References and source policy

DEVELOPMENT REVIEW DEPLOYMENT - NOT READY FOR RELEASE

Source policy

This module teaches people to check claims. It would be indefensible if its own claims were unchecked. Three rules apply to everything written here.

1. Historical and mathematical claims carry a citation or an uncertainty label. Every factual claim in the module maps to an entry below or is plainly labelled contested or unsourced. A claim with neither is a defect that blocks release. Author-only drafting flags must be resolved before material reaches this reference list.

2. Claims about current AI carry a date. Model capabilities, tool names, pricing, institutional policy, benchmark results, and industry practice all age badly. Any such claim is written either so it does not date (a statement about how next-token prediction works) or with an explicit date attached (a statement about what a particular system did in a particular month). Undated claims about "what AI can do now" do not belong in the materials.

3. Named commercial systems are avoided unless pedagogically necessary. The module teaches a method that outlives any particular product. Naming a model version in the teaching text creates a maintenance burden and implies an endorsement. Where a learner needs an AI system for an activity, the material says "an AI assistant of your choice".

Verification status labels

Used throughout course/historical-notes.md and the unit materials:

label meaning
established Supported by a primary source, or by consistent secondary sources with no serious dispute.
contested Sources disagree, or the claim is widely repeated but its evidential basis is unclear. State the disagreement; do not pick a side.
unsourced The source is not currently known. This does not mean false; state what evidence could settle it.

An author-only drafting flag is not a learner-facing uncertainty category and must not survive to release. Unsourced is the module's student-facing bucket, defined in Unit 16, for a claim deliberately left open together with the evidence that could move it.

Pre-release freshness review

Before any student-facing release, a reviewer must re-read every dated claim and either refresh it or remove it. This is a named task in TODO.md, Stage 6.

Primary sources

[P1] Menabrea, L. F. (1842). "Notions sur la machine analytique de M. Charles Babbage." Bibliothèque Universelle de Genève, new series, vol. 41, no. 82, October 1842. The original French account of Babbage's Analytical Engine, based on Babbage's 1840 lectures in Turin. This is the article Lovelace translated.

[P2] Menabrea, L. F., trans. A. A. L. [Augusta Ada Lovelace] (1843). "Sketch of the Analytical Engine Invented by Charles Babbage... with notes by the translator." In R. Taylor (ed.), Scientific Memoirs, vol. 3, London: Richard and John E. Taylor, pp. 666–731. The translation plus Lovelace's Notes A–G. The Notes are commonly characterised as roughly three times the length of the article — a secondary characterisation, consistent with the page counts here but not something this page-run itself states. Note G contains the Bernoulli-number diagram and the "no pretensions whatever to originate anything" passage. Public domain. The diagram itself is headed "for the computation by the Engine of the Numbers of Bernoulli. See Note G (page 722 et seq.)", so the Bernoulli material sits at page 722 and following; the surrounding Notes span the pp. 666–731 range commonly cited for the whole translation. Confirmed against the facsimile [P2a].

[P3] Babbage, C. (1864). Passages from the Life of a Philosopher. London: Longman, Green, Longman, Roberts, & Green. Babbage's own retrospective account, including his description of the Analytical Engine and of Lovelace's work on the Notes. A primary source that is also an interested party — useful, not neutral.

Facsimiles

The Note G table must be checked against a facsimile rather than a transcription, because the error the module discusses is precisely the kind of thing a transcription silently corrects.

[P2a] Facsimile of the Note G diagram. Photograph of the printed "Diagram for the computation by the Engine of the Numbers of Bernoulli" from Scientific Memoirs vol. 3, as offered by Sophia Rare Books: https://www.sophiararebooks.com/pictures/3544a.jpg (accessed 2026-07-23; 1000×654 JPEG, retained at media/unit-16-historical-bug/facsimile-noteg.jpg so the check does not depend on the URL persisting). Legible from it: the diagram heading and page reference (722 et seq.); that the error is in operation 4; and that operation 4 acts on ²V₅ and ²V₄ in that order. Not resolvable at this resolution: the fine typography of the division operator, which is easy to misread. For anything turning on the exact glyph, a higher-resolution facsimile or the physical volume is needed. See course/historical-notes.md §4.

Higher-resolution scanned copies of Scientific Memoirs vol. 3 are also available through several digital library collections and should be recorded here if used.

Secondary sources

[S1] Bromley, A. G. (1982). "Charles Babbage's Analytical Engine, 1838." Annals of the History of Computing, 4(3), 196–217. Technical reconstruction of the Engine's architecture. The standard reference for what the machine could actually do.

[S2] Stein, D. (1985). Ada: A Life and a Legacy. Cambridge, MA: MIT Press. A sceptical assessment of Lovelace's mathematical contribution. Represents one pole of a genuine and continuing scholarly disagreement.

[S3] Toole, B. A. (1992). Ada, the Enchantress of Numbers: A Selection from the Letters of Lord Byron's Daughter and Her Description of the First Computer. Mill Valley, CA: Strawberry Press. A collection of correspondence, taking a considerably more expansive view of Lovelace's contribution than [S2]. Read alongside it, not instead of it.

[S4] Swade, D. (2000). The Difference Engine: Charles Babbage and the Quest to Build the First Computer. New York: Viking. On Babbage's machines and the modern reconstruction effort.

