# References

The consolidated bibliography for ARD's [defensibility trace](README.md) — every external source the [positions](positions/) rest on, de-duplicated into one numbered list. A position's `[handle]{N}` citation resolves to the correspondingly-numbered entry below.

This file is **bibliographic only**: author, title, venue, engagement scope, and substrate-confidence. It carries **no verbatim source reproduction** — load-bearing quotations live inside the positions that argue with them (bounded quotation for criticism and comment), not in a standalone extract layer. That is the deliberate IP posture: the *findings* that defend each commitment are ARD's own; the source attestations that reproduce external copyrighted passages stay in the originating deployment and are not republished here.

**On the `{N}`.** Each source has one **stable entry number** here, and every `[handle]{N}` citation across the positions uses it — so a handle carries the same `{N}` in every position, and `{N}` resolves to the entry of that number below. Numbers are **append-only**: a source added by extension takes the next integer and existing numbers never change, so live citations never break. (A handle's `{N}` is a whole-source pointer; where a position engages specific chapters or sections of a multi-piece source, the chapters are named in the prose and in the entry's engagement scope.)

**Substrate-confidence.** Entries engaged below source-direct depth (paywalled, bot-blocked, or read via secondary sources) are flagged inline. A load-bearing claim citing such a source carries a matching `{confidence: search-summary}` marker in the position body.

This list **grows by extension** as more commitments earn a ported rationale.

---

## Standards & specifications

- **[w3c-prov-dm]{1}** — W3C, *PROV-DM: The PROV Data Model*, W3C Recommendation, 30 April 2013. <https://www.w3.org/TR/prov-dm/>. Engaged source-direct: §1 Introduction; Components 1–4 (Entities/Activities, Derivations, Agents/Responsibility, Bundles).
- **[w3c-prov-o]{2}** — W3C, *PROV-O: The PROV Ontology*, W3C Recommendation, 30 April 2013. <https://www.w3.org/TR/prov-o/>. Engaged source-direct: Starting Point terms; Expanded terms (derivation + temporal properties); Qualified terms (Qualification Pattern).
- **[w3c-web-annotation]{3}** — Sanderson, Ciccarese & Young (eds.), *Web Annotation Data Model*, W3C Recommendation, 23 February 2017. <https://www.w3.org/TR/annotation-model/>. Engaged: §3.1 Annotations; §3.2 Bodies and Targets; §3.3.5 Motivation/Purpose; §3.4 Provenance. (TextQuoteSelector body shape.)
- **[tei-certainty]{4}** — TEI Consortium, *TEI P5: Guidelines*, Ch. CE "Certainty, Uncertainty, and Responsibility" (current release, accessed 2026-05-31). <https://tei-c.org/release/doc/tei-p5-doc/en/html/CE.html>. Engaged: `<certainty>`; `cert`/`resp`/`locus`/`degree` attributes; standoff placement.
- **[dcmi-elements]{5}** — Dublin Core Metadata Initiative, *DCMI Metadata Element Set* (DCMES), v2012-06-14. <https://www.dublincore.org/specifications/dublin-core/dces/>. Normalized at ISO 15836, IETF RFC 5013, ANSI/NISO Z39.85. Engaged source-direct: the 15-element base (all optional/repeatable); qualified DC (`dcterms:`); DC TAP obligation-constraint mechanism (`mandatory` / `mandatory-if-applicable`); dumb-down principle.
- **[nanopub-guidelines]{6}** — Nanopublication Guidelines (Working Draft), nanopub.net, 2014 (accessed 2026-05-31). <https://nanopub.net/guidelines/working_draft/>. Engaged: tripartite structure (assertion / provenance / publication_info); graph contents and requirements; integrity and trustworthiness.
- **[iso-iec-11179]{7}** — ISO/IEC, *ISO/IEC 11179: Information Technology — Metadata Registries*, 2023 edition (Parts 1, 3, 6; ISO/IEC JTC 1/SC 32). Engaged: Framework (administered-item concept); basic-attribute categories (Identifying/Naming/Definitional/Administrative/Relational); registration-status lifecycle. *Substrate confidence: search-summary — normative text paywalled; architecture cross-confirmed across multiple independent secondary sources.*
- **[niso-metadata-typology]{8}** — NISO, *Understanding Metadata* (2004 original; 2017 primer by Jenn Riley). Engaged via secondary sources (<https://www.metadataetc.org/metadatabasics/types.htm>; DCC "What Are Metadata Standards" briefing). Three-type typology (descriptive / administrative / structural) + administrative sub-types. *Substrate confidence: search-summary — typology cross-confirmed; exact NISO text not directly read.*
- **[arc42-fail-safe-defaults]{9}** — arc42 Quality Model, "Fail-Safe Defaults." <https://quality.arc42.org/approaches/fail-safe-defaults>. Engaged: full entry — safe-state-holds-independent-of-configuration; directional asymmetry toward restriction under ambiguity; atomic safe-state transitions.

