Bibliography

Bibliography — the whole paper, one list, grouped by source

RES-1188 ruled: one bibliography for the whole paper, grouped by source, not by chapter. This document is that consolidation. It absorbs the existing 37-entry Generative-lane list (AIJM-S12_BIBLIOGRAPHY-GENERATIVE_v0.1.md, entries F1–F7, C1–C6, B1–B5, A1–A4, E1, G1–G3, D1–D2, N1–N9 — that document's own numbering is kept as a cross-reference key below rather than re-verified) and adds every source named in prose across the other eight live sections: Intro (v0.20), Field-history (v0.7), How-these-systems-work (v0.7), Method-Predictive (v0.18), Method-Agentic (v0.11), Terminology (v0.12), Appendix (v0.9) and Conclusion (v0.18).

Constraint honored throughout. Every entry below comes from a document already on disk in this project — a case dossier, a source-pull file, a research dossier, a verification pass, or the generative bibliography itself. Nowhere below is a URL or date supplied from this assistant's own knowledge of the source. Where the project's own record holds no link, the entry says so, in bold, and that is reported as a finding rather than completed. Two sentences in the live paper — the Sculley NeurIPS quotation and one unattributed Reiter quotation carried since the trends-brief era — have no locatable source anywhere in the project's record at all, not even a thin one; both are flagged under "Passages with no usable citation" at the end, because that absence is more serious than a thin entry.

Two decisions, made before building the list

(a) Short-form in-text reference format

Decision: (Source, short tag, Year) — the publishing organization or named author, a two-to- five-word tag distinguishing that document from others by the same source, and the year. Examples: (Bloomberg, LQA launch release, 2016) · (Bloomberg, BAI white paper, 2020) · (Reiter, "still not trustworthy", 2024) · (Chua, Semafor column, 2025) · (Sculley et al., NeurIPS 2015) · (SEC, BVAL order, 2023). Where a source has more than one document from the same source and year, add a letter: (Reiter, 2026a) / (Reiter, 2026b), keyed to the numbered entry below.

Why this form and not another. The paper's own prose already names people and organizations in-line ("John Micklethwait, Bloomberg, December 2024"), so a numbered footnote scheme would duplicate work the prose already does and would force renumbering on every edit. A full author-year form (APA-style) fails here because a third of the record's entries have no named author — a press release, a product page, a corporate blog post — and "Bloomberg, n.d." repeated sixty times is less useful than a short descriptive tag a reader can match to the entry by eye. The tag form also matches the register distinction this whole exercise turned on: (Bloomberg, ...) and (Press Gazette, reporting Bloomberg, ...) read differently on the page, which is the point — collapsing them into one undifferentiated author-year form would erase exactly the distinction RES-1188 asked this document to preserve. This is a recommendation for the format only; per scope, no in-text marker is added to the body prose in this pass.

(b) Register vocabulary — reconciled from 11 down to 9

The generative bibliography's own closing note states the authoritative six-category scheme this project declared for its case records: first-party Bloomberg, peer-reviewed, trade press, mainstream press, aggregator-relayed, sponsored content (AIJM-S12_BIBLIOGRAPHY-GENERATIVE_v0.1.md, opening paragraph and closing register-mix note). That same document then needed five more to cover sources outside the case records: practitioner expert blog, preprint (not peer-reviewed), research-institute report, conference paper held without a locator, and user-generated — eleven in total, "not a project source" recorded twelfth as an exclusion flag rather than a register.

Pulling in the Predictive and Agentic lanes' own dossiers, plus the newly reviewed Field-history, Terminology, Appendix and Conclusion prose, adds two categories that recur too often and carry too different an evidentiary weight to fold into any of the six: a U.S. government regulatory action (the SEC's BVAL order) and preprints that are never going to become peer-reviewed within this paper's window (BloombergGPT, the Hagar/Diakopoulos/Gilbert paper, several GHG-adjacent arXiv postings). This document's reconciled set is nine registers, collapsed and extended as follows:

1. First-party Bloomberg — unchanged from the original six.
2. Peer-reviewed — unchanged, and now also absorbs the generative list's "conference paper held without a locator" (F5, Sculley) — it is peer-reviewed (NeurIPS 2015) regardless of whether this project holds its locator; "held without a locator" was describing this project's access, not the source's own status, so it does not need to be its own register.
3. Preprint / working paper, not peer-reviewed — kept as its own category rather than folded into peer-reviewed, because the generative bibliography's own BloombergGPT entry (G1) is emphatic that this distinction is load-bearing for how the paper cites it.
4. Practitioner expert blog / independent commentary — kept as its own category. It recurs across every lane that touches Reiter, Harrell, Karpathy or Chua's personal-voice pieces, and folding it into trade press (as the six-category scheme would require) would misstate the authorship — these are the source's own first-person statement, not a reporter's account of one.
5. Trade press — unchanged, and now absorbs the generative list's "research-institute report" (F7, the Tow Center report cited inside Graefe 2016) — a nonprofit research-institute report about the trade functions evidentially the same way a trade-press piece does here: independent of Bloomberg, written for an industry audience, not peer-reviewed.
6. Mainstream press — unchanged.
7. Aggregator-relayed — unchanged, and now absorbs the generative list's "user-generated" category (the single Apple App Store review, A5) — both are third-party content produced for a purpose other than reporting on the source, and both reach this study at one or more removes from anyone who could be asked a follow-up question. Collapsing them loses only the distinction between an anonymous relay and an anonymous review, which this project's own record does not otherwise use anywhere else.
8. Regulatory — new. The SEC's BVAL order is categorically different from a trade-press account of that order (S06_PULL-IBVAL S6, S7): it is the primary enforcement document itself, carries legal weight none of the other eight registers share, and appears only in the Predictive lane and the Conclusion.
9. Sponsored content — unchanged (a genuine register in the LQA and GHG pulls' Risk.net awards entries, even though the generative list's own case record used it zero times).

"Not a project source" stays outside the register system entirely, as an exclusion flag rather than a category — a document identified as contamination (aicerts.ai, G3) is not weighed as evidence at all, so giving it a register would imply it belongs in this list on the same footing as the other 90-plus entries. It stays here only because the paper names and rebuts it.

Bloomberg — first-party channels

Every document Bloomberg published itself, self-hosted or through a wire service, across all four lanes that cite Bloomberg. Grouped chronologically by first attestation.