[S5] Hollings, C., Martin, U., and Rice, A. (2018). Ada Lovelace: The Making of a Computer Scientist. Oxford: Bodleian Library. A recent treatment drawing on the Lovelace archives, notable for engaging directly with the disputed questions rather than taking a side by omission. Useful for Unit 16's distinction between established and contested claims.

[S6] The printed errors, and what each reading computes. Operation 4 is printed as V₅ ÷ V₄ where V₄ ÷ V₅ is correct, commonly read as a typesetting rather than an authorial error; its location and operand inversion are established by the facsimile [P2a] and the reconstructions [S7], [S8]. It is not the only printed fault: [S7] §5.2 also lists operation 24, where "the negative value of V₁₃ must be transferred to V₂₄", and defects in operation 21. A phrase like "the table as printed" therefore does not name one object.

The module reconstructed the 25-operation table and ran it with exact arithmetic (course/lab/noteg.py), which reproduces the documented −1/30 when both faults are repaired and then reports each other reading: 139/630 (operation 4 as printed, 24 repaired, earlier values supplied), −139/630 (both faults literal), 1/30 (only 24 as printed), and −25621/630 when the faulty routine generates the earlier Bernoulli numbers it consumes.

The last of those is the value repeated in encyclopaedia and blog sources (https://en.wikipedia.org/wiki/Note_G, accessed 2026-07-23). Correction of record: this module previously stated that 139/630 was what the as-printed table computes and that −25621/630 "does not survive recomputation". Both were wrong for the same reason — neither said which object was executed — and the 139/630 run silently repaired operation 24. What those sources can still be faulted for is not stating their interpretation; so could this module, until 2026-07-26.

[S7] Glaschick, R. (2016). Ada Lovelace's Calculation of Bernoulli's Numbers. Paderborn, 29 September 2016. https://rclab.de/_media/analyticalengine/aal_noteg_glaschick_v1.2.pdf (accessed 2026-07-23). A detailed independent reconstruction, self-published — a step-by-step working of the Note G table that states the operation-4 operand swap directly ("V₄ must be divided by V₅") and groups it with typographical-domain errors; independently confirms the target value B₇ = −1/30.

[S8] Campbell-Kelly, M. (ed.) (1989). The Works of Charles Babbage, vol. 3. London: Pickering. The archival correction of the Note G table (p. 159, note a); cited by [S7] and by Misa (2015) as the fullest correction of record.

A note on tertiary sources

Popular articles, encyclopaedia entries, and blog posts about Note G's table error are abundant, mutually inconsistent, and frequently trace back to each other rather than to the table. They are useful for finding out what is commonly claimed — which Unit 16 asks learners to examine — and are not sufficient support for a factual assertion in the materials. Where the module reports a common claim, it says that it is a common claim.

Mathematical references

[M1] Bernoulli numbers, conventions. Two standard conventions differ in the value of B1: the first Bernoulli numbers (B_n^-) take B1 = -1/2, and the second (B_n^+) take B1 = +1/2. All other values agree. This module fixes the first convention; see course/lab/reference_bernoulli.py.

[M2] Published value tables. The oracle's PUBLISHED_TABLE was checked against NIST Digital Library of Mathematical Functions, Release 1.2.7 of 2026-06-15, F. W. J. Olver et al. (eds.), Chapter 24, §24.2(iv), Tables 24.2.1 and 24.2.3, National Institute of Standards and Technology, https://dlmf.nist.gov/24.2.T1 and https://dlmf.nist.gov/24.2.T3 (accessed 2026-07-24). The subsection identifies its printed source as M. Abramowitz and I. A. Stegun (eds.) (1964), Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, National Bureau of Standards Applied Mathematics Series 55, Washington, DC: U.S. Government Printing Office, pp. 809–810; the electronic source corresponds to the corrected tenth printing of December 1972.

The comparison covers B0–B12, every even value through B30 = 8615841276005/14322, and the odd zeros represented in the oracle. The table and DLMF equation 24.2.2 use the module's convention B1 = -1/2. All 22 stored values agree exactly; no correction was needed.

[M3] Defining recurrence. The oracle uses sum_{j=0}^{n} C(n+1, j) * B_j = 0 for n >= 1, with B0 = 1, which yields B1 = -1/2 directly. course/lab/examples/correct_example.py deliberately uses a different method (the Akiyama–Tanigawa transform) so that agreement between them is evidence rather than duplication.

Language models and current AI systems

These references support the compact account in Unit 3. They describe distinct layers of a modern assistant: transformer architecture, pre-training and post-training, retrieval, and calibrated self-evaluation. Unit 3 does not imply that every product uses every technique.

[L1] Vaswani, A. et al. (2017). "Attention Is All You Need." Advances in Neural Information Processing Systems, 30. arXiv 1706.03762. https://arxiv.org/abs/1706.03762 (accessed 2026-07-24). Introduced the Transformer architecture. Its attention mechanism combines information from different token positions to produce context-dependent representations. This citation supports the architecture sketch in Unit 3, not a claim that attention provides a truth check or a human-readable explanation.

