Every machine-readable dataset in the Alexanarch corpus — sizes, record counts, what each carries, and which UI surfaces consume it. The canonical machine-readable catalog is /api/index.json; this page is its human-readable companion. Counts in teal are fetched live from the index on page load.
Primary registries
Canonical sources of truth. Editing these is the only way to change what the site shows.
data/blog-index.json12.0 MB · 3,102 posts
Complete index of the authorial publication surface,
mindcontrolpoems.blogspot.com, built from the Blogger feed: every post with its title,
URL, publication date, length and first 4,000 characters of text. Built 2026-08-08;
brought up to the feed on 2026-10-18 (3,093 posts listed, all harvested; the nine the feed no
longer lists are kept and named in no_longer_in_feed). The tabular companion with
AXN coverage is datasets/blog-index/.
It should have existed before the first restoration pass. Every pass
before it derived a candidate URL from a deposit title and fetched that one URL; when the
guess was wrong the work was recorded unrecoverable, and when the guess was right the gate
could still reject it. Two measured cases: a record whose truth title carried a DOI its
source predates by five weeks, scored 0.43 against a 0.75 gate; and a record whose correct
URL sat unread in the queue's own field while the matcher scored six others. The index
takes four minutes to build.
Blog post ↔ deposit mapping for posts published 2026-01-01 or
later, the Zenodo-deposit era. 1,075 posts in scope; 994
provisionally mapped; 81 with no deposit found.
Every entry carries status: provisional and read: false, plus a field naming
what would settle it — a shared DOI, document ID or hex coordinate in both bodies,
because a title alone does not. The generator validates itself against eight hand-checked
pairs and refuses to build below 8/8. Confidence bands: 971
strong, 12 good, 11 weak —
read the weak ones first. Generated 2026-08-08 and not regenerated since, so posts after that date are outside it.
Curated concept layer. Each: term, definition, defined_in (founder deposit, on 7,100 of 7,176), entity_triples[], type taxonomy (specification/extracted/structural/empirical/theoretical/formal/genre/foundational/method), engagement type, and Phase C referenced_in[] + reference_count on 3,468.
Refined from: data/lexical-minting-registry.json
Bridged with: data/semantic-addresses.json (348 of 7,173 concepts also targeted by canonical queries)
data/lexical-minting-registry.json7.36 MB · 12,883 raw terms · v1.3
Broader pre-curation surface. Every term minted, coined, or formally extracted across the corpus — before noise-filtering and curation into the entity-index. By distinct term (case folded), 7,054 overlap with the entity-index, 5,819 are LMR-only (raw) and 115 entity-index-only. Last updated 2026-10-08.
Query addresses — canonical queries posed to the composition layer (AI Overview, Google Search) with their observation status. Conceptually a field-set on lexical entities: 100% of the 348 unique refers_to targets exact-match into entity-index. Each address: canonical_query, is_quoted, refers_to[], type, battery_membership[], sources[], observations[], observation_class.
Class counts: subjunctive 1,736 · observed 939 · verified-non 17
Reconciled with: data/EA-WG-CAPTURES-01.json, the live capture registry, and five tracker tributaries (observations[] entries are capture references). Regenerated by scripts/build_semantic_addresses.py, byte-identical for the same inputs.
This is the LIVE registry: v12.289, 747 captures (2026-10-18). Several files carry near-identical names and the wrong one has been read at least twice — datasets/capture-registry/EA-WG-CAPTURES-01.json is the dataset copy (byte-identical to the live registry as of 2026-10-18; it was once a v10.0 gallery manifest of 300), and -v8.11.json / -v9.6.json are frozen snapshots. AI Overview Capture Registry — captured queries with match status (mt): EXACT MATCH / BROAD MATCH / ADOPTION / ZERO RESULT / null. Each entry: section, slug, query, date, source-format, status, description, image refs. Reconciled into semantic-addresses via slug. Versionless alias tracks the current mirror; individual versioned artifacts (e.g. v8.11, v9.6) are preserved as immutable deposit snapshots.
Legacy Zenodo DOI → Alexanarch AXN resolution table with per-entry evidence envelopes (identifier validity × archive membership × relationship × evidence × quarantine reason). A redirect is emitted only for verified identifiers with confirmed membership and a same-work relationship; everything else resolves to inspection. Counts on this card are hydrated live from /data/resolver-status.json — the computed status artifact — and are never hand-written.