1. PR Newswire (carrying Bloomberg), "Bloomberg Delivers First Quantitative Model For Calculating Liquidity Risk." 9 March 2016, dateline "NEW YORK, March 9, 2016." No author (corporate release). https://www.prnewswire.com/news-releases/bloomberg-delivers-first-quantitative-model-for-calculating-liquidity-risk-300233068.html. Register: first-party Bloomberg. Lanes: Predictive. Source of the LQA launch date and the launch description. (AIJM-S06_PULL-LQA_v0.1.md S1.)
2. Bloomberg L.P., "Bloomberg Regulatory Solutions" client deck. March 2016 (title slide); copyright block reads "©2015 Bloomberg Finance L.P." No author. https://data.bloomberglp.com/professional/sites/12/2017/05/Reg_solutions.pdf. Register: first-party Bloomberg (technical/marketing). Lanes: Predictive. Source of the "Problem : A lack of trade data..." three-line client-deck quote confirmed by the S19 Predictive audit. (AIJM-S06_PULL-LQA_v0.1.md S9.)
3. Bloomberg L.P., LQA Fact Sheet. Copyright block "©2019 Bloomberg 673876 DIG 1219," i.e. December 2019. No author. https://data.bloomberglp.com/professional/sites/10/LQA-Fact-Sheet-December-2019.pdf. Register: first-party Bloomberg. Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S10.)
4. Bloomberg L.P., "Liquidity & regulation: SEC 22e-4" fact sheet. Copyright block "©2017 Bloomberg L.P. S771321102 0517." No author. https://data.bloomberglp.com/professional/sites/10/SEC-22e-4-Fact-Sheet.pdf. Register: first-party Bloomberg (regulatory-facing). Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S11.)
5. Bloomberg L.P., JFSA/JITA Liquidity Guidelines Brochure. Copyright block "©2021 Bloomberg 1006892 DIG 0421." No author. https://assets.bbhub.io/professional/sites/10/JFSA-JITA-Liquidity-Guidelines-Brochure.pdf. Register: first-party Bloomberg (regulatory-facing). Lanes: Predictive, Conclusion. Source of the bucket boundaries and of the Conclusion's directly quoted "can be fully customized to reflect the fund manager's determination of a 'significant' market impact." (AIJM-S06_PULL-LQA_v0.1.md S12.)
6. Bloomberg Professional Services, LQA product page. meta-article:modified_time 26 February
7. No author. https://professional.bloomberg.com/products/risk/lqa. Register: first-party Bloomberg marketing. Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S13.)
8. John Micklethwait, staff memo, reprinted in full by Talking Biz News. 26 April 2016 (article:published_time 2016-04-26T16:18:23Z). https://talkingbiznews.com/they-talk-biz-news/micklethwait-bloomberg-to-expand-story-automation-efforts/. Register: trade press reprinting a first-party document — grouped here by content-origin, listed again under Talking Biz News below by venue. Lanes: Generative (C2), Agentic. Full detail: generative bibliography C2.
9. Bloomberg L.P., "Bloomberg Media's Innovation Lab launches 'The Bulletin' – An AI-Powered News Feed for Bloomberg Mobile App Users." bloombergmedia.com press release, 18 September 2018 (article:modified_time 2018-09-17T22:03:14Z). https://www.bloombergmedia.com/press/bloomberg-medias-innovation-lab-launches-bulletin/. Register: first-party Bloomberg press release. Lanes: Generative (B1). Full detail: generative bibliography B1. Title and
10. Karen Johnson, BHIVE / Medium. 13 December 2018. https://medium.com/bhive/how-bloomberg-medias-bhive-team-keeps-users-at-the-center-4842042cd2e8. Post title not stated in the project's own record; URL slug reads "how-bloomberg-medias-bhive-team- keeps-users-at-the-center." Register: first-party Bloomberg team blog. Lanes: Generative (B3). Full detail: generative bibliography B3.

11. Karen Johnson, "BHIVE 2020 Look Ahead: 5 News Experience Goals," BHIVE / Medium. 27 February 2020 ("Last published Feb 27, 2020" per the page itself, confirming this is BHIVE's final post). https://medium.com/bhive/bhive-2020-look-ahead-5-news-experience-goals-e295372eee26. Register: first-party Bloomberg team blog. Lanes: Generative (B4). Full detail: generative bibliography B4.

12. Mohammad Fesanghary and Arun Verma, "Predictive Analysis of Bloomberg Automated Intelligence." Bloomberg Quant Research white paper, June 2020 (by its own copyright block). https://data.bloomberglp.com/professional/sites/10/BQUANT_AutomatedInsights.pdf. Register: first-party Bloomberg quant white paper — marketing register, not peer-reviewed. Lanes: Generative (A1), Agentic (same BAI case, per the S19 agentic audit). Full detail: generative bibliography A1.