[L2] Ouyang, L. et al. (2022). "Training Language Models to Follow Instructions with Human Feedback." Advances in Neural Information Processing Systems, 35, 27730–27744. arXiv 2203.02155. https://arxiv.org/abs/2203.02155 (accessed 2026-07-24). Shows one influential post-training pipeline: supervised demonstrations followed by preference data and reinforcement learning from human feedback. It supports the distinction between next-token pre-training and training an assistant to follow instructions.

[L3] Lewis, P. et al. (2020). "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks." Advances in Neural Information Processing Systems, 33, 9459–9474. arXiv 2005.11401. https://arxiv.org/abs/2005.11401 (accessed 2026-07-24). Combines a pre-trained generator with an external document index. It supports Unit 3's distinction between generation from model parameters and a system that also retrieves inspectable sources.

[L4] Kadavath, S. et al. (2022). "Language Models (Mostly) Know What They Know." arXiv 2207.05221. https://arxiv.org/abs/2207.05221 (accessed 2026-07-24). Finds that specially elicited or trained self-evaluation scores can discriminate correct from incorrect answers on some evaluated tasks, with calibration and out-of-distribution limits. This is why Unit 17 says ordinary confident wording is not a correctness defence while avoiding the stronger, false claim that every model-derived confidence measure carries no information.

Second-order reasoning

Background for course/second-order.md and Unit 18. Read that page first: it marks the boundary between the established results below and the module's own synthesis, and the module's claims discipline applies to its own capstone exactly as it applies to everything else.

[T2] Tarski, A. (1936). "Der Wahrheitsbegriff in den formalisierten Sprachen." Studia Philosophica, 1, 261–405. The undefinability of truth: under the theorem's formal conditions, arithmetical truth for the object language is not definable within that same language. This is a scoped result about formal languages. Applying it to a language model, a person, or an organisation requires a separate argument and is not a corollary of Tarski's theorem.

Publication history, checked. Volume and page range are established: the German text is Studia Philosophica vol. 1, pp. 261–405, Leopold Blaustein's translation of the Polish original with a postscript added (Stanford Encyclopedia of Philosophy, "Alfred Tarski", bibliography, https://plato.stanford.edu/entries/tarski/, accessed 2026-07-23; that bibliography lists the German text as 1935, which is the datum the contested-year discussion below turns on). The year is contested, and the disagreement is recorded here rather than settled. The volume is dated 1935 and the Stanford Encyclopedia of Philosophy cites it as 1935; Peter Milne, "On the Year of Publication of Tarski's 'Der Wahrheitsbegriff in den formalisierten Sprachen'", History and Philosophy of Logic, 46(2), 273–286 (DOI 10.1080/01445340.2024.2334173, published online 26 April 2024), argues from correspondence that the journal issue did not appear until 1936, and that what circulated in late 1935 were preprints lacking two corrections and a short addendum. The module cites 1936; a reader who finds 1935 elsewhere has not found an error.

The other versions, and why a page number depends on which one. The Polish original is Pojęcie prawdy w językach nauk dedukcyjnych, Warsaw: Nakładem Towarzystwa Naukowego Warszawskiego, 1933, VII + 116 pp., in the series Prace Towarzystwa Naukowego Warszawskiego, Wydział III: Nauk Matematyczno-Fizycznych, no. 34 (catalogue record, Kujawsko-Pomorska Biblioteka Cyfrowa, permalink https://kpbc.umk.pl/dlibra/publication/265521/edition/262893, accessed 2026-07-23, which confirms the series title, the Warsaw imprint and the VII + 116 pagination; the "Wydział III" number 34 as cited by the Stanford Encyclopedia of Philosophy). The English translation, "The Concept of Truth in Formalized Languages", is in Logic, Semantics, Metamathematics, 2nd edn 1983 [1956], pp. 152–278. Three texts, three paginations: a page reference to "Tarski" means nothing unless it says which one.

[T3] Gödel, K. (1931). "Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I." Monatshefte für Mathematik und Physik, 38, 173–198. The first incompleteness theorem. Closely related to [T2]; the two are frequently conflated in secondary accounts, and Unit 18 should keep them distinct.

[T7] von Foerster, H. (2003). Understanding Understanding: Essays on Cybernetics and Cognition. New York: Springer-Verlag. Collected essays from the tradition that includes the observer in the system being described. Confirmed against the book's own front matter: "© 2003 Springer-Verlag New York, Inc.", ISBN 0-387-95392-2, xii + 362 pp., with the printing line "9 8 7 6 5 4 3 2 1" — the first printing, and there is no later edition. The Library of Congress CIP data on the same page carries the year 2002 (Q315.5 .V64 2002), which is why some catalogues give 2002; the imprint year is 2003. Chapter and page references used by the module are to this printing. The 2003 Springer edition is also online with per-chapter DOIs under the stem 10.1007/0-387-21722-3 (e.g. chapter 14, "Ethics and Second-Order Cybernetics", pp. 287–304, DOI 10.1007/0-387-21722-3_14; the chapter DOI, its title, page range and 2003 year confirmed via the Crossref record, accessed 2026-07-23).