## Reference wikis (CC-BY-SA)

- **[wikipedia-controllability]{10}** — Wikipedia, "Controllability." <https://en.wikipedia.org/wiki/Controllability> (CC-BY-SA-4.0). Engaged: §State controllability (core definition); §Continuous LTI systems (Kalman rank condition).
- **[wikipedia-faceted-classification]{11}** — Wikipedia, "Faceted classification." <https://en.wikipedia.org/wiki/Faceted_classification> (CC-BY-SA-4.0). Engaged: Ranganathan's PMEST; analysis–synthesis structure; facet-combination notation; relation to faceted search.
- **[wikipedia-faceted-search]{12}** — Wikipedia, "Faceted search." <https://en.wikipedia.org/wiki/Faceted_search> (CC-BY-SA-4.0). Engaged: faceted-search definition; facet values derived via text analysis or DB fields; commercial-adoption evidence; relation to Ranganathan's tradition.
- **[saltzer-schroeder-principles]{13}** — HandWiki, "Saltzer and Schroeder's design principles." <https://handwiki.org/wiki/Saltzer_and_Schroeder%27s_design_principles> (CC-BY-SA-4.0). **Secondary source** for Saltzer, J. H. & Schroeder, M. D., "The Protection of Information in Computer Systems," MIT, 1975. Engaged: the fail-safe-defaults principle and the eight design principles; HandWiki notes these are design guidelines grounded in experience, not invariants.