13. "Using Bloomberg automated news stories to predict market events," Bloomberg Professional Insights. Body-stated 29 June 2020; CMS date field 2022-03-31. https://www.bloomberg.com/professional/insights/data/using-bloomberg-automated-news-stories-to-predict-market-events/. Author not named in record. Register: first-party Bloomberg marketing. Lanes: Generative (A2), Agentic. Full detail: generative bibliography A2.
14. Bloomberg Terminal news product page. https://www.bloomberg.com/professional/products/bloomberg-terminal/news/, article:modified_time 5 January 2024. No author; no title beyond the product name. Register: first-party Bloomberg product marketing. Lanes: Generative (A3), Agentic. Full detail: generative bibliography A3.
15. "Bloomberg Pro Tips: Get instant snapshots of ETFs and more with AID," Bloomberg Professional. CMS date 2024-10-17T19:49:22 (index date matches exactly). https://www.bloomberg.com/professional/insights/markets/bloomberg-pro-tips-get-instant-snapshots-of-etfs-and-more-with-aid-go/. No individual byline (Pro Tips video series). Register: first-party Bloomberg marketing. Lanes: Generative (A4), Agentic.
16. Quinonez and Meij, "A new era of AI-assisted journalism at Bloomberg." AI Magazine 45(2):187–199, first published 5 June 2024. DOI 10.1002/aaai.12181. https://onlinelibrary.wiley.com/doi/full/10.1002/aaai.12181. Register: peer-reviewed and first-party Bloomberg simultaneously (both authors are Bloomberg executives). Lanes: Generative (C1), Agentic (confirmed by the S19 agentic audit as the source of "more than a hundred currencies" and "hundreds of software bots"), Conclusion (the "reporters would previously wait for an earnings press release..." block quote and "months of research by an editorial domain expert..."). This is the single most cross-lane-cited document in the paper. Full detail: generative bibliography C1.
17. "Bloomberg Launches AI-Powered Earnings Call Summaries," PR Newswire. 22 January 2024. https://www.prnewswire.com/news-releases/bloomberg-launches-ai-powered-earnings-call-summaries-302040670.html. Register: first-party Bloomberg press release. Lanes: Generative (D1). Full detail: generative bibliography D1.
18. "Bloomberg Accelerates Financial Analysis with Gen AI-Powered Document Insights," PR Newswire / Bloomberg. 7 April 2025 (PRN transmission ID 302421875; Bloomberg's own page dates it 6 April 2025, one day earlier). PRN: https://www.prnewswire.com/news-releases/bloomberg-accelerates-financial-analysis-with-gen-ai-document-insights-302421875.html. Bloomberg-hosted: https://www.bloomberg.com/company/press/bloomberg-accelerates-financial-analysis-with-gen-ai-document-insights/. Register: first-party Bloomberg press release. Lanes: Generative (D2).
19. Bloomberg LP — Tech At Bloomberg, "How Bloomberg Uses Machine Learning For Intraday Pricing Transparency." 4 December 2023. Quotes David Krein, Timothy O'Brien, Junling Wang, Camilo Ortiz. https://www.bloomberg.com/company/stories/how-bloomberg-uses-machine-learning-intraday-pricing-transparency. Register: first-party Bloomberg engineering. Lanes: Predictive, Conclusion (Ortiz "scale better" quote; the < 6-hour retrain figure; the 6–7 billion ticks/day and 300,000 ticks/second figures). (AIJM-S06_PULL-IBVAL_v0.1.md S1.)
20. PR Newswire (carrying Bloomberg), "Bloomberg Launches Front Office Pricing Solution..." 17 October 2023, quotes Tony McManus. https://www.prnewswire.com/news-releases/bloomberg-launches-front-office-pricing-solution-providing-greater-visibility-into-fixed-income-trading-301957528.html. Register: first-party Bloomberg marketing. Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S2.)
21. PR Newswire (carrying Bloomberg), "Bloomberg Expands IBVAL Front Office Pricing to Europe..." 16 April 2024, quotes Eric Isenberg. https://www.prnewswire.com/news-releases/bloomberg-expands-ibval-front-office-pricing-to-europe-backed-by-contextual-metadata-for-greater-transparency-302116423.html. Register: first-party Bloomberg marketing. Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S3.)
22. Bloomberg LP (asset CDN), "Fixed Income Cash Pricing Sources" fact sheet. Copyright block "©2016 ... S739104556 1216." https://data.bloomberglp.com/professional/sites/10/Fixed-Income-Cash-Pricing-Sources.pdf. Register: first-party Bloomberg marketing. Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S9.)
23. Bloomberg Professional Services, IBVAL Front Office product page. Modified 9 February 2024. https://professional.bloomberg.com/products/data/enterprise-catalog/pricing/ibval-front-office/. Register: first-party Bloomberg marketing. Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S10.)
24. Bloomberg Finance L.P., "Bloomberg's Greenhouse Gas Emissions Estimates Model: A Summary of Challenges and Modeling Solutions." Copyright "©2022," document code "2238051 0223" (reads February 2023). No author. https://assets.bbhub.io/professional/sites/10/GHG.pdf. Register: first-party Bloomberg marketing/technical. Lanes: Predictive. Source of the "nearly 70,000 companies," "over 100,000... vs approximately 4,500 self-reported," and "over 800 individual features" figures. (AIJM-S06_PULL-GHG_v0.1.md S3.)
25. PR Newswire (carrying Bloomberg), "Bloomberg Increases Carbon Emissions Data to Cover 100,000 Companies." 24 October 2022, quotes Brad Foster. https://www.prnewswire.com/news-releases/bloomberg-increases-carbon-emissions-data-to-cover-100-000-companies-301656554.html. Register: first-party Bloomberg marketing. Lanes: Predictive. (AIJM-S06_PULL-GHG_v0.1.md S4.)
26. Bloomberg L.P. (Tech at Bloomberg), "Our published AI research" index page. Undated; content current to at least 2026 (page robot-blocks direct fetch; confirmed via two independent search-index snapshots showing 2026 entries, including a July 2026 ACL 2026 papers post — GLM/Perplexity verification, 2026-09-12). https://www.bloomberg.com/company/values/tech-at-bloomberg/artificial-intelligence-ai/our-published-ai-research/. Register: first-party Bloomberg engineering. Lanes: Predictive. (AIJM-S06_PULL-GHG_v0.1.md S5.)
27. "Bloomberg Launches Gen AI Summarization for News Content." bloomberg.com/company/press, page dated 15 January 2025. https://www.bloomberg.com/company/press/bloomberg-launches-gen-ai-summarization-for-news-content. Author not named. Register: first-party Bloomberg press release. Lanes: Generative (N1). Full detail: generative bibliography N1.
28. "Investors Harness Bloomberg's Expanded AI Tools to Discover and Summarize News." 24 November
    2025. https://www.bloomberg.com/company/press/investors-harness-bloombergs-expanded-ai-tools-to-discover-and-summarize-news/ — confirmed rendering in full (GLM/Perplexity verification, 2026-09-12): on-record quote from Adela Quinones (Global Head of News Discovery), the 175,000-source / 1.5 million-stories-daily figures, launch of AI Summary for Company News and expanded AI-Powered News Summaries. The original research-time "would not render on five attempts" note was a transient bloomberg.com anti-bot block, not a property of the source — corrected from aggregator-relayed to first-party. Register: first-party Bloomberg press release. Lanes: Generative (N7). Full detail: generative bibliography N7.
29. "Bloomberg Pro Tips: Get up to speed on any company with AI Summary in CN," Bloomberg Professional. CMS date 25 June 2026 (exact match). https://www.bloomberg.com/professional/insights/markets/bloomberg-pro-tips-get-up-to-speed-on-any-company-with-ai-summary-in-cn/. No individual byline. Register: first-party Bloomberg marketing. Lanes: Generative (N9). This entry's claim (the product was still being marketed in June 2026) does not depend on subject matter, and the CMS date is what identifies the correct page — a second June 2026 Pro Tips episode exists ("Go from questions to answers faster with ASKB," 12 June 2026) but its date does not match this entry's CMS timestamp, so it is not the source.
30. Bloomberg (self-published), ASKB launch release. 23 February 2026 — "Bloomberg introduced ASKB, a powerful new conversational AI interface, now in beta..." Best verbatim reprint at https://www.marketsmedia.com/bloomberg-introduces-agentic-ai-to-the-terminal/, corroborated at https://www.thetradenews.com/bloomberg-embeds-agentic-ai-into-the-terminal/. Register: first-party Bloomberg press release, reaching this study via verbatim trade-press reprint. Lanes: Agentic. (AIJM-S04_CASE-THREADS-BC_v0.1.md, CASE C2.)
31. "Bloomberg Unveils ASKB Roadmap for Clients to Augment their Investment Process with Agentic AI," Bloomberg (self-published). 16 April 2026 — confirms ASKB's sell-side/independent research provider coverage expanding from "over 800" to "over 1,200." https://www.bloomberg.com/professional/insights/press-announcement/bloomberg-unveils-askb-roadmap-for-clients-to-augment-their-investment-process-with-agentic-ai/. Register: first-party Bloomberg press release. Lanes: Agentic.
32. "Bloomberg Extends ASKB on Mobile; Investment Ideas Don't Stop When Financial Professionals Leave the Desk," Bloomberg via PR Newswire. 18 August 2026. PR Newswire transmission ID 302853450 (embedded in the URL). https://www.prnewswire.com/news-releases/bloomberg-extends-askb-on-mobile-investment-ideas-dont-stop-when-financial-professionals-leave-the-desk-302853450.html. Quotes Fahd Arshad, Head of Product for ASKB. Register: first-party Bloomberg press release, wire-distributed. Lanes: Agentic.
33. "AI on Bloomberg," Bloomberg product page. https://professional.bloomberg.com/products/bloomberg-terminal/ai/, retrieved 20 August 2026; confirmed resolving at this exact path with title "AI on Bloomberg" and text introducing ASKB (beta). Register: first-party Bloomberg marketing. Lanes: Agentic.
34. Remaining generative-lane Bloomberg entries carried forward without re-verification, cross-listed by lane where the audits confirm shared use: B2 (no — Digiday, listed under Trade press), C3 (Poynter reprint, listed under Trade press), C4 (Press Gazette, listed under Trade press, cross-listed Agentic/Conclusion), C5 (Global Editors Network Q&A, listed under Trade press, cross-listed Agentic), C6 (NYT, listed under Mainstream press), G2 (A-Team Insight, listed under Trade press, cross-listed Agentic), G3 (aicerts.ai, listed under Excluded), N2–N6, N8 (listed under Trade/mainstream/aggregator press below).