[T8] Maturana, H. R. and Varela, F. J. (1980). Autopoiesis and Cognition: The Realization of the Living. Boston Studies in the Philosophy of Science, vol. 42. Dordrecht, Holland: D. Reidel. The companion reference for the same tradition. Confirmed against the book's own copyright page: published by D. Reidel Publishing Company, P.O. Box 17, Dordrecht, Holland; series Boston Studies in the Philosophy of Science, vol. 42; copyright 1980; ISBN 90-277-1015-5 (cloth) and 90-277-1016-3 (paperback). The same 1980 volume is reprinted by Springer under the stable DOI 10.1007/978-94-009-8947-4, whose Crossref record (accessed 2026-07-23) confirms the title, the 1980 date, ISBN 978-90-277-1016-1 and the Boston Studies series — the Crossref record uses the later series name "Boston Studies in the Philosophy and History of Science", the same renaming noted below. No edition statement is printed — this is the first English book publication. Sold and distributed in the USA and Canada by Kluwer Boston Inc., Hingham, MA, which is why several catalogues record the place of publication as Boston; the imprint is Dordrecht. The series was later renamed Boston Studies in the Philosophy and History of Science, and records that use the newer name for volume 42 are describing the same book. The second of the two essays, "Autopoiesis: The Organization of the Living" (from p. 63), first appeared in Spanish as De Máquinas y Seres Vivos, Editorial Universitaria S.A., 1972.

[T9] Madaan, A. et al. (2023). "Self-Refine: Iterative Refinement with Self-Feedback." Advances in Neural Information Processing Systems, 36. Experiments on seven tasks found that iterative feedback and revision using the same language model could improve outputs over one-step generation. This is dated, task-bounded evidence that same-model revision can be useful; it does not make the revision independent verification. Primary paper: https://proceedings.neurips.cc/paper_files/paper/2023/hash/91edff07232fb1b55a505a9e9f6c0ff3-Abstract-Conference.html (accessed 2026-07-24).

A note on how these are used. [T2] and [T3] are established results with settled citations; they are stated as established. [T7] and [T8] are named as a research tradition rather than as results. [T9] is dated empirical evidence, not a general guarantee about self-correction. Unit 18's checkpoint asks the learner to sort these categories themselves, so the categories must be visible in the material rather than smoothed into a single voice.

Second-order cybernetics

Background for course/second-order.md and Unit 18, which cite these as [C1] to [C7]. They are the sources for the cybernetic tradition itself, as distinct from the formal results at [T2] and [T3]. All of it is established, published material, and none of it is a claim about language models.

[C1] von Foerster, H. — second-order cybernetics, the cybernetics of observing systems. The distinction between the cybernetics of observed systems, where the observer stands outside, and the cybernetics of observing systems, where the observer is included in what is described, is formulated in his work. The usual English source is the collected papers Understanding Understanding: Essays on Cybernetics and Cognition, New York: Springer, 2003 — the same volume listed above as [T7]. Cite [C1] for the first-order/second-order distinction and [T7] for the volume; they are one source under two uses.

The formulation, located. It is stated in von Foerster's own words in "Ethics and Second-Order Cybernetics", chapter 14 of [T7] (DOI 10.1007/0-387-21722-3_14, Crossref record accessed 2026-07-23), pp. 287–304 — an interview with Yveline Rey, first published in French in Systèmes, Éthique, Perspectives en thérapie familiale, ed. Y. Rey and B. Prieur, Paris: ESF éditeur, 1991, pp. 41–55, per the chapter's own source note. On p. 303: "I would say, first-order cybernetics is the cybernetics of observed systems, while second-order cybernetics is the cybernetics of observing systems." The first half of the pair is given separately on p. 299: "the whole conceptual machinery of 'early' cybernetics, first-order cybernetics; or as I would say, the cybernetics of observed systems." Cite one of those two pages; do not attribute the formulation to the volume at large. (Chapter 13 of the same volume, pp. 283–286, is von Foerster's own "Cybernetics of Cybernetics", which is a different text from Mead's at [C2].)

The Biological Computer Laboratory. Much of the work was done there. It was founded by von Foerster at the University of Illinois at Urbana-Champaign in 1958 and closed in 1976 — "founded by Professor Heinz von Foerster in 1958, made Illinois an international center of cybernetics research until the lab's demise in 1976" (Illinois Distributed Museum, University of Illinois, https://distributedmuseum.illinois.edu/exhibit/biological_computer_laboratory/, accessed 2026-07-23). The 1958 founding is confirmed independently by the University of Illinois Archives: "he established the Biological Computer Laboratory in 1958" (https://archives.library.illinois.edu/2016/12/06/heinz-von-foerster-and-the-bcl/, accessed 2026-07-23). Give 1958–1976 rather than "the late 1950s to the mid 1970s".