## Papers

- **[harnad-symbol-grounding-problem]{14}** — Stevan Harnad, "The Symbol Grounding Problem," *Physica D* 42 (1990): 335–346. arXiv:cs/9906002. Engaged: abstract; §2 (the grounding problem; Chinese Room / dictionary-go-round); §3 (iconic + categorical representations); §5 (conclusions); fn.6.
- **[bender-koller-climbing-towards-nlu]{15}** — Bender & Koller, "Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data," ACL 2020: 5185–5198. ACL Anthology. Engaged: §1; §3 (form/meaning; symbol grounding); §4 (octopus test); §7 (distributional semantics; Wittgenstein meaning-as-use); §9.
- **[bender-stochastic-parrots]{16}** — Bender, Gebru, McMillan-Major & Shmitchell, "On the Dangers of Stochastic Parrots," FAccT '21: 610–623. Engaged: §1; §5 (form/meaning; Saussure); §6.1 (coherence-in-the-eye-of-the-beholder; parrot definition).
- **[shanahan-talking-about-llms]{17}** — Murray Shanahan, "Talking About Large Language Models," arXiv:2212.03551v5 (Dec 2022, rev. Feb 2023). Engaged: §2 (Wittgenstein, language as collective behaviour); §3 (intentional stance; conditional scope); §5; §13 ("form of life").
- **[mitchell-krakauer-debate-over-understanding-llms]{18}** — Mitchell & Krakauer, "The Debate Over Understanding in AI's Large Language Models," arXiv:2210.13966v3 (2023). Engaged: the three-question framing; the Michael et al. NLP-community survey (51/49 split); functional-equivalence possibility.
- **[kaplow-rules-vs-standards]{19}** — Louis Kaplow, "Rules Versus Standards: An Economic Analysis," *Duke Law Journal* 42:3 (1992): 557–629. *Substrate confidence: search-summary — primary text inaccessible (HTTP 403/405); key passages (pp. 557, 563) confirmed as verbatim cite-throughs via two directly-fetched secondaries (Bodansky 2003; Parchomovsky & Stein 2015).*
- **[bodansky-rules-standards-intl-law]{20}** — Daniel Bodansky, "Rules and Standards in International Law," NYU Law School preliminary draft, March 2003. <https://www.iilj.org/wp-content/uploads/2017/02/Bodansky-Rules-and-Standards-in-International-Law-2003.pdf>. Used **only** as cite-through intermediary for Kaplow (1992) passages (pp. 557, 563–64); the draft is marked "Not for quotation or citation" and is not quoted for its own content.
- **[parchomovsky-stein-catalogs]{21}** — Gideon Parchomovsky & Alex Stein, "Catalogs," *Columbia Law Review* 115 (2015): 165. <https://www.columbialawreview.org/wp-content/uploads/2016/04/Parchomovsky-Stein.pdf>. Used as cite-through intermediary for Kaplow (1992) passages (footnotes at pp. 563, 569, 609).
- **[freiman-ai-testimony-2024]{22}** — Ori Freiman, "AI-Testimony, Conversational AIs and Our Anthropocentric Theory of Testimony," *Social Epistemology* 38(4): 476–490, 2024. DOI:10.1080/02691728.2024.2316622. *Substrate confidence: search-summary — body paywalled (Taylor & Francis).* Coins "AI-testimony"; names three anthropocentric assumptions behind testimony theory's rejection of LLM outputs.
- **[he-testimony-by-llms-2025]{23}** — Jinhua He & Chen Yang, "Testimony by LLMs," *AI & SOCIETY* 40(8): 6201–6213, 2025. DOI:10.1007/s00146-025-02366-y. *Substrate confidence: search-summary — body paywalled (Springer).* Argues LLM outputs can be a knowledge source if Reid's veracity/cautiousness principles are robustly met; critiques Lackey's assurance theory.
- **[petruzella-inconsistency-critique-2025]{24}** — Gerol Petruzella, "The Inconsistency Critique: Epistemic Practices and AI Testimony About Inner States," arXiv:2601.08850 (Dec 2025). *Substrate confidence: source-direct for bibliographic claims; search-summary for body.* The inconsistency critique; deploys Fricker's informant-vs-source and Goldberg's obligation account. (Coady absent from its bibliography — verified at body-fetch.)
- **[fierro-defining-knowledge-llm-2024]{25}** — Fierro, Dhar, Stamatiou, Søgaard & Garneau, "Defining Knowledge: Bridging Epistemology and Large Language Models," EMNLP 2024 (2024.emnlp-main.900). arXiv:2410.02499. *Substrate confidence: source-direct for the five definitions + epistemologist list; search-summary for the testimony-engagement question.* Reviews five definitions of knowledge; surveys ~100 philosophers/CS researchers (JTB-and-successors lineage — Sartwell, Nozick, Williamson, Zagzebski, Austin).
- **[lackey-learning-from-words-ndpr]{26}** — NDPR review of Jennifer Lackey, *Learning from Words: Testimony as a Source of Knowledge* (OUP, 2008). *Substrate confidence: source-direct for the NDPR review; cite-through for Lackey's primary text.* Characterizes Lackey's "hybrid" view and the generation-vs-transmission distinction.
- **[zhan-pies-ping]{40}** — Zhan, Fan, Huang, Guo & Huang, "Why Your Deep Research Agent Fails? On Hallucination Evaluation in Full Research Trajectory," arXiv:2601.22984. Engaged source-direct: **v1** (30 Jan 2026, "PIES" — the Explicit/Implicit × Planning/Summarization taxonomy + production hallucination rates across multiple deep-research agents) and **v2** (23 May 2026, "PING" — Propagation · Intent · Noise-induced · Grounding; the taxonomy reorganized by *locus of failure*). Load-bearing for the coordinate-system position: the v1→v2 rename + locus-of-failure reorganization (the de-anchoring trigger), and v2's independent adoption of the locus-of-failure cut (the convergence argument). *(Number {40} follows the append-only rule — added by extension after the Books block; resolution is by number, not file position.)*