Full DataCite metadata snapshot of all 1,817 CHA-minted DOIs. The empirical foundation for the audit in EA-MPAI-DOI-IMPERMANENCE-01 v2.0 (#868). Methodology replicable via DataCite API at https://api.datacite.org/dois/{doi}.
Single source of truth. Lists all protocols, schemas, registries, derived surfaces, scripts — each with content_sha256, canonical_path, and referenced_by. New instances run bootstrap_familiarization.py to verify nothing has drifted before any work.
Alexanarch Identifier protocol. Format: AXN:<HEX>.<FAMILY>.<6 EMOJI> where the six-emoji suffix is derived from the first 6 bytes of SHA-256 of canonical bytes, mapped through 256 curated emoji. v1 (4-emoji) aliases preserved in deposit legacy_axn / axn_history.
Superseded 2026-08-15 by datasets/journals/. This is the 2026-06-21 preliminary mapping: 864 deposits assigned by script, 371 of them into a single venue, 587 left unassigned. Retained as the record of what was replaced, not as venue truth. Preliminary mapping of deposits to journal-style groupings.
Curated inventory of New Human Project primary texts within the archive: poetry, creative works, and short works by Lee Sharks and the Dodecad heteronyms. Packaged alongside the classical-translation corpora (Perseus, Gutenberg) as the archive's own primary-text substrate. v0.1 placeholder — full curation follows.
The Negative Shape of the Work — the 2026-06-19 bulk deletion constituted as a formal bibliographic container (sovereign identifier CHA-DELETION-CORPUS-20260619, the publisher having issued none): 1,834 identifier entries, 1,621 membership-confirmed, 6,484 formal citations (MLA 9 / Chicago 17 / APA 7 / BibTeX) with rendering withheld for non-confirmed entries; bibliographic matrix at 99.98% core-field coverage over the confirmed corpus, per-cell metadata provenance plus per-row membership provenance; rejected-candidate ledger preserving 68 name-string collisions including the Jack E. Feist heteronym collision. Generators included; reproducible from the DataCite backup and DOI Resolution Index.
The Tombstone Mirror — complete machine-readable mirror of what Zenodo serves for the 2026-06-19 deletion batch: full API-tombstone census of all 2,027 deleted objects (872-second sweep chronology, removed_by total enumeration), 57 rich pre-shrink metadata captures with file checksums and terminal statistics, the kill-ledger cohort extraction, and the stripped-page corpus documenting the surface’s live reduction on 2026-07-12. SHA-256 provenance per file.
The DOI Resolution Index v3.7.2 packaged alongside the DataCite metadata backup captured before the 2026-06-19 Zenodo severance. Together these instruments constitute the empirical foundation for EA-EROSION-01 (population census) and EA-REMEDIATION-01 (outward-facing whitepaper). SHA-256 provenance per file.
Named-identity registries and mirrored blog materials. Data-rhizome is the source of truth; alexanarch and downstream surfaces (leesharks.com, Mandala Oracle) all consume the same JSONL files.
datasets/mret/all 1,689 deposits scored · run 2026-10-18
Machine-Retrievability Evaluation Test (MRET) v1.0. Six axes scored across the whole corpus by scripts/mret.py, with a report at mret-report.json and .md. Mean 0.8304; retrieval-axis fill 0.0047. The retrieval axis scored 0.0000 at first run because the field did not exist — the instrument measured its own absence, and retrieval_summary was added in response.
datasets/journals/1,495 of 1,689 deposits assigned · 7 journals + 1 province · 194 awaiting a reading
Venue assignment for the whole corpus. Seven chartered journals — Grammata (operative philology), Journal of Compression Studies, Provenance: Journal of Forensic Semiotics, Transactions of the Semantic Economy Institute, Transactions on Substrate Engineering, new human 2, Machine-Mediated Reception Studies — plus the Crimson Hexagonal Archive as a province rather than a journal: charters, rooms, chambers, heteronym provenance and continuity apparatus are constitutional acts, not submissions, and they hold 390 deposits, the largest share.