[C2] Mead, M. (1968). "Cybernetics of Cybernetics." In H. von Foerster, J. D. White, L. J. Peterson and J. K. Russell (eds), Purposive Systems: Proceedings of the First Annual Symposium of the American Society for Cybernetics, New York: Spartan Books, pp. 1–11. Commonly cited as the point at which the reflexive turn — cybernetics applied to itself — was named. The address is associated with the founding of the American Society for Cybernetics. The module states only that it is commonly cited as that point, which is what the available sources support. The venue, editors and year are given in this form by the American Society for Cybernetics itself (https://asc-cybernetics.org/CofC/wp-content/uploads/2010/08/mead.html, accessed 2026-07-23, which prints the citation "Mead, M (1968) Cybernetics of Cybernetics, in Foerster, H von, White, J, Peterson, L and Russell, J (eds) (1968) Purposive Systems, New York, Spartan Books"); the page range is given by P. Pangaro, "Why do we want to live in cybernetics?", Cybernetics and Human Knowing, 23(2), 2016, 9–21, whose reference list reads "Mead, M. (1968). Cybernetics of cybernetics. In von Foerster, H., White, J. D., Peterson, L. J., & Russell, J. K. (Eds.), Purposive Systems (pp. 1–11). New York: Spartan Books", and who quotes Mead at pp. 9 and 10 — in-text page citations that fall inside the stated range. Contested: the occasion of the address. The society's first annual symposium is usually dated 26–27 October 1967 at the National Bureau of Standards, Gaithersburg, Maryland, but descriptions of the volume also place the symposium itself in 1968. The module's claim rests only on the printed text, whose date, 1968, is not in dispute. Do not assert a delivery date in the materials.

[C3] Maturana, H. R. and Varela, F. J. (1980). Autopoiesis and Cognition: The Realization of the Living. The same volume as [T8]: cite [C3] for the observer, [T8] for autopoiesis. Publisher, place, series and edition are established and are recorded at [T8] (Boston Studies in the Philosophy of Science, vol. 42; Dordrecht: D. Reidel, 1980).

The formulation, located — and corrected. It belongs to Maturana's essay "Biology of Cognition", first issued as Biological Computer Laboratory Research Report BCL 9.0, Urbana, IL: University of Illinois, 1970, and reprinted in this volume at pp. 5–58. In section III, "Cognitive Function in General", under the sub-heading "The Observer", numbered statement (1), on p. 8, the sentence reads:

Anything said is said by an observer. In his discourse the observer speaks to another observer, who could be himself; whatever applies to the one applies to the other as well.

Checked against a page facsimile of the 1980 Reidel text, and against the 1970 BCL report, which has the same wording.

So the module's own working phrase was slightly wrong. The commonly quoted form everything said is said by an observer is not the wording on p. 8. It is the title of a later Maturana essay — "Everything said is said by an observer", in W. I. Thompson (ed.), Gaia: A Way of Knowing: Political Implications of the New Biology, Lindisfarne Press, 1987. Both forms are Maturana's; they are not the same text. Quote "Anything said is said by an observer" and cite p. 8 of this volume, or quote "Everything said is said by an observer" and cite the 1987 essay by title. Do not attach the 1987 wording to the 1980 page. Unit 16 asks learners to watch a claim drift as it is repeated; this is the same thing happening inside this file, and it is worth saying so in the teaching.

[C4] The Macy Conferences on Cybernetics (1946–1953). A series of ten meetings funded by the Josiah Macy Jr. Foundation and chaired by Warren McCulloch, deliberately mixing mathematicians, engineers, neurophysiologists, anthropologists and psychiatrists. The standard secondary history in English is Heims, S. J. (1991), The Cybernetics Group, Cambridge, MA: MIT Press. The surviving transcripts have been edited by Claus Pias, which is the route to the primary material.

The Pias edition — there are two, and they are not the same book. The bilingual original is Claus Pias (ed.), Cybernetics — Kybernetik: The Macy-Conferences 1946–1953, Zürich and Berlin: Diaphanes, in two volumes: vol. 1, Transactions / Protokolle, 2003, 734 pp., ISBN 978-3-935300-35-3; vol. 2, Essays and Documents / Essays und Dokumente, 2004, 512 pp., ISBN 978-3-935300-36-0 (records of the Deutsche Nationalbibliothek). The later English-only edition is Claus Pias (ed.), Cybernetics — The Macy Conferences 1946–1953: The Complete Transactions, Zurich and Berlin: Diaphanes, 2016, first printing, 734 pp., ISBN 978-3-03734-598-6 (Deutsche Nationalbibliothek). It is distributed in North America by the University of Chicago Press, whose catalogue page is https://press.uchicago.edu/ucp/books/book/distributed/C/bo23348570.html (accessed 2026-07-23); note that that page lists its printing as ISBN 978-3-03734-607-5, 752 pp., differing from the Diaphanes printing above, so cite the exact printing consulted. Cite whichever was actually consulted: "the Pias edition" on its own is ambiguous, and the two have different contents and different pagination.

The count of ten. Ten is the core series and does not include the supplementary meetings. The ten conferences on "Circular Causal and Feedback Mechanisms in Biological and Social Systems", later "Cybernetics", ran from 8–9 March 1946 to 22–24 April 1953 (American Society for Cybernetics, Macy conference summary, https://www.asc-cybernetics.org/foundations/history/MacySummary.htm, accessed 2026-07-23, which lists all ten with dates — first conference 8–9 March 1946, tenth 22–24 April 1953). The separately sponsored related meetings are not among them: a 1946 sociological sub-conference organised by Bateson, and the New York Academy of Sciences symposium on teleological mechanisms, whose papers appeared as Annals of the New York Academy of Sciences, 50(4), 1948, 187–278. The Pias volumes use the same count of ten. "A series of ten meetings" as written above is therefore correct, read as the core series.