- **[kim-correlated-errors-llms]{41}** — Kim, Garg, Peng & Garg, "Correlated Errors in Large Language Models," ICML 2025. arXiv:2506.07962. Load-bearing for the model-diversity property (§7): same-/same-family LLM evaluators share errors, *and* the most capable models correlate even across distinct architectures/providers — so cross-model review is partial decorrelation, not paradigm-independence. *Substrate confidence: search-summary (existence + abstract verified; findings via abstract/search).*
- **[zheng-judging-llm-as-judge]{42}** — Zheng, Chiang, et al., "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," NeurIPS 2023 (Datasets & Benchmarks). arXiv:2306.05685. The canonical LLM-as-judge "self-enhancement bias" (a judge favouring outputs like its own). *Substrate confidence: search-summary.*
- **[bommasani-picking-same-person-monoculture]{43}** — Bommasani, Creel, Kumar, Jurafsky & Liang, "Picking on the Same Person: Does Algorithmic Monoculture Lead to Outcome Homogenization?", NeurIPS 2022. arXiv:2211.13972. The component-sharing → outcome-homogenization lineage (shared training data/models → correlated failures) — the systemic parent of the correlated-verifier shape. *Substrate confidence: search-summary.*
- **[huang-llms-cannot-self-correct]{44}** — Huang, Chen, Mishra, Zheng, Yu, Song & Zhou, "Large Language Models Cannot Self-Correct Reasoning Yet," ICLR 2024. arXiv:2310.01798. Intrinsic self-correction (no external signal) fails and can degrade performance — the verifier must be *external/different*, grounding the cross-model fence over a same-model re-run. *Substrate confidence: search-summary.*
- **[chelli-hallucination-reference-accuracy]{45}** — Chelli, Descamps, Lavoué, et al., "Hallucination Rates and Reference Accuracy of ChatGPT and Bard for Systematic Reviews," *J. Medical Internet Research* 26:e53164 (2024). DOI:10.2196/53164. Keystone for `GR.9`: on *real* (non-hallucinated) papers, DOI accuracy was ~16–20% while titles were ~99% — the metadata tier fails independently of the existence tier. *Substrate confidence: search-summary.*
- **[walters-fabrication-citations-chatgpt]{46}** — Walters & Wilder, "Fabrication and errors in the bibliographic citations generated by ChatGPT," *Scientific Reports* 13:14045 (2023). DOI:10.1038/s41598-023-41032-5. ~24–43% of *real* citations carried substantive metadata errors (dates, authors, volume/pages) — metadata recall-sourcing on genuine sources. *Substrate confidence: search-summary.*
- **[li-good-abandonment-search]{47}** — Li, Huffman & Tokuda, "Good Abandonment in Mobile and PC Internet Search," SIGIR 2009: 43–50. DOI:10.1145/1571941.1571951. The good-vs-bad abandonment distinction; `AQ.4` is *bad abandonment* — quitting with the need unmet and the source still acquirable. *Substrate confidence: search-summary.*
- **[browne-cognitive-stopping-rules]{48}** — Browne, Pitts & Wetherbe, "Cognitive Stopping Rules for Terminating Information Search in Online Tasks," *MIS Quarterly* 31(1): 89–104 (2007). Stopping is a heuristic that can fire prematurely — grounds the mode-exhaustion fence as a stopping rule that resists firing while a stronger acquisition mode remains untried. *Substrate confidence: search-summary.*
- **[howard-information-value-theory]{49}** — Ronald A. Howard, "Information Value Theory," *IEEE Trans. Systems Science and Cybernetics* 2(1): 22–26 (1966). DOI:10.1109/TSSC.1966.300074. The canonical value-of-information formalization — the ex-ante "is more investigation worth it?" gate behind `decision_relevance`. *Substrate confidence: search-summary.*
- **[stigler-economics-of-information]{50}** — George J. Stigler, "The Economics of Information," *J. Political Economy* 69(3): 213–225 (1961). DOI:10.1086/258464. Search theory's marginal stop (search until marginal expected benefit equals marginal cost) — the diminishing-returns basis for right-sizing depth. *Substrate confidence: search-summary.*

