Research and Source Record — The Rope and the Algorithm

I wrote The Rope and the Algorithm as a historical and analytical inquiry into procedure, evidence, and trust. The book begins with the Śulba Sūtras, texts associated with the construction of Vedic fire altars, and moves toward a modern question: what must people be able to inspect, test, and correct before they are justified in relying on a complex technical system?

The connection between the two parts is methodological. I am not arguing that the Śulba Sūtras were early computer programs, that Vedic practitioners anticipated Artificial Intelligence, or that a historical resemblance proves a line of technological descent. I am asking what becomes visible when two very different settings are placed beside one another: a material procedure involving cords, marks, surfaces, trained action, and standards; and a contemporary technical system involving data, models, evaluation, deployment, institutions, and governance.

This record explains how I researched that comparison, what kinds of sources I used, what each group of sources contributes, and where the evidence requires restraint. It is intended to make the research trail available to readers without turning the book into a catalogue of notes.

I began by separating the questions the book needed to answer. The historical chapters ask what the Śulba traditions are, how they relate to the wider Vedic ritual setting, what their texts and translations preserve, and how their dates and mathematical procedures should be described. The later chapters ask how modern technical systems are built, evaluated, explained, governed, and held accountable. The comparison is made only after those two bodies of material have been treated in their own terms.

For each chapter, I identified the claims that required evidence. I then looked for the strongest source appropriate to the claim: a translated primary text or critical edition for textual material; a university-press book, peer-reviewed article, or recognised scholarly chapter for historical and mathematical interpretation; an institutional archive for collection-level facts; a government or intergovernmental framework for policy language; and an original research paper or working paper for empirical findings.

I checked more than the reputation of a source's author. I considered the author's standing in the relevant field, the publication venue, the kind of claim the source could support, the availability of an inspectable record, and whether an independent source was needed. A catalogue record can establish that a work exists, but it cannot substitute for the work itself. An abstract can establish the broad result reported by a study, but it cannot support every detail of the study's method. An OCR record can help locate a passage, but a number or quotation that matters should be checked against the page image or a reliable edition.

I also kept different kinds of statement apart. A primary text or archive may provide evidence. A scholar may interpret that evidence. I may then draw an analytical comparison. Those are three different activities. A modern reconstruction can demonstrate that an expression is mathematically equivalent to another expression; it cannot, by itself, establish what an ancient author intended, what notation was used, or what theory a practitioner possessed.

The Śulba Sūtras are treated here as texts and practices within a religious and ritual tradition. I have not treated them merely as a storehouse of mathematical curiosities, and I have not used religious material as decorative proof for a modern technological argument. I distinguish text from performance, translation from interpretation, later reconstruction from ancient intention, and archival description from my own conclusion.

The comparison with Artificial Intelligence is deliberately limited. It concerns procedure, conditions, reproducibility, error, standards, authority, correction, and responsibility. It does not claim that ancient practitioners encoded a hidden theory of machine learning, predicted modern computation, or stood in a demonstrated genealogical line to contemporary AI systems.

The same discipline applies to claims about priority. A result can be technically impressive without being evidence that a culture made a modern discovery in modern terms. A documented ritual performance can be historically important without proving unchanged continuity from antiquity. A high model score can be useful without proving that a system is trustworthy in every setting.

The Preface establishes the method of the book. My principal historical source is Sen and Bag's translated edition of the Śulba texts, supported by Dani's scholarly mathematical and historical review. Henderson's discussion provides a modern reconstruction of the square-root material, while Staal's work provides a documentary and ethnographic context for the later discussion of Agnicayana.

These sources support a general account of the Śulba corpus, its constructional concerns, the importance of distinguishing related textual witnesses, and the need to separate evidence from analogy. They do not support a claim that the texts contain modern computer science or that a modern mathematical reconstruction reveals the intentions of the original authors.

The Preface therefore makes a modest argument. Before comparison can be useful, the objects being compared must be described accurately. The historical material must remain historical, the modern technical material must remain modern, and the resemblance between them must be treated as an analytical question rather than as proof of descent.

Chapter 1 places the Śulba traditions within the wider Vedic ritual setting. Sen and Bag provide the main translated edition and commentary for the principal Śulba witnesses: Baudhāyana, Āpastamba, Kātyāyana, and Mānava. Dani supplies a broader scholarly account of the mathematics and of the difficulty of reading later mathematical categories back into earlier material. These sources support the claim that the Śulba texts belong to the wider ritual world without making them identical to the four Vedas themselves.