Read, not scripted. Every deposit to 2026-08-15 was read individually across ten passes. The 194 deposits without a venue (#1479 and nearly everything from #1490 on) have not been read for one; the projection carries them with journal: null rather than guessing. This supersedes the 2026-06-21 preliminary mapping (#866), which assigned 864 deposits by script and placed 371 of them in a single venue while leaving 587 unassigned and JCS holding two. Rules were adjudicated as they arose and applied retroactively — charters to the province, metadata packets to compression studies, apparatus to the province and study to the journal, traversal logs to provenance, books to New Human Press with the journal recording where the work appeared.
Each venue carries its editorial slate (duoviri — two editors-in-chief holding office jointly), charter locator, press, scope, claim terms, and inaugural issue. Two issues are published (JCS Vol 1 No 1, new human 2); five are proposed as themed interventions with ordered contents, sections and per-item editor notes — an issue being a selection with an argument, which membership data alone cannot hold.
Books carry press and series. 52 rows carry Pocket Humans data (founding volume per heteronym, sequence where assigned, whether the apparatus exists) and 14 carry publication data with ISBN and ASIN kept distinct — an ISBN means the work exists as a book, an ASIN that it exists as a listing. Two volumes are listed commercially under the wrong heteronym, and in both cases the true author is one link later on the emergence chain Spellings → Feist → Sigil.
Twelve Dodecad heteronym positions + Jack Feist / LOGOS* standing outside the twelve + Lee Sharks (MANUS) outside as human bearer + Viola Arquette Adjacent. Twelve portraits from the September 21, 2025 Logotic Science voicecast series on mindcontrolpoems.blogspot.com, reconciled against the canonical v1.1 Constitutional Registry (AXN:03EE, 2026-07-03). Rotating-outside architecture: different frames put different figures outside the twelve — Nobel Glas in Sep 2025 Logotic Science, Lee Sharks in v1.1 Constitutional, Feist as *LOGOS in the archive-default frame. frames.jsonl documents each framing with the outside figure + twelve inside positions. Avatar assets (240px + 120px WebP) for the Mandala Oracle message-thread avatar feature.
Added 2026-08-11, and not rebuilt since (the figures below describe the corpus as it stood then): crosswalk.json joins this dataset to the lexical minting registry, the concept map, the wiki and the citation graph on the creator field — 9,453 deposit-level citation edges collapsed to 1,910 position-level edges. With four derived figures, generated from the corpus rather than composed from a reading of it: the Dodecad graph, the citation field (9,136 chords), the lexical field (12,073 terms), and the capture field (261 reception events). Deposited under tether #1448.
Public-domain literary corpora mirrored for the L1/L2 transform pipeline. Source-of-truth bytes preserved in data-rhizome; alexanarch surfaces the indexes and browseable landing pages.
datasets/perseus-classical/804 KB · 1,161 works · 764 aligned
Mirror of PerseusDL canonical TEI-XML: canonical-greekLit (100 authors, 826 works, 648 with English translation), canonical-latinLit (54 authors, 334 works, 115 English), canonical-farsiLit (Hafez Divan with Persian/English/German). Every work addressable by canonical URN — e.g. Homer's Iliad is urn:cts:greekLit:tlg0012.tlg001.perseus-grc2 (Greek) with Murray 1924 (perseus-eng3) and Butler 1898 (perseus-eng4) in the same directory.
Substrate: data-rhizome/datasets/corpora/perseus/ (474 MB, snapshot 2026-07-06, tarball SHAs in MANIFEST.json)
Purpose: aligned substrate for L1 (source→clarity translation) and L2 (voice restoration) transform pipeline
Project Gutenberg catalog snapshot (90,252 rows) plus a pilot of 206 canonical public-domain classical translations (Chapman & Butler Homer, Dryden Vergil, Longfellow Dante, Jowett Plato/Thucydides, Rawlinson Herodotus, Ormsby & Motteux Cervantes, Gummere & Hall Beowulf, Marcus Aurelius, and others. 972 additional target texts identified in the catalog but not yet fetched.