## Books (all rights reserved)

- **[saussure-course-in-general-linguistics]{27}** — Ferdinand de Saussure, *Course in General Linguistics* (posthumous 1916; Bally & Sechehaye, from 1906–11 lecture notes). Engaged in the Roy Harris trans., Bloomsbury Revelations 2013 (Payot 1922 pagination in margins). Engaged: Intro ch.3–4; Part One; Part Two ch.3–5. Primary: sign-as-dyad; invariability under arbitrariness; language as form not substance; value as differential.
- **[peirce-essential-peirce-vol-2]{28}** — C. S. Peirce, *The Essential Peirce, Vol. 2 (1893–1913)*, ed. Peirce Edition Project, Indiana UP, 1998. Engaged: ch.2 (*What Is a Sign?*); ch.21 (*Nomenclature… Triadic Relations*); ch.22 (*New Elements*); ch.32 (*Letters to Lady Welby*). Primary: Representamen/Object/Interpretant triad; Immediate/Dynamic/Normal Interpretant; Immediate/Dynamical Object; the endless-series-with-limit passage.
- **[wittgenstein-philosophical-investigations]{29}** — Ludwig Wittgenstein, *Philosophical Investigations*, 3rd ed., trans. Anscombe, Blackwell, 1953. Engaged: §§1–7; §§19–23; §§185–202; §§217–242. Primary: §199, §201–202 (rule-following as practice); §241 (agreement in form of life).
- **[gadamer-truth-and-method]{30}** — Hans-Georg Gadamer, *Truth and Method* (1960), rev. 2nd ed., trans. Weinsheimer & Marshall, Bloomsbury Academic, 2013. Engaged: Part Two, ch.4 ("Elements of a Theory of Hermeneutic Experience"), pp. 278–376.
- **[ricoeur-interpretation-theory]{31}** — Paul Ricoeur, *Interpretation Theory: Discourse and the Surplus of Meaning*, Texas Christian UP, 1976. Engaged: all four essays + Intro/Conclusion; primary focus Essay 4 ("Explanation and Understanding," pp. 71–88) and Conclusion (pp. 89–95).
- **[ahrens-smart-notes]{32}** — Sönke Ahrens, *How to Take Smart Notes*, 2nd ed., 2022. Engaged: Introduction; ch.2; ch.3; Appendix (§§1–15). Primary for the directionality position: §6 ("Simplicity Is Paramount") and §10.1 ("Read With a Pen in Hand").
- **[luhmann-communicating-slip-boxes]{33}** — Niklas Luhmann, "Communicating with Slip Boxes" (1992; German orig. c.1980/81), trans. Manfred Kuehn. <https://luhmann.surge.sh/communicating-with-slip-boxes>.
- **[coady-testimony]{34}** — C. A. J. Coady, *Testimony: A Philosophical Study*, Clarendon Press / OUP, 1992 (pbk 1994). Engaged source-direct: Parts I–III in full, Parts IV–V selectively. Load-bearing: ch.2 (natural testimony; believing-the-witness); ch.4 (Reductionist Thesis critique); ch.7 (Reid); ch.8 (testimony as basic, non-inferential); ch.11 (anti-decay argument vs Locke).
- **[glaser-discovery-of-grounded-theory]{35}** — Barney Glaser & Anselm Strauss, *The Discovery of Grounded Theory*, 1967 (AldineTransaction reprint 1999). Engaged: ch.I (founding claim, pp.1–19); ch.III (Theoretical Sampling, pp.45–79); ch.V (Constant Comparative Method, pp.101–125).
- **[charmaz-constructing-grounded-theory]{36}** — Kathy Charmaz, *Constructing Grounded Theory*, SAGE, 2006. Engaged: ch.1–8.
- **[corbin-basics-of-qualitative-research]{37}** — Juliet Corbin & Anselm Strauss, *Basics of Qualitative Research*, 4th ed., SAGE, 2015. Engaged: ch.1, 2, 5, 7, 8.
- **[miles-qualitative-data-analysis]{38}** — Miles, Huberman & Saldaña, *Qualitative Data Analysis: A Methods Sourcebook*, 4th ed., SAGE, 2020. Engaged: ch.1, 2, 4, 11.
- **[cochrane-handbook]{39}** — Higgins & Thomas (sr. eds.), *Cochrane Handbook for Systematic Reviews of Interventions*, v6.5, Cochrane, 2024. <https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current>. Engaged: ch. I, II, X, XII, XIII, XIV (GRADE / Summary of findings).
- **[raiffa-schlaifer-applied-statistical-decision-theory]{51}** — Howard Raiffa & Robert Schlaifer, *Applied Statistical Decision Theory*, Harvard Business School, 1961. The operational machinery + the canonical terms EVPI / **EVSI** (expected value of sample information — the value of a *specific, costly depth* of inquiry) behind `decision_relevance`'s depth-right-sizing. *Substrate confidence: search-summary (existence verified; engaged via catalog/secondary).*
- **[wald-sequential-analysis]{52}** — Abraham Wald, *Sequential Analysis*, Wiley, 1947 (the SPRT). Optimal-stopping / when-to-stop: observe sequentially, stop the moment accumulated evidence settles the decision — grounds the dynamic stop-deepening complement to the yield gate. *Substrate confidence: search-summary.*