The chapter also uses these sources to describe the material setting in which constructional procedures matter: cords, marks, pegs, ground, lengths, areas, right-angle relations, and trained action. I present the resulting account as an analytical summary of a material practice, not as a claim that the ancient texts use my modern vocabulary of "specification," "inspection," or "verification."

The chapter does not attempt to settle the detailed square-root reconstruction, the Mānava 25/8 material, the dating debate, the full history of Agnicayana performance, or the later AI frameworks. Those questions belong elsewhere in the book because combining them too early would make the historical opening carry claims it cannot itself establish.

Chapter 2 returns from historical setting to textual and mathematical content. Sen and Bag remain the principal translated source. Dani and Pingree provide scholarly context for the place of the Śulba traditions within Indian mathematical literature. Kashikar's edition of the Baudhāyana Śrautasūtra helps situate the wider Śrauta environment without being used as a substitute for a Śulba-specific source. Plofker supports the description of Indian mathematics as practical and constructive, while Kichenassamy contributes a modern scholarly reconstruction of Baudhāyana's rule for the quadrature of the circle.

A central concern of this chapter is textual plurality. Baudhāyana, Āpastamba, Kātyāyana, and Mānava are related traditions, not one uniform manual. Differences among them matter. They prevent the reader from treating "the Śulba Sūtras" as if the phrase referred to a single, unchanging document with one voice and one purpose.

The Mānava material requires particular care. The value 25/8 has been discussed through modern reconstructions, including work associated with Gupta and Dani. I use those reconstructions as scholarly interpretations of a textual problem, not as an unqualified statement that the ancient text simply declares π to equal 25/8. The MacTutor History of Mathematics Archive is useful as a university-hosted guide and lead, but a guide or catalogue is not enough by itself to establish a precise verse-level claim.

The chapter's broader conclusion is therefore narrower than a priority claim. The texts preserve difficult and useful constructional procedures. Later scholars can reconstruct mathematical relationships in them. Neither fact requires us to attribute modern mathematical theory or modern intentions to the original practitioners.

Chapter 3 addresses chronology and the temptation to make dates do more work than the evidence permits. The chapter draws on the translated Śulba corpus, Pingree's history of Indian mathematical literature, Dani's review, and the wider historiographical discussions associated with the dating of Vedic and mathematical texts.

I distinguish several dates that are often collapsed: the date of a practice, the date of composition, the date of a manuscript, and the date of a printed or translated edition. A text may preserve an older practice without having been written down at the moment the practice began. A later manuscript may transmit an earlier composition. A modern edition may make a text accessible without changing the date of the underlying tradition.

The chapter also treats origins and transmission cautiously. Similar procedures in different cultures can raise an important comparative question, but similarity alone does not demonstrate contact, borrowing, or shared descent. The appropriate historical conclusion depends on evidence for transmission, not only on the fact that two traditions solved related constructional problems.

This is why I keep the language of possibility separate from the language of proof. A proposed connection may be worth investigating. It is not thereby established. Honest uncertainty is not a refusal to learn; it is a way of preventing a persuasive story from outrunning its sources.

Chapter 4 asks whether a modern term such as "algorithm" can be used carefully in relation to an ancient constructional procedure. Cormen, Leiserson, Rivest, and Stein provide the modern definition and computational context. Sen and Bag, Dani, Pingree, and Kashikar provide the historical and textual material from which the comparison begins.

I use "algorithm" as a modern analytical category, not as a claim about ancient terminology or consciousness. A procedure can have an ordered sequence, inputs, operations, checks, and an intended result without being a computer algorithm in the modern sense. The comparison becomes useful only when the difference is kept visible.

The chapter's constructional list is likewise a modern analytical reconstruction. It asks whether a procedure can be described through a required form, materials, operations, comparison, and correction. It does not claim that an ancient practitioner wrote that list or possessed a formal theory of algorithms.

The chapter therefore draws a line between resemblance and identity. A modern word may illuminate a feature of a historical procedure if it is introduced as an analytical tool. It becomes misleading when it replaces the historical object with a modern one.

Chapter 5 examines approximation, iteration, and correction. Henderson is the principal source for the modern reconstruction of the square-root material. Bailey and Borwein provide numerical and historiographical caution. Fowler and Robson offer a comparative reference through the Babylonian tablet YBC 7289. The National Academies' report on reproducibility and replicability supplies a modern vocabulary for distinguishing the repeatability of a result from the repeatability of a larger research process.

The sequence 17/12 → 577/408 → 665857/470832 is treated as a modern analytical reconstruction and a successive divide-and-average or Newton–Heron sequence. I do not attribute Newton's formalism, modern algebra, or modern convergence theory to the Śulba authors.