[C5] Bateson, G. (1972). Steps to an Ecology of Mind: Collected Essays in Anthropology, Psychiatry, Evolution, and Epistemology. San Francisco: Chandler Publishing Company. Bateson was a Macy participant [C4], and this collection is the standard source for his part in the tradition. This entry is the reference for the Bateson attribution in course/second-order.md and Unit 18, which previously carried no source at all.

The 1972 imprints, and why the pagination does not carry. The first edition is Chandler Publishing Company, San Francisco, 1972, 545 pp., ISBN 0-8102-0447-9, LCCN 75-169581. A Ballantine Books paperback, New York, 1972, 541 pp., appeared the same year by arrangement with Chandler. Records that merge the two — giving "Ballantine Books; Chandler Pub. Co., New York, 1972" in one line — are conflating imprints, and the two differ in pagination. The current reissue is University of Chicago Press, 2000, 565 pp., ISBN 978-0-226-03905-3, with a new foreword by Mary Catherine Bateson (publisher's page https://press.uchicago.edu/ucp/books/book/chicago/S/bo3620295.html, accessed 2026-07-23, which confirms the 2000 year, the 565 pp., the ISBN and the foreword). The 1972 Chandler first edition carries LCCN 75-169581 and ISBN 0-8102-0447-9; that LCCN and OCLC 258014 are recorded in OpenLibrary edition OL17854018M (https://openlibrary.org/books/OL17854018M, accessed 2026-07-23), though that OpenLibrary record itself merges the Chandler and Ballantine imprints — the exact conflation warned against just above — so it is cited for the identifiers, not the pagination. Cite the edition actually used and give its pagination; page numbers do not transfer between these three.

[C6] Gödel (1931), first incompleteness theorem. Listed in full above as [T3] and not repeated here. course/second-order.md cites it as [C6] and Unit 18 cites it as [C6]; [T3] is the same paper under its earlier label.

[C7] Tarski (1936), undefinability of truth. Listed in full above as [T2] and not repeated here, including the record of that volume's contested date and of the three separately paginated versions of the text. course/second-order.md cites it as [C7], and Unit 18 now cites [C7] directly; [T2] is the same paper under its earlier label. Anything said about the paper's year or pagination belongs at [T2], where the disagreement is set out, and not restated here.

A note on how these are used. [C1] to [C5] support one claim and no more: that second-order cybernetics is a named research tradition with a literature and a history going back to the 1940s. They are not evidence about what any AI system does, and course/second-order.md says so on its own account.

Gothic and Romantic context

Supporting the visual-register decision recorded for the module and the short passage in Unit 2. These entries exist so that the register carries a citation rather than an atmosphere.

[G1] Marchand, L. A. (1957). Byron: A Biography. 3 vols. New York: Alfred A. Knopf. The standard scholarly biography, and the reference for Byron's departure from England in 1816 and his subsequent life abroad. Ada Lovelace was born in December 1815; Byron left in 1816 and did not see her again.

Confirmed from the book itself (volume III front matter, contents and index): Alfred A. Knopf, New York, 1957, "a Borzoi book", first edition, LC catalog card number 57-7547, published simultaneously in Canada by McClelland & Stewart. The 1957 Knopf edition is catalogued at OpenLibrary as edition OL42986375M (https://openlibrary.org/books/OL42986375M, accessed 2026-07-23), under the work record for Marchand's Byron: A Biography (LC card 57-7547 is from the volume itself, above). Three volumes, continuously paginated — vol. 1, chapters I–XII, pp. 3–476; vol. 2, chapters XIII–XXIII, pp. 477–942; vol. 3, chapters XXIV–XXX, pp. 943–1264, followed by notes, a list of sources and the index. Because the pagination is continuous, a page reference identifies the volume by itself.

The 1816 departure falls in volume 2, at the end of chapter XV, "1816 The Separation" (pp. 563–608), running into chapter XVI, "1816 Switzerland" (from p. 609). The index is explicit: "Dover, 607–8", and under Byron, "preparations for going abroad, 602; engaged Dr. Polidori, 602–3; ... signed deed of separation, 605; ... to Dover in Napoleonic coach, 607; ... at Churchill's grave, 607–8; ... through lane of spectators to boat, 608; ... end of an epoch, 609; ... at Ostend, 610". Cite pp. 605–610 for the departure as a whole, or pp. 607–8 for Dover.

[G2] The Villa Diodati, summer 1816. The gathering on Lake Geneva at which Byron, Percy Bysshe Shelley, Mary Godwin (later Mary Shelley), Claire Clairmont and John Polidori spent part of the summer, and from which the ghost-story challenge that produced [G3] and [G4] is conventionally dated. The nearest thing to a primary account is Mary Shelley's own introduction to the 1831 edition of Frankenstein; Polidori's diary is the other contemporary source.

A citable edition of the diary. Polidori, J. W., The Diary of Dr. John William Polidori, 1816, Relating to Byron, Shelley, etc., edited and elucidated by William Michael Rossetti, London: Elkin Mathews, 1911. This is the standard text, with the standing caveat that Rossetti worked from the copy expurgated by Polidori's sister Charlotte, the original having been destroyed. Reissued in the Cambridge Library Collection (Literary Studies), Cambridge University Press, 2014, 240 pp., ISBN 978-1-108-07228-1, DOI 10.1017/CBO9781107444713. The 1911 text is in the public domain and is available complete as Project Gutenberg ebook 55017 (https://www.gutenberg.org/ebooks/55017, accessed 2026-07-23; confirmed as Rossetti's edition of Polidori's 1816 diary), which is the copy the dates below were checked against.