Regulatory

34. U.S. Securities and Exchange Commission, Office of Public Affairs, press release 2023-14. 23 January 2023. Quotes Osman Nawaz. https://www.sec.gov/newsroom/press-releases/2023-14. Underlying order at https://www.sec.gov/litigation/admin/2023/33-11150.pdf (not fetched by this project — cited via the press release only). Register: regulatory. Lanes: Predictive, Conclusion. Source of the $5 million BVAL penalty and the "at least 2016 through October 2022" period; the lane correctly notes the order predates IBVAL. (AIJM-S06_PULL-IBVAL_v0.1.md S7.)

Peer-reviewed and preprint literature

35. Han, Gopal, Ouyang and Key, "Machine Learning for GHG Emissions Prediction." ICML 2021 workshop, Climate Change AI (CCAI). Venue page https://www.climatechange.ai/papers/icml2021/4; preprint https://arxiv.org/abs/2109.04318 (v1, 9 Sept 2021). Register: peer-reviewed (reviewed workshop paper). Lanes: Predictive. THE primary source for the GHG case's 2.27%, 24,052/3,960, 619,703/61,467 and 97%/20% figures. (AIJM-S06_PULL-GHG_v0.1.md S1/S2.)
36. Ouyang and Key, sibling paper on Patterned Dropout / missingness shift. NeurIPS 2021 DistShift workshop. https://arxiv.org/abs/2111.10344. Register: peer-reviewed (workshop). Lanes: Predictive (background for the GHG methodology). (AIJM-S06_PULL-GHG_v0.1.md S6.)
37. Nguyen, Diaz-Rainey, Kitto, McNeil, Pittman and Zhang, "Scope 3 emissions: Data quality and machine learning prediction accuracy." PLOS Climate, published 15 November 2023. DOI 10.1371/journal.pclm.0000208. https://journals.plos.org/climate/article?id=10.1371%2Fjournal.pclm.0000208. Register: peer-reviewed, third-party — with a declared competing interest: three authors work for EMMI, a company selling carbon risk data. Lanes: Predictive. Source of "remain largely a black box" and the disclosed-vs-modeled Scope 3 distinction. (AIJM-S06_PULL-GHG_v0.1.md S7.)
38. Maddela, Kulkarni and Preoţiuc-Pietro, "EntSUM." ACL 2022 (main conference). https://aclanthology.org/2022.acl-long.237/; preprint https://arxiv.org/abs/2204.02213 (v1 only, 5 April 2022). Register: peer-reviewed. Lanes: Generative (E1). Full detail: generative bibliography E1.
39. "BloombergGPT: A Large Language Model for Finance." https://arxiv.org/abs/2303.17564, v3 (21 December 2023). Register: arXiv preprint, not peer-reviewed (no Journal-ref field). Lanes: Generative (G1), Conclusion (512 GPUs, 53 days of training figures). Full detail: generative bibliography G1.
40. Hagar, Diakopoulos and Gilbert. arXiv:2509.25494. Title, author given names and date not in record. Register: preprint, logged REPORTED not VERIFIED (paraphrase, not a quotation read from the paper). Lanes: Generative (F1). Full detail: generative bibliography F1.
41. D. Sculley et al., "Hidden Technical Debt in Machine Learning Systems." NeurIPS 2015, pp. 2503–2511. https://proceedings.neurips.cc/paper/2015/file/86df7dcfd896fcaf2674f757a2463eba-Paper.pdf (also indexed at https://www.semanticscholar.org/paper/Hidden-Technical-Debt-in-Machine-Learning-Systems-Sculley-Holt/1eb131a34fbb508a9dd8b646950c65901d6f1a5b). Both quotations Field-history uses are confirmed verbatim against the primary text (GLM/Perplexity verification, 2026-09-12): the §5 glue-code line ("Because a mature system might end up being (at most) 5% machine learning code and (at least) 95% glue code, it may be less costly to create a clean native solution rather than re-use a generic package") and the Figure 1 caption. Register: peer-reviewed (NeurIPS main conference), first-party. Lanes: Generative (F5), Field-history, Terminology, Conclusion. Cross-check: AIJM-S11_RESEARCH-DOSSIER_practitioner-recency_v0.1.md confirms Sculley's identity (now Kaggle CEO); this entry's own prior "no locator" note is superseded by this correction.
42. D. Sculley, William Cukierski, et al., "Position: AI Competitions Provide the Gold Standard for Empirical Rigor in GenAI Evaluation." arXiv:2505.00612, posted 1 May 2025; ICML 2025 (Position Paper Track), oral, 16 July 2025. https://arxiv.org/abs/2505.00612 · https://icml.cc/virtual/2025/poster/40139. Register: peer-reviewed / preprint (conference accepted, arXiv version cited). Not used in the live paper text — surfaced only in the S11 recency dossier as an adjacent, not directly on-point, 2025 Sculley source. Lanes: none currently (background research only). Listed here because it is the only 2024–2026 Sculley document this project holds a URL for. (AIJM-S11_RESEARCH-DOSSIER_practitioner-recency_v0.1.md.)

Trade and financial press

43. Benjamin Mullin, "Bloomberg EIC: Automation is 'crucial to the future of journalism,'" Poynter. 27 April 2016 (one day after the 26 April memo). https://www.poynter.org/tech-tools/2016/bloomberg-eic-automation-is-crucial-to-the-future-of-journalism/. Register: trade press. Lanes: Generative (C3). Full detail: generative bibliography C3.
44. Bron Maher, Press Gazette. 18 December 2024. https://pressgazette.co.uk/media_business/john-micklethwait-bloomberg-paywalls-ai-zxc/. Register: trade press. Lanes: Generative (C4), Agentic (Micklethwait pre-launch quotations), Conclusion (the "roughly the same number of people to look at earnings" quotation is attributed in the Conclusion simply to "John Micklethwait, Bloomberg, December 2024" — this is its source). Full detail: generative bibliography C4.
45. "Bloomberg's Micklethwait: 'Future of news to be more digital, personalised, automated, paid for,'" Global Editors Network / Medium. 18 October 2018. https://medium.com/global-editors-network/bloombergs-micklethwait-future-of-news-to-be-more-digital-personalised-automated-paid-for-9d6460e757f4. Byline is the institutional "Global Editors Network"; no individual interviewer is named on the page (GEN's then-CEO Bertrand Pecquerie is the likely questioner but is not credited there — not stated as fact). Register: trade press Q&A. Lanes: Generative (C5), Agentic (byline-disclosure date), Field-history/How-these-systems-work (cross-checked, not separately quoted). Full detail: generative bibliography C5.
46. "Bloomberg Media has a robot writing story summaries," Digiday. 21 September 2018 (exact date match). https://digiday.com/media/bloomberg-media-robot-writing-story-summaries/. Staff byline (no individual credit). Register: trade press. Lanes: Generative (B2). Full detail: generative bibliography B2.
47. Wayne Barlow, interviewed by A-Team Insight. 28 April 2026. https://a-teaminsight.com/blog/inside-bloombergs-askb-roadmap-the-data-moat-behind-the-ai/. Register: trade press carrying an on-record first-party executive. Lanes: Generative (G2, BloombergGPT "not used at all in any of our products" quote), Agentic (the "we expect the human to be the one directing" and post-processing/hallucination-check quotes, plus the "orchestrated portfolio" analyst framing that the ASKB re-extraction pass used to strike the fabricated "core innovation" claim), Conclusion. This is the single most cross-lane-cited trade-press interview in the paper. Full detail: generative bibliography G2.
48. Bron Maher / WatersTechnology, John Brazier, "Bloomberg launches liquidity assessment tool for bond liquidity risk." 9 March 2016. https://www.waterstechnology.com/trading-tech/2450311/bloomberg-launches-liquidity-assessment-tool-for-bond-liquidity-risk. Register: trade press. Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S3.)
49. Terry Flanagan, Markets Media, "Bloomberg Delivers First Quantitative Model..." 9 March 2016. https://www.marketsmedia.com/bloomberg-delivers-first-quantitative-model-calculating-liquidity-risk/. Register: first-party marketing republished as trade press (verbatim republication of the launch release). Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S2.)
50. Pablo C Conde-Herman, Structured Retail Products. 16 March 2016. https://www.structuredretailproducts.com/insights/21955/bloomberg-releases-cross-asset-security-level-liquidity-risk-model. Register: trade press (editorial rewrite of the launch release). Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S4.)
51. Risk.net (Buy-side Awards 2016). No byline. 1 December 2016. https://www.risk.net/asset-management/2478745/liquidity-risk-technology-provider-year-bloomberg. Register: trade press — award coverage, unlabelled, built on a vendor interview (Naz Quadri). Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S5.)
52. Sebastian Day, Risk.net, "Quants turn to machine learning to model market impact." 5 April
    2017. https://www.risk.net/asset-management/4644191/quants-turn-to-machine-learning-to-model-market-impact. Register: trade press (editorial feature). Lanes: Predictive. Source of the Naz Quadri quotations and the Jean-Philippe Bouchaud material — Bouchaud's first name is not given in this article per the S19 predictive audit's own check; it is carried in the paper as a longstanding, unverified addition and flagged [UNVERIFIED] rather than confirmed here. (AIJM-S06_PULL-LQA_v0.1.md S6.)
53. Risk.net (Asia Risk Awards 2021). "Risk Staff." 27 September 2021. https://www.risk.net/awards/7879741/liquidity-risk-solution-of-the-year-bloomberg. Register: trade press — award coverage, unlabelled, built entirely on vendor quotes (Brad Foster). Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S7.)
54. Risk.net (Risk Markets Technology Awards 2026). 16 December 2025. https://www.risk.net/awards/7962728/market-liquidity-risk-product-of-the-year-bloomberg. Register: sponsored content — explicitly tagged iim-sponsored: sponsored on the page itself, and the record notes it carries an "LLM-generated summary box, 'edited by humans.'" Lanes: Predictive. (AIJM-S06_PULL-LQA_v0.1.md S8.)
55. Environmental Finance (Field Gibson Media), Sustainable Investment Awards 2024. No byline. https://www.environmental-finance.com/content/awards/sustainable-investment-awards-2024/winners/esg-data-provider-of-the-year-americas-bloomberg-esg-data.html. Register: trade press — awards coverage, carrying vendor-supplied claims; checked for Advertisement/Sponsored tags: none present, so it is not classed sponsored despite the awards format. Lanes: Predictive. (AIJM-S06_PULL-GHG_v0.1.md S8.)