The derivation of 17/12 is also a modern calculation. Starting with a rectangle whose sides are 1 and 2, the arithmetic mean of those sides is 3/2. The area-preserving complement is 4/3, and the mean of 3/2 and 4/3 is 17/12. These operations show how a modern analyst can examine a relation. They do not tell us what tolerance an ancient practitioner used, how the expression was understood, or what theoretical language was available at the time.

The chapter's error taxonomy distinguishes measurement error, construction error, transcription error, interpretive error, and model error. The point is not that these errors are identical. The point is that "error" becomes useful only when a target, standard, tolerance, purpose, and possible corrective action are specified.

Chapter 6 studies ritual performance, transmission, and documentation through the 1975 Nambudiri Agnicayana performance at Panjal in Kerala. The Smithsonian Human Studies Film Archives finding aid provides a collection-level record of the film and associated materials. Staal's Agni provides scholarly and ethnographic documentation, and Schechner's review addresses the performance and its mediation. Kashikar supplies wider Śrauta context. UNESCO's record on Vedic chanting and Gerety's study of traditional knowledge transmission in Kerala are used comparatively, not as direct substitutes for evidence about every aspect of the Panjal event.

The archival record describes a twelve-day performance documented through filming and related materials. It makes it possible to study material preparation, construction, recitation, ritual action, interviews, scholarly framing, and the conditions under which an event becomes an archive. The record does not justify calling the performance an untouched survival or proof of unbroken continuity from antiquity.

The chapter separates three records: the performance itself, the archive that preserves and frames it, and the later interpretation through which readers encounter it. A camera preserves some things and selects others. A catalogue describes a collection from a particular institutional and historical position. A scholarly account adds interpretation. None of these layers can simply be removed to reveal an unmediated past.

I also use Gerety's work on transmission in Kerala to examine how knowledge may involve voice, gesture, face-to-face teaching, social regulation, and living authority. That comparative case does not prove that every ritual tradition works in the same way. It helps clarify why a recording cannot replace the competence and relationships through which a practice is learned.

Chapter 7 moves into contemporary technical systems. Vaswani and colleagues provide the Transformer architecture as a concrete example of how a model architecture can make certain forms of computation practical. Pineau and colleagues provide research on reproducibility in machine learning. Mitchell and colleagues contribute model cards; Gebru and colleagues contribute datasheets for datasets; Amershi and colleagues contribute research on human–AI interaction. NIST's Artificial Intelligence Risk Management Framework provides a broader organisational and lifecycle perspective.

The central distinction is between a model and a system. A model is one component of an arrangement that also includes problem formulation, data, objectives, software, hardware, evaluation, deployment, interfaces, users, and institutions. A model's architecture specifies possibilities and constraints, but training and deployment conditions shape what happens in practice.

I describe this as the difference between a model's designed form and its operational form. The designed form includes architecture, objectives, and intended use. The operational form emerges under actual data, computation, users, interfaces, incentives, and institutional conditions. The difference between them is not a minor technical detail; it is where many questions of evaluation and responsibility arise.

Documentation can preserve part of the surrounding knowledge. Model cards may describe intended uses, limitations, and evaluation conditions. Datasheets may record aspects of dataset provenance and characteristics. Reproducibility work emphasises code, data, experimental details, and reporting practices. None of these documents guarantees safety. They make some conditions more visible.

Chapter 8 asks what it means to understand a system whose internal operations are difficult to inspect. Doshi-Velez and Kim frame the need for a rigorous science of interpretability. Ribeiro, Singh, and Guestrin provide the LIME approach to local explanations. Sundararajan, Taly, and Yan provide integrated gradients and its axiomatic attribution framework. Olah contributes a mechanistic-interpretability perspective. Mitchell and colleagues provide model cards, while NIST provides a system-level risk framework.

The chapter distinguishes several questions that are often collapsed into the single word "explanation." An explanation may be behavioural, attributive, mechanistic, human-grounded, or system-level. It may help a person decide what to do without faithfully describing the mechanism that produced an output. It may identify an important feature without providing a complete account of the model's internal computation.

I organise the evaluation of explanations around usefulness, faithfulness, completeness, and stability. These dimensions do different work. A useful explanation helps a particular reader or user. A faithful explanation tracks the relevant model behaviour. A complete explanation covers the parts of the mechanism needed for the question at hand. A stable explanation remains materially similar under appropriate changes in input, method, or context.

The same gap that separates a model's designed form from its operational form reappears here between an explanation's claimed fidelity and the computation it is meant to describe. A clear story may be useful without being faithful. A faithful local account may be incomplete. A mechanistic hypothesis may be technically strong without answering the institutional question of who should be allowed to rely on the system.

