Retrieval Knowledge Commons (LM-Dawn class)

Class card for the LM-Dawn cluster of search/retrieval infrastructures operated as public-good commons rather than as ad-funded platforms (DM form). The class encodes the LM pattern where retrieval over structural-knowledge corpora — web archives, scientific literature, cultural heritage, biodiversity observations, and machine-readable knowledge graphs — is community-governed, transparency-preserving, and capture-resistant. Distinction from adjacent cards: machine:decentralized-science-platform-class is the DeSci PUBLICATION substrate (preprints, cooperative IP, formal proofs); machine:dawn-machine-substrate-knowledge-class is the broad epistemic-substrate machine. THIS card is the OPERATIONAL RETRIEVAL LAYER — the indexing, embedding, and search mechanisms that make corpora findable and navigable without platform capture. The canonical LM structural signal is that retrieval governance is separated from corpus ownership, creating a commons form distinct from both the MM library-catalog tradition (curator-as-gatekeeper) and the DM platform form (retrieval-as-advertising-OPP). Named instances [EXTRAP unless noted]: (1) Internet Archive (1996+; Brewster Kahle; San Francisco 501(c)3; ~735 billion web captures in Wayback Machine 2024; ~47M documents in Open Library; ~1M daily users; $30M+ annual operating budget from donations and grants). The canonical high-capture-resistance instance: non-profit legal structure + mission-oriented leadership + distributed community support + court-contested preservation mandate (Hachette v. Internet Archive 2023 — defeated in lending; Wayback Machine legally validated as fair use in 2024 Hachette appeal). [CANON] (2) Wikidata + Wikipedia retrieval layer (~100M structured items 2024; free/open SPARQL endpoint; Wikimedia Foundation 501(c)3; CC0 data license; community- curated structured knowledge; Query Service for SPARQL retrieval; OpenRefine reconciliation; Wikibase reuse across 1000+ community instances). [CANON] (3) Library Genesis / Sci-Hub (~80M papers, ~3M books — controversial but the largest instance of retrieval-commons resisting paywall capture; Sci-Hub bypasses Elsevier/Springer/Wiley paywalls to ~8M researchers/month 2024; Library Genesis mirrors across multiple jurisdictions). Contested-commons instance: high community value, legally precarious in Berne-Convention jurisdictions. [EXTRAP-classification] (4) HathiTrust (~18M digitized volumes; Ann Arbor; multi-university consortial commons; Authors Guild v. HathiTrust 2014 — full text search for accessibility found lawful; 180+ institutional partners; $12M+ annual budget). [CANON] (5) Europeana (~50M cultural-heritage items; pan-European commons; EU-funded; Open GLAM — galleries, libraries, archives, museums; CC licenses; SPARQL endpoint; JSON-LD linked data retrieval). [CANON] (6) OpenAlex (post-Microsoft Academic Graph 2015; OurResearch 2022+; ~250M works; ~200M authors; DOI-based open index; CC0; REST API; fully open scholarly metadata graph). [CANON] (7) Common Crawl (~3.5 billion web pages crawled monthly; 501(c)3; petabytes of open web corpus available for AI training, research, and retrieval; used as training substrate by virtually all major LLMs including BLOOM, GPT, Llama). [CANON] (8) ROAR Map (~5K+ registered open repositories in global registry). [EXTRAP] (9) Zenodo (CERN Geneva 2013+; ~3M+ deposits 2024; EU OpenAIRE integration; DOI minting; all research outputs; GitHub integration; CC licenses; no paywall; 20+ year data guarantee). [CANON] (10) Permaculture Research Institute (~100K members; bioregional knowledge retrieval for land-management commons; operational retrieval layer over traditional ecological knowledge). [EXTRAP] (11) Indigenous Knowledge Network (UNESCO + local partners; ~30 partner orgs; retrieval-commons for traditional knowledge while maintaining community governance over access). [EXTRAP] (12) Prime Radiant retrieval layer (post-Phase-2 DuckDB+VSS+gemma3-embedding stack as commons-substrate instance: HNSW index over 768-dim embeddings; PageRank-based graph retrieval over ~360 Machine Cards; community-governed via open-source schema). [EXTRAP-self-referential] Mechanism pillars of the LM-class form: (1) Non-profit retrieval governance: search/indexing infrastructure governed by 501(c)3, consortial, or inter-governmental bodies rather than ad-funded corporations. The Ostrom design principle of "graduated sanctions" and "collective-choice arrangements" applied to digital retrieval: community members participate in policy development (Wikimedia community governance; HathiTrust member governance; Common Crawl advisory board). Theoretical anchor: Hess & Ostrom "Understanding Knowledge as a Commons" (2007) — the knowledge commons requires governance institutions at the level of retrieval architecture, not just corpus content. (2) Open APIs and transparent indexing: retrieval infrastructure exposes open SPARQL/REST/OAI-PMH endpoints rather than black-boxing ranking algorithms. Anti-PageRank-opacity commitment: commons retrieval indices are auditable, reproducible, and forkable. OpenAlex REST API (CC0); Wikidata SPARQL endpoint (CC0); Common Crawl S3-direct access (petabyte-scale open); Europeana API (CC). This is the capture-resistance mechanism: a retrieval commons that black-boxes its algorithm becomes a platform. (3) Vector-embedding + graph retrieval as commons substrate: the LM-Dawn retrieval layer adds semantic/vector retrieval over open corpora as a commons service. Embedding models (sentence-transformers; gemma3-embedding via llama.cpp; FAISS/DuckDB+VSS index) applied to open corpora generate retrievable knowledge-graph substrate. Prime Radiant's DuckDB+VSS stack is the self-referential instance of this pattern: community-governed retrieval over a hand-curated corpus. (4) Preservation mandate and temporal retrieval: retrieval-commons includes temporal retrieval over archived states — Wayback Machine as the canonical form. Commons-retrieval preserves the retrievability of past states against link-rot, corporate discontinuation, and memory-holing. The MM-Library tradition's preservation mandate reterritorialized into digital-temporal retrieval substrate. Emergence subtype [v0.2-gap — recorded here]: commons_retrieval_governance. The LM class instantiates community-coordinated retrieval without a central ad-funded platform OPP. Unlike crowdsourced (Wikipedia) or meritocratic_ hierarchy (OSS), the retrieval-commons form is constituted by the governance of access policies, indexing standards, and API availability. Capture-resistance [EXTRAP]: capture_resistance_index MEDIUM-HIGH (0.65). Non-profit + consortial governance + open API commitments provide structural capture-resistance. Internet Archive faces active legal threat from publishers; HathiTrust requires institutional membership for full-text access (partial capture); Wikidata is strongest (CC0 + Wikimedia Foundation governance). The proletarianization_risk is MEDIUM (0.52): vector-index and graph-retrieval infrastructure requires living technical competence to maintain; if the maintenance community thins, the retrieval layer stagnates even if the corpus persists (Stiegler technical-memory degradation pattern). LM mechanism signatures: capture_resistance_index 0.65 [EXTRAP]; proletarianization_risk 0.52 [EXTRAP]; liveness_temporal_coupling HIGH 0.68 (commons-retrieval must track corpus updates in real-time; Wayback Machine crawls continuously; OpenAlex ingests new papers daily; Wikidata edits in real-time); coordination_yield_index MEDIUM-HIGH 0.62 (commons retrieval layer reduces redundant indexing costs across community institutions; OCLC WorldCat as MM analog; cooperative cataloging yields coordination surplus). All quantitative state-variable values are [EXTRAP]. CANON framing applies to: Internet Archive 1996+ existence and Wayback Machine statistics; Wikidata ~100M items; HathiTrust ~18M volumes; Europeana ~50M items; OpenAlex ~250M works; Common Crawl ~3.5B pages; Zenodo ~3M deposits (all as cited).