56. Michael Thrasher, Institutional Investor. 17 October 2023. https://www.institutionalinvestor.com/article/2cc01z896579jxju9f08w/portfolio/bloomberg-says-its-using-machine-learning-to-deliver-near-real-time-bond-prices. Register: trade press (original reporting, adversarial questions, launch-day). Lanes: Predictive, Conclusion (the "combination of machine learning and more traditional rules-based models" quote). (AIJM-S06_PULL-IBVAL_v0.1.md S4.)
57. Alex Pugh, The DESK (Markets Media Europe). 16 April 2024. https://www.fi-desk.com/bloomberg-expands-ibval-front-office-pricing-solution/. Register: trade press (original interview with Eric Isenberg). Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S5.)
58. Dan Barnes, The DESK (Markets Media Europe). 24 January 2023. https://www.fi-desk.com/sec-fines-bloomberg-us5-million-for-failure-to-disclose-single-quotes-as-bval-source/. Register: trade press. Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S6.)
59. A-Team Insight, unbylined brief. 18 October 2023. https://a-teaminsight.com/briefs/bloomberg-adds-intraday-bval-front-office-pricing-solution-for-fixed-income-trading/. Register: trade press (rewritten press release). Lanes: Predictive. (AIJM-S06_PULL-IBVAL_v0.1.md S8.)
60. Mark Samuels, Diginomica, "How Bloomberg uses machine learning to create a competitive advantage." Published 27 December 2018. https://diginomica.com/how-bloomberg-uses-machine-learning-to-create-a-competitive-advantage. Register: trade press. Lanes: Agentic (background — Shawn Edwards table-reading quote; not confirmed as load-bearing in the live v0.11 lane text, but present in the case dossier). (AIJM-S04_CASE-THREADS-BC_v0.1.md, line 295.)
61. Ted Merz, ted-merz.com, "Bloomberg launches its own ChatGPT." 21 January 2026. https://ted-merz.com/2026/01/21/bloomberg-launches-its-own-chatgpt/. Register: trade press / independent observer — Merz is a former Bloomberg global head of news product who explicitly disclaims Bloomberg contact for this piece. Lanes: Agentic, Conclusion (background for the "325,000 financial professionals" and beta-timing figures). (AIJM-S04_CASE-THREADS-BC_v0.1.md.)
62. MarketsMedia, verbatim reprint of the ASKB launch release. https://www.marketsmedia.com/bloomberg-introduces-agentic-ai-to-the-terminal/, 23 February
    2026. Register: first-party content republished verbatim by trade press. Lanes: Agentic. Listed here by venue; content grouped with Bloomberg above.
63. TheTradeNews, rewrite (not a reprint) with accurate direct quotes, corroborating the ASKB launch date. https://www.thetradenews.com/bloomberg-embeds-agentic-ai-into-the-terminal/, 23 February 2026, 5:00 PM GMT. Register: trade press. Lanes: Agentic.
64. "Bloomberg Unveils ASKB Roadmap for Clients to Augment their Investment Process with Agentic AI," Radical Data Science (WordPress), reprint of the 16 April 2026 ASKB roadmap release. https://radicaldatascience.wordpress.com/2026/04/16/bloomberg-unveils-askb-roadmap-for-clients-to-augment-their-investment-process-with-agentic-ai/, 16 April 2026 (matching the original). Register: trade press (verbatim reprint). Lanes: Agentic. Since the Bloomberg original (entry 30) is now recovered live, this reprint is supplementary, not load-bearing.
65. Joel Khalili, "The Bloomberg Terminal Is Getting an AI Makeover, Like It or Not," Wired. 28 April 2026. https://www.wired.com/story/the-bloomberg-terminal-is-getting-an-ai-makeover-like-it-or-not/. Quotes Shawn Edwards ("It has become more and more untenable...", "not magical... 10 mediocre ideas") and states the "roughly a third of the software's 375,000 users" beta figure. Register: mainstream/trade press. Lanes: Agentic. Direct fetch of wired.com is blocked from this session (as it was for the original research); the URL, author and date are corroborated by independent search-index quotations of both required Edwards phrases and by a secondary summary citing Wired's Joel Khalili.
66. Talking Biz News, full reprint of the Micklethwait memo. 26 April 2016. https://talkingbiznews.com/they-talk-biz-news/micklethwait-bloomberg-to-expand-story-automation-efforts/ (same URL as entry 7 above — this is the venue-indexed duplicate listing). Register: trade press reprinting a first-party document in full — the record's designated copy of record. Lanes: Generative (C2), Agentic. Full detail: generative bibliography C2.

Mainstream press

67. Jaclyn Peiser, "The Rise of the Robot Reporter." The New York Times, 5 February 2019. https://www.nytimes.com/2019/02/05/business/media/artificial-intelligence-journalism-robots.html (reproduced same-day at https://talkingbiznews.com/they-talk-biz-news/the-rise-of-the-robot-business-reporter/; corroborated at https://www.techmeme.com/190206/p3). Register: mainstream press. Lanes: Generative (C6). Full detail: generative bibliography C6. Recorded UNSUPPORTED AS CITED for the "roughly a third" automation-share figure — it is Peiser's own sentence with no Bloomberg attribution.
68. Katie Robertson, "Bloomberg Has a Rocky Start With A.I. Summaries." The New York Times, 29 March 2025. https://www.nytimes.com/2025/03/29/business/media/bloomberg-ai-summaries.html. Register: mainstream press, carrying a first-party Bloomberg statement. Paywalled — every quotation from it in this project is relay-verified, not read from the original, via N3 and via Nieman Lab of 31 March 2025. Lanes: Generative (N2). Full detail: generative bibliography N2.