Dates that are established, from the diary and from the Shelley-Godwin Archive's Frankenstein chronology (which follows Nora Crook's "Chronology of Life and Works"):

Contested: the date of the ghost-story challenge. The conventional date is 16 June 1816. The Shelley-Godwin Archive chronology (https://shelleygodwinarchive.org/contents/frankenstein/frankenstein-chronology/, accessed 2026-07-23) declines to fix it, recording three possibilities — before 15 June, on 15 June, or on 16 June — and noting only that 16 June is "the date often given". The diary fixes an outer bound rather than a date: on 17 June, "The ghost-stories are begun by all but me"; on 18 June, "Began my ghost-story after tea", the same evening as Byron's recitation of Christabel and Shelley's fit. Rossetti's note on the 17 June entry points out that this contradicts the standard sequence, in which the Christabel incident preceded the challenge — "The ghost-stories ... had already been begun ... not later than June 17, whereas the Christabel incident happened on June 18." Give a range, not a night, and do not write "on the night of 16 June" as settled fact.

[G3] Shelley, M. (1818). Frankenstein; or, The Modern Prometheus. In Three Volumes. London: Printed for Lackington, Hughes, Harding, Mavor, & Jones, 1818. Published anonymously. The revised 1831 edition carries the author's introduction describing the Villa Diodati summer, and differs from the 1818 text in ways that matter if the novel is quoted rather than merely cited.

Imprint, confirmed. Three volumes; London; "Printed for Lackington, Hughes, Harding, Mavor, & Jones"; 1818. The imprint is given in this form by the New York Public Library's record for its Berg Collection copies and by the catalogue record for the Internet Archive facsimile of volume I (https://archive.org/details/shelleyfrankenstein01, accessed 2026-07-23, whose record gives the bracketed printer "Printed [by Macdonald and Son] for Lackington..."), and the publisher and year are given the same way by the Frankenstein Variorum, a scholarly digital edition (https://frankensteinvariorum.org/about, accessed 2026-07-23) collated against a photographic facsimile of the 1818 text. The NYPL record adds that the print run was about 500 copies, of which it holds seven. The Internet Archive record supplies the printer in brackets — "Printed [by Macdonald and Son] for Lackington..." — i.e. as a cataloguer's addition rather than from the title page, so do not print it as if it were on the book. The publication date of 1 January 1818 is standardly given but was not confirmed from a primary source here; if the module needs the day, check it before printing it.

[G4] Polidori, J. W. (1819). "The Vampyre; A Tale." The New Monthly Magazine and Universal Register, vol. XI, no. 63 (1 April 1819), pp. 195–206. London: Henry Colburn. First published in April 1819 and initially, and wrongly, attributed to Byron — which is itself a usable example of an attribution that was widely repeated before it was checked.

Checked against the scanned issue, page by page. The scanned issue used is the scan-backed Wikisource edition of The New Monthly Magazine and Universal Register, Volume XI, No. 63 (https://en.wikisource.org/wiki/The_New_Monthly_Magazine_and_Universal_Register/Volume_11/Number_63/The_Vampyre, accessed 2026-07-23), transcribed from the Wikimedia Commons scan of Volume XI (file The New Monthly Magazine and Universal Register - Volume 011.djvu). The running head on the first page of the tale reads "1819.] The Vampyre; a Tale by Lord Byron. 195"; the tale ends on p. 206, where "A Pedestrian Tour round Florence" begins under the running head "[April 1,", which also fixes the issue date. The volume and number are confirmed by the signature line on p. 193: "New Monthly Mag.—No. 63. Vol. XI". The tale is preceded, from p. 193, by the unsigned "Extract of a Letter from Geneva, with Anecdotes of Lord Byron", which supplies the Villa Diodati frame; booksellers' catalogues that cite the piece as "pp. 193–194" are citing that introduction rather than the tale, so pp. 195–206 is the range for the tale itself. The false attribution is on the title as printed, not merely in circulation: "a Tale by Lord Byron". A disclaimer by Polidori is reported in the following month's issue; that follow-up was not verified here.

The module makes no claim that this literary inheritance influenced Lovelace's own work. That the family connection exists is established; what she made of it is not a claim this module supports.

Music quoted

Unit 18 carries short song quotations as part of its register. They are recorded here so that each one has a source and a rights position rather than appearing as unattributed text on a slide.

The three entries below mark the three steps of that unit's argument, in the order the unit takes them: [Q1] the position that everything is a lie, [Q3] the realisation that no standpoint outside the system is available, [Q2] what is left once that is accepted. Each work is quoted to at most two short lines, and the argument at every step is stated in the module's own words beside the quotation.

Rights position. The position is to reference the song and quote a minimal set of lyrics, rather than to clear rights or to drop the material. The lyrics remain in copyright; the module reproduces no work in whole or in substantial part.

This module is produced in the UK, so the framework is fair dealing under the Copyright, Designs and Patents Act 1988, section 30 — which permits fair dealing for criticism or review and, since 2014, for quotation. It is not US fair use, which is a different test, and this module should not be read as relying on one.