An explanation should therefore have a standard, a test, and a correction pathway. Plausibility is not enough.

Chapter 9 examines trustworthiness through accuracy, error distribution, operating conditions, affected people, and correction. The principal case is the introduction of more stringent biometric identification requirements in India's public distribution system, studied by Muralidharan, Niehaus, and Sukhtankar. I checked the peer-reviewed article against the authors' NBER working-paper record. I also used Drèze, Khalid, Khera, and Somanchi's work on Aadhaar and food security in Jharkhand, and the India Policy Forum record on biometric authentication in welfare programmes.

I selected the case because it makes a narrow but important distinction visible. A system can reduce certain forms of corruption or leakage and still impose costs on legitimate beneficiaries. Administrative improvement and trustworthiness are not identical questions.

The research record keeps the study's phases separate. The first phase involved electronic point-of-sale machines and Aadhaar-based biometric authentication for ration collection and was evaluated experimentally. The second phase used authenticated transaction records in an event-study framework to examine changes in grain disbursal and leakage. These phases should not be collapsed into one undifferentiated "Aadhaar effect."

The study's reported design included randomisation across 132 sub-districts in 10 districts, an evaluation population of approximately 15.1 million beneficiaries in 17 of Jharkhand's 24 districts, and a survey sample of 3,840 Public Distribution System beneficiaries. The baseline consisted of 3,410 successful interviews, reported as 86 percent of the eligible sampled households. Follow-up counts were reported separately rather than treated as the same denominator.

Several figures require their qualifiers. By the first follow-up, 97 percent of beneficiary households in treated areas had at least one member with an Aadhaar number seeded to the Public Distribution System account, and 90 percent reported that transactions at their fair-price shop were authenticated. These are implementation and reported-experience measures, not proof that every transaction succeeded or that every entitlement was received.

The reported Phase 1 transaction-cost increase was 7 rupees on a base of 41 rupees, or approximately 17 percent. For unseeded households, the study reports a 49-rupee fall in the mean value of rice and wheat received. The percentage attached to that result depends on the denominator and the location in the study record; the research record preserves the relevant subgroup and does not silently combine summary figures from different versions.

The published summary reports that 1.5–2 million legitimate beneficiaries lost access at some point during the reforms. I attribute that figure to the authors and do not present it as a new national calculation or as proof of permanent exclusion for every person in the reported population.

The chapter's conclusion is deliberately bounded. The evidence does not establish that biometric authentication is inherently harmful, that every Aadhaar deployment has the same effect, or that every failed authentication was caused by a fingerprint mismatch. It shows instead that consequences depend on purpose, transition protocol, operating conditions, affected users, and the definition of success.

Chapter 10 draws on NIST's Artificial Intelligence Risk Management Framework, the OECD Recommendation on Artificial Intelligence, and research on human–AI interaction by Amershi and colleagues. It asks what follows once a technical system has been understood as an arrangement rather than as a model alone.

Governance is not only a final review gate. Decisions about purpose, data, metrics, deployment, access, monitoring, and remedy shape a system before a final output is produced. NIST's framework organises risk management through the functions GOVERN, MAP, MEASURE, and MANAGE. I use that structure as a way to discuss lifecycle responsibility, not as a guarantee that any system following a framework will be safe.

The OECD recommendation frames accountability in relation to the roles of AI actors and connects it with transparency, explainability, robustness, security, and safety. Human–AI interaction research adds the practical questions of feedback, reliance, control, and the conditions under which a person can meaningfully contest an output.

A system may be technically transparent and still institutionally unanswerable. Someone must be able to state what the system is for, what evidence counts, who is authorised to use it, who monitors performance, who hears a challenge, and who can suspend or withdraw it. Responsibility cannot disappear into the phrase "the model decided." It must remain connected to an actor with sufficient authority to act.

Chapter 11 is a synthesis rather than a new case study. It returns to the rope as a bounded methodological analogy and brings together the earlier distinctions among material procedure, error, documentation, explanation, deployment, governance, and correction.

Sen and Bag, Plofker, and Dani support the historical and mathematical side of the comparison. Mitchell, NIST, Doshi-Velez and Kim, the National Academies, the OECD, and Amershi support the modern discussion of models, explanations, reproducibility, governance, and human interaction.

The chapter's claims are deliberately modest. A rope can make a relation such as distance, direction, alignment, or proportion inspectable within a material practice. An AI output is not the whole system, and a benchmark is not the whole deployment. A reconstruction can demonstrate mathematical equivalence without proving ancient intention. An explanation can illuminate one behaviour without providing a complete account of mