man photograph already in Hands. - The pictures in "A theory from the south" now use the same framed-print treatment as the galleries: a fixed-height mount with the picture contained, and the caption inside the frame. Text and picture still sit side by side; the alternating left–right layout was dropped. - Identifications after text and image searches, recorded in the image register, the captions and the transcript's addendum: the Yinganna painting is the Yingarna figure on Injalak Hill, a recent painting; the painted man in the churinga montage is a Wellcome Collection print; the news graphic is an All That's Interesting article of 2 April 2021; the four Olmec heads are San Lorenzo heads; the bush-bucket passage is confirmed in Lommel's book by a search inside it. What was not found is also recorded. - The transcript has an addendum with the September 2026 exchange and its screenshot. - Bird-men gained a supplied photograph of the whole Gate of the Sun at Tiwanaku, registered as the seventh reference image, with a caption and a note on its forty-eight winged attendants, sixteen of them bird-headed, and on what a comparison with the Assyrian genii would need. - Civilisers gained "A navigator in Micronesia": a reported Pohnpeian tradition of a teacher who arrived and left, with the post's two pictures cropped from its screenshot, set beside the recorded Nan Madol traditions (Olisihpa and Olosohpa; Isokelekel) and marked as unsourced until a recorded telling is found. # Indus & rongorongo merged into this repository — 10 September 2026 - Investigation 01 no longer lives in a separate repository. The `ejhong/rongo` repository was merged here with its history: its page is `docs/rongo/`, its pipeline `pipeline/rongo/`, its data `data/rongo/`, its starting screenshots and briefs `inputs/rongo/`, its method and run order `research/rongo/README.md`. The original repository is archived and its page redirects here. - The study's recorded paths were rewritten to the new locations (`data/rongo/…`, `inputs/rongo/…`) in the scripts and the data manifests; rebuilding the page from the saved results reproduces it exactly. No result, scan or crop was changed. - The study card, the catalogue page and the README now link to the local page, which carries the same scope-and-limitations panel as the other studies. Its page copy no longer names people. # Title and cover — 10 September 2026 - The site is titled Deep Memory; the former title, "Bird-men, handbags and the navel of the world", opens the deck. The masthead carries the Deep Memory cover from the companion site Aletheia, labelled as AI-generated editorial artwork, and every page now has link-preview tags that use it.