Practitioner / independent expert commentary

69. Ehud Reiter, personal blog, 10 August 2020 post. https://ehudreiter.com/2020/08/10/ (protocol added to the record's own bare-domain locator, AIJM-S19 v0.4). Register: practitioner expert blog. Lanes: Generative (F2). Full detail: generative bibliography F2.
70. Ehud Reiter, personal blog, 29 June 2023 post. https://ehudreiter.com/2023/06/29/ (protocol added to the record's own bare-domain locator, AIJM-S19 v0.4). Register: practitioner expert blog. Lanes: Generative (F3). Full detail: generative bibliography F3.
71. Ehud Reiter, "In 2019 LM output was fluent but not trustworthy: still true in 2024." Ehud Reiter's Blog, 27 February 2024. https://ehudreiter.com/2024/02/27/lm-output-still-fluent-not-trustworthy/. Register: practitioner expert blog. Lanes: Conclusion (the "But ultimately I wonder if the fact that my ultra-high-level description of LLMs has not changed in 5 years..." block quote). (AIJM-S11_RESEARCH-DOSSIER_practitioner-recency_v0.1.md, item confirming the 27 Feb 2024 date and URL.)
72. Ehud Reiter, "Real-world usage of LLMs in Journalism." Ehud Reiter's Blog, 23 April 2024. https://ehudreiter.com/2024/04/23/usage-llms-journalism/. Register: practitioner expert blog. Lanes: none currently in the live paper text — surfaced in the recency dossier as background; listed for completeness since it was fetched in full and is directly on-topic.
73. Ehud Reiter, "Defining hallucination is not straightforward." Ehud Reiter's Blog, 10 September 2025. https://ehudreiter.com/2025/09/10/defining-hallucination-is-not-straightforward/. Register: practitioner expert blog. Lanes: none currently quoted in the live text; background only.
74. Ehud Reiter, "Comparing performance of LLMs is not very interesting." Ehud Reiter's Blog, 24 March 2026. https://ehudreiter.com/2026/03/24/comparing-performance-of-llms-is-not-very-interesting/. Register: practitioner expert blog. Lanes: none currently quoted; background only.
75. Ehud Reiter, "AI for Healthcare: Great benchmark scores but minimal impact." Ehud Reiter's Blog, 5 August 2026. https://ehudreiter.com/2026/08/05/aimed-great-benchmarks-minimal-impact/. Register: practitioner expert blog. Lanes: Field-history (the "benchmarks and test-sets... can serve as screening tools" quote, confirmed verbatim here — GLM/Perplexity verification, 2026-09-12; this connection had never been made before).
76. Ehud Reiter, "Future of NLG evaluation." Ehud Reiter's Blog, 26 June 2026. https://ehudreiter.com/2026/06/26/future-of-nlg-evaluation-2/. Register: practitioner expert blog. Lanes: Conclusion (the "I hope that we will see good progress towards these goals by 2030..." block quote). (AIJM-S11_RESEARCH-DOSSIER_future-direction-quotes_v0.1.md, item 1a.)
77. Frank Harrell, "Road Map for Choosing Between Statistical Modeling and Machine Learning." https://www.fharrell.com/post/stat-ml/ (protocol added to the record's own bare-domain locator, AIJM-S19 v0.4), published 30 April 2018, modified 24 December 2023. Register: practitioner expert blog. Lanes: Generative (F4), Field-history (the "quick feedback while training" and classification quotations). Full detail: generative bibliography F4.
78. Frank Harrell, talk abstract. Statistical Thinking (fharrell.com), talk given at the Regression Modeling Strategies Course 2026. https://www.fharrell.com/talk/ai/; slide deck at https://hbiostat.org/talks/ai.html. Register: practitioner expert blog / talk materials. Lanes: Conclusion (the "For statistical programming, success has come when I play the role of specification writer..." block quote, dated 19 May 2026 in the paper's own attribution). (AIJM-S11_RESEARCH-DOSSIER_future-direction-quotes_v0.1.md.)
79. Andrej Karpathy, "Software 2.0." Medium, published 11 November 2017. https://karpathy.medium.com/software-2-0-a64152b37c35. Register: practitioner essay. Lanes: Generative (F6), Field-history ("Karpathy named the shift 'Software 2.0' in 2017..."), Conclusion (referenced by name but not actually quoted in the live v0.18 Conclusion text; a quotable block was cut by Hugh at v0.18 per that file's own Provenance). Full detail: generative bibliography F6.
80. Andrej Karpathy, bearblog post, "Sequoia Ascent 2026." https://karpathy.bearblog.dev/sequoia-ascent-2026/. Register: practitioner expert blog. Not cited anywhere in the live paper text — this is the 2026 recency source the S11 dossier located for Karpathy, but the Conclusion's "Karpathy... has too [said something about where the field is headed]" sentence (line 81) currently points to no quote at all, because the quoted block was cut in the v0.18 edit pass. Listed here as a finding: the paper names Karpathy as a fourth voice in "Where practitioners say their own field is headed" but carries no quotation for him, while a usable, project-verified 2026 source exists and is unused. See SUGGESTED TICKETS. (AIJM-S11_RESEARCH-DOSSIER_future-direction-quotes_v0.1.md, item on Karpathy.)
81. Gina Chua, quoted in the paper's practitioner bio (undated composite). Register: practitioner expert blog / column, multiple pieces — see the three dated entries below (82–84), which now supply all three quotations the paper uses. The bio's own closing quotation ("The form of words is increasingly less valuable, because machines create the form of words.") was resolved at v0.3 to entry 84 (Nieman Reports, 12 January 2026) — see that entry's own note.
82. Gina Chua, "View / How AI will upend the news." Semafor, 1 September 2025, 7:49pm ET. https://www.semafor.com/article/09/01/2025/view-how-ai-will-upend-the-news. Register: practitioner expert commentary (signed opinion column). Lanes: Conclusion (the "Generative AI promises to revolutionize how people interact with information..." block quote). (AIJM-S11_RESEARCH-DOSSIER_practitioner-recency_v0.1.md.)
83. Gina Chua, "Change Agent." (Re)Structured News, her own Substack, 8 June 2026. https://restructurednews.substack.com/p/change-agent. Register: practitioner expert commentary. Lanes: Conclusion (the "That the vast majority of people will turn to agentic systems..." block quote). (AIJM-S11_RESEARCH-DOSSIER_future-direction-quotes_v0.1.md, item 4a.)
84. Gina Chua, interviewed, "'How can we have the best of machines and the best of humans?'" Nieman Reports (Live @ Lippmann, interviewer Megan Cattel), 12 January 2026. https://niemanreports.org/gina-chua-ai-news-industry-disruption/. Register: practitioner expert commentary (edited Q&A, her words, first-person). Lanes: Field-history — this is the direct source of the closing bio quote ("The form of words is increasingly less valuable, because machines create the form of words"), confirmed against AIJM-S11_RESEARCH-DOSSIER_future-direction-quotes_v0.1.md line 214, which carries the full sentence including a clause the paper's quotation trims ("...which, by the way, AI-generated text is not copyrightable"). This entry was already correctly sourced in the project's own record; the connection to this specific quotation had simply never been made. (AIJM-S11_RESEARCH-DOSSIER_future-direction-quotes_v0.1.md, item 4b.)