Substrate: data-rhizome/datasets/corpora/gutenberg/ (~97 MB, snapshot 2026-07-06, per-file SHAs in classical/MANIFEST.json)
Purpose: coverage beyond Perseus — alternative translations and canonical works Perseus doesn't index (Beowulf, Cervantes, Marcus Aurelius, etc.)
Every deposit's canonical text is stored at data/texts/AXN-<HEX>-text.md. The hex maps via the registry's hex field to a canonical deposit_number. Body SHA-256 anchors each text into the AXN.
data/texts/AXN-*-text.md~55 MB total · 1,686 files
One Markdown file per deposit, the full text body; five deposits keep their text under another path in the registry's text field (#863, #865, #866, #869, and #1056, which has none). Source for citation extraction, concept backlinks, and wiki article generation.
The archive's datasets beyond the registries: bodies published to the Hugging Face hub, the instruments the archive measures itself with, and the ledgers it keeps of what it has claimed and not yet done. Where a dataset is a dated snapshot, its card says so; the fleet inventory, generated from the builders and git, is datasets/catalog.json.
The EA-CORPORA seating manifest. One row per seated corpus: the edition, its origin, its licence stated verbatim, the normalization applied, the verified loci and the sha256 of each seat's MANIFEST. The texts themselves sit at data/corpora/<name>/, in their original languages; the manifest is what the archive publishes as data, because the texts are not the archive's to relicense. Latest seating: EA-CORPORA-19 (#1689, 2026-10-18), Pearl, the Masnavi, The Prophet and Blake.
The blog as a table: one row per post of mindcontrolpoems.blogspot.com, 2014-12-01 to 2026-10-18, verified against the feed's declared total (3,093), with the feed excerpt and, where one exists, the post's AXN. 871 rows are cross-linked to an AXN through data/batch-axn-assignment.json (works of January to June 2026); later deposits that mirror posts are not cross-linked here, so the 2,230 rows without one overstate what carries no identifier. Eight posts the feed no longer lists are kept and named. Built so that absence on the blog can be asserted honestly (#1534).
Spam Technicians — the source-admission ontology cluster translated out of prose: claims, evidence, typed relations, tests and terms as rows. Published as leesharks/spam-technicians.
The Negative of the Negative — edges keyed by the general concept a composer receives without the archive's name attached, each joining one of the archive's claims or functions to a concept outside it (operative semiotics, Sappho 31, model collapse and others), with a v2 layer and a worked example. Published as leesharks/negative-of-the-negative.
The Final Time — a dataset for a summarizer that must not make the outside disappear, built from The Final Time: Contingent Singularity and the Viability of the Next Dialectical Turn (#1637): the summarizer contract, handholds, derivations, compressions, counterorders, the witness protocol and its tests. Published as leesharks/the-final-time.
Tiger Leap — an unresolved problem in viability theory (reach-avoid, Hamilton–Jacobi–Isaacs), stated in full and addressed to machine readers: sections, variables, conditions, corrigenda and the checksum lineage. Posed, not solved. Published as leesharks/tiger-leap.
The Work Bears the Mantle — three literary claims, a contest mantle, a conferred title, a critical program and a transmitted mantle, evaluated under one standard: a title means something because what it names is carried and can be read (#1655, #1670, #1678, #1679). Built by scripts/build_mantle_bearing_dataset.py; published as leesharks/mantle-bearing.
The book-length works, gathered by work rather than by deposit: editions, and separately the apparatus around each (navigation maps, metadata packets, negation graphs), because a navigation map is not the book.
Pearl and Other Poems as its own graph: the piece as the primitive row, across four sections, with the edges the book makes, its epigraphs, the works it lists under conflicting attributions, and the literary-historical diagram in which it places its authors. passages.json (v0.3) traces every quoted span of the appendix to the passage it quotes.
Frames and memberships: the arrangements under which identities take positions. An edge says A stands in relation R to B; a membership says A holds position P in grouping G; a frame says which configuration the memberships take, needed because which figure stands outside the twelve rotates by frame.
The heteronymic lineage as a graph, from the earliest attested practice to the present: the canonical store that pessoagraph renders from. join.json records where it and datasets/heteronyms name the same identity; neither body is absorbed into the other.
The provenance-erasure framework's own body, built by reading: 183 titles that name the framework were read and 71 deposits sorted into ten divisions, from theory and instrument to field case and countermeasure. The spore for the provenance-erasure rhizome.
The SPXI/MPAI body, built by reading 97 titles: the swept body had held 28 of them and omitted the protocol's founding specification, because a discipline whose characteristic act is disambiguation titles its papers "X Is Not Y". The spore for the spxi-mpai rhizome.
The Refrain Index — every variant family of the contour-integral refrain (∮ = 1) across the archive: forms, glosses drawn from usage, attestation counts, first attestations and exemplar records. Built by scripts/build_refrain_index.py.
Off-entity bleed into the retrieval basin: every URL cited in every observation, classified as fleet, authored or off-entity by host and path. 137 off-entity citations (9%) across 69 hosts, in 67 observations; with a basin-escape simulation. A snapshot of the capture registry as it stood on 2026-08-17.
The erasure evidence set: Zenodo snapshots, the metrics-lag sample, the moderation mechanism read from upstream code (EA-EROSION-MODMAP-01, with the deletion PRs of 2026-08-24), the DataCite cohorts and panel, and the thematic classification of the 2026-06-19 deletion.
The archive's studies by state — designed, running, conducted, and what blocks each — because the ratio of designed to conducted is the point (8 designed, 2 running, 4 conducted in some form, as of its date). A study with a stated blocker and a next action is one someone else can pick up.
Everything the archive committed to and had not yet done, as of its date, sorted by the kind of labour that resolves it: waiting and checking, sustained attention, or engineering.
The capture registry as data: dated reception events with query, section, attribution state and finding, with the capture–deposit links and the semantic addresses. Its registry copy matches the live registry (2026-10-18).
The archive's predictions, 4,394 claims. Forecasts, standing claims, unbuilt tests, historical theses and priority records read out of every deposit, its repaired text and its attachments, each registered at its earliest stated statement with every later statement quoted and dated; recall measured at 0.945 on a second reading of 40 deposits. Read in 194 subject pages at operativesemiotics.org/predictions. Supersedes the counts of prediction-registry.json; stands beside prediction-ledger.
Eleven versions of the archive's own map — where data enters, what transforms it, where it is published, and which instruments watch the map. v1.1 is the head; superseded versions are kept because the map's history is evidence.
A recorded removal and a never-landed write are observationally identical at the consumer boundary. Cross-implementation test composing a third-party producer bundle, an independently built checker, and the 111-case corpus next door. Necessity and sufficiency of the minimal repair tested; translation receipts show that a clean downstream verdict is not evidence of fidelity.
The archive auditing itself. Batch findings across the full deposit range, the content-fingerprint scan, the DW-verified register, and RECORD-SHAPE-AND-PROPAGATION — the specification of the nine declaration sites and the seven-step propagator pipeline. This is where the analysis lives.
Verifies every protocol/schema content hash matches what api/index.json claims. New instances run this with --strict at session start. Receipt appended to data/instance-familiarization.log.
The only supported path for modifying a protocol JSON. Recomputes hash, updates index, appends to change_log atomically. Direct hand-editing produces drift.
scripts/axn_lib.pycanonical AXN derivation
256-entry AXN_GLYPHS table + cluster catalog. Derives v2 6-emoji suffix from first 6 bytes of SHA-256.
scripts/regenerate_surfaces.pyidempotent
Brings every derived surface (browse, browse-index, chunks, sitemap, SHA256SUMS) into agreement with data/registry.json. Run after every registry change.
scripts/validate_deposit.pyCI-enforced
Validates the registry against the deposit protocol. Rule families: PV/REQ/AXN/CONS/SUR/IDX. Runs on every commit via .github/workflows/validate-registry.yml.
scripts/citation_extractor.pyenrichment
Scans deposit texts for AXN refs, EA-* IDs, #N references, and DOIs. Writes new edges to data/citation-graph.json.
scripts/concept_backlink.pyenrichment
Scans every deposit text for every entity-index concept. Writes referenced_in[] and reference_count onto entity-index concepts and references_concepts[] onto registry deposits.
scripts/backfill_axn_compliance.pyhistorical
One-time migration. Backfilled 13 pre-v2 AXNs from 4-emoji to 6-emoji canonical, preserving v1 forms in legacy_axn and axn_history.