The conditions that framework attaches, and how each is met:

A quotation that fails any of them does not go in, and the argument never depends on a lyric, so removing one would cost the module nothing. By the module's own rule the register may carry an argument and may never substitute for one.

The licence does not extend to them. The AGPL covers this module's original work only:

Third-party material — in particular the song lyrics quoted in Unit 18 — remains the copyright of its respective owners, is included strictly for criticism, review and educational illustration, and is excluded from the AGPL grant. Reusing this module does not give you any right to reuse that quoted material.

Where this stands. On 2026-07-27 the director judged the three quotations below to meet the conditions above, these being short quotations from lyrics. He said so in those terms, including the qualification "as far as I can tell", and that is what is recorded: the judgement of the person responsible for the module, not settled legal advice. On 2026-07-28, having considered a reviewer's argument that an open worldwide release carries a more complicated rights context than a closed educational presentation, he confirmed the decision and paired it with the carve-out above. Employer ownership and the copyright status of generated output remain open for a dedicated copyright review. A reader following the chain from a quotation to its basis should be able to see exactly how far the basis goes, which is why the qualification is quoted rather than smoothed away.

[Q1] Lacrimosa. "Alles Lüge." Title track of the single Alles Lüge, Hall of Sermon, Switzerland, 1993. Referenced, and quoted minimally under the educational quotation exception. Not from the album, despite the usual attribution. The band's own discography lists Alles Lüge (1993) as a single, separately from the studio album Satura (1993), and the original six-track Satura — "Satura", "Erinnerung", "Crucifixio", "Versuchung", "Das Schweigen", "Flamme im Wind" — does not contain it (band's own discography, https://lacrimosa.com/en/project/alles-luege-1993/, accessed 2026-07-23; MusicBrainz release group for the 1993 single, type Single, https://musicbrainz.org/release-group/1e938aaf-09e1-37d6-b004-2af4e6213d91, and for the album Satura, https://musicbrainz.org/release-group/d320ced3-5fa5-3725-94ca-8be674182bd0, both accessed 2026-07-23; the single-versus-album distinction and the six-track original Satura are corroborated by both, so the 1993 single date is not single-sourced). The song was added to later Satura reissues as a bonus track, which is where the common "from Satura" attribution comes from. Cite the 1993 single. MusicBrainz also dates the recording's first appearance to a label sampler, German Mystic Sound Sampler, Volume III (1992); that is single-sourced, is not relied on, and should not be printed without a second source.

[Q2] Einstürzende Neubauten. "Sabrina." Opening track of the album Silence Is Sexy, Mute (UK) / Potomak (Germany), 2000. Referenced, and quoted minimally under the educational quotation exception. Release date 3 April 2000 on the UK Mute release (MusicBrainz release https://musicbrainz.org/release/e0ceeda4-fa04-30f0-b511-75117b673c7d, accessed 2026-07-23, giving date 2000-04-03, country GB, label Mute, catalogue CD STUMM 182, opening track "Sabrina"). This exact release date rests on that single database and is recorded here as single-sourced, not independently corroborated. Two lines are quoted in Unit 18: "It's not the red of which we bleed" and "It is as black as Malevich's Square". Both were checked against two independent lyric transcriptions (https://readdork.com/lyrics/einsturzende-neubauten-sabrina and https://www.letras.com/einstuerzende-neubauten/675817/, both accessed 2026-07-25). The transcriptions agree on the wording and differ on the spelling of the painter's name (Malevich and Malevitch); the module prints Malevich, the usual English form, and the picture referred to is Kazimir Malevich's Black Square. As with [Q3], lyric sites are not authoritative editions and the wording is a review item for the sleeve text.

[Q3] Einstürzende Neubauten. "Haus der Lüge." Title track of the album Haus der Lüge, Some Bizzare (UK, Bart 333) / Rough Trade (Germany, RTD 126), 1989. Referenced, and quoted minimally under the educational quotation exception: Unit 18 quotes the refrain couplet "Gott hat sich erschossen / ein Dachgeschoss wird ausgebaut" and glosses it in English, as the step where the standpoint above is gone and the building work carries on regardless. On the German release the title track is track 4, listed as "Haus der Lüge / Epilog" (MusicBrainz release group https://musicbrainz.org/release-group/f30172ff-a6de-341f-9871-8614c4ccec66 and release https://musicbrainz.org/release/bf441775-8718-4e4d-bae7-7616cbe088c8, giving date 1989-09-04, country DE, labels Some Bizzare and Rough Trade with those catalogue numbers, both accessed 2026-07-25). The 1989 year is corroborated across the fifteen releases in that release group; the exact 4 September date rests on the single release entry and is recorded here as single-sourced. The couplet's wording was checked against two independent lyric transcriptions (https://readdork.com/lyrics/einsturzende-neubauten-haus-der-luge and https://www.songtexte.com/songtext/einsturzende-neubauten/haus-der-luge-2bd2a476.html, both accessed 2026-07-25), which agree; both record it as a repeated refrain rather than a single line. Lyric sites are not authoritative editions, so the wording is recorded as transcription-sourced and is a review item for the sleeve text.

Citing in module materials

In-text, cite by bracket label: (Lovelace 1843, Note G) for [P2], [S1] for a technical point. Every citation must resolve to an entry on this page. If it does not, it is a defect.