Aggregator-relayed

85. RETIRED. Formerly an Apple App Store user review (storefront CA, user "Iarya55," 2 February 2022) cited to support a sentence in the Generative section about The Bulletin's undetermined persistence. That sentence was removed from the paper (AIJM-S20) because no stable public URL exists for an individual App Store review; this entry is kept and marked retired, rather than deleted, so the numbering of every other entry stays stable.
86. Slashdot, "Bloomberg's AI-Generated News Summaries Had At Least 36 Errors Since January." 30 March 2025. https://news.slashdot.org/story/25/03/30/1946224/bloombergs-ai-generated-news-summaries-had-at-least-36-errors-since-january. Author not named. Register: aggregator-relayed. Lanes: Generative (N3). Full detail: generative bibliography N3.
87. "Bloomberg's use of AI summaries for its articles leads to numerous corrections," Washington Times. 29 March 2025 (exact date match). https://www.washingtontimes.com/news/2025/mar/29/bloombergs-use-ai-summaries-articles-leads-numerous-corrections/. Byline reported as Brad Matthews via the paper's own story archive rather than confirmed from the rendered article header (both this session's and Perplexity's direct fetch were blocked) — carried as attributed, not certain. The count inside the piece ("at least 20 instances") is the article's own tally, not Bloomberg's or the Times's. Register: aggregator-relayed. Lanes: Generative (N4). Full detail: generative bibliography N4.
88. Oliver Darcy, "Bloomberg and the Bots," Status. 2 April 2025 (confirmed this session directly — resolves the 1-vs-2-April discrepancy the recovery pass initially flagged). https://www.status.news/p/bloomberg-news-ai-summaries. Body behind a hard paywall; the visible opening confirms Bloomberg's "full control" claim and Bloomberg staff's skepticism of it. Register: trade press reaching this study at a remove — kept under Generative's own "trade press" tag there; cross-listed here because its relay function matches this category's evidentiary posture. Lanes: Generative (N5). Full detail: generative bibliography N5.
89. Sarah Scire, "Let's get to the point: Three newsrooms on generating AI summaries for news," Nieman Lab. 5 June 2025. https://www.niemanlab.org/2025/06/lets-get-to-the-point-three-newsrooms-on-generating-ai-summaries-for-news/. Register: trade press, carrying four distinct facts the Generative chapter treats as load-bearing: Chris Collins (Bloomberg's chief product officer of news) saying "Reader feedback — both in comments to Bloomberg and what the newsroom has seen in the data — has been very positive" (quote confirmed verbatim this session, resolving the paper's paraphrase), the coverage rule, the plan to expand to opinion pieces, and the Wall Street Journal's CMS-workflow answer (all four independently confirmed on the live page this session). Lanes: Generative (N6). Full detail: generative bibliography N6.
90. "Investors harness Bloomberg's expanded AI tools to discover and summarize news," InPublishing. 2 December 2025 (sitemap records 10:09 UTC; some search indexes normalize this to 1 December). https://www.inpublishing.co.uk/articles/investors-harness-bloombergs-expanded-ai-tools-to-discover-and-summarize-news-26052. House byline (no individual credit) — a near-verbatim relay of Bloomberg's own product-announcement language (entry 27). Register: aggregator-relayed. Lanes: Generative (N8). Full detail: generative bibliography N8.
91. "The Bloomberg Terminal Is Getting an AI Makeover," Slashdot mirror of the Wired ASKB piece (28 April 2026). https://news.slashdot.org/story/26/04/28/1832202/the-bloomberg-terminal-is-getting-an-ai-makeover. Preserves the "not magical... 10 mediocre ideas" passage. See entry 65 above for the now-also- recovered Wired original. Register: aggregator-relayed. Lanes: Agentic.

Excluded — not a project source

92. aicerts.ai, "Bloomberg AI Summaries Reshape Financial Tech." 25 November 2025. Register: not a project source — flagged "purge on sight" by the project's own record; carries an AI-generated content disclaimer and asserts, unsourced, a claim Wayne Barlow directly contradicts. Lanes: Generative (G3) — listed there only because the chapter names and rebuts it. Full detail: generative bibliography G3.

Count by completeness tier

• Fully cited (organization/title/date/author-if-any + a working URL or permanent identifier such as a DOI or arXiv ID): 88 of the 92 numbered entries above.
• Thin (organization and date established, but title, author and/or a working URL still missing): 2 of 92 (entries 33 and 81) — both are index/pointer entries that name no source of their own rather than genuine citation gaps.
• No usable citation at all: 0, unchanged since v0.3.
• Excluded (not a project source): 1 (entry 92).
• Retired (citation removed; no longer supports any sentence in the paper): 1 (entry 85).

Of the 91 entries that still function as citations in the paper, 88 are fully sourced and 2 are thin index/pointer entries — and, as of this version, zero are documented permanent gaps: the paper's one previously unfixable citation (entry 85, an Apple App Store review with no stable public URL) was retired along with the sentence it supported, rather than carried indefinitely as a documented gap.

Full recovery method, per-entry verification detail, and version history: AIJM-PAPER.json (paper-wide provenance record) and the internal working file AIJM-S20_BIBLIOGRAPHY-FULL_v0.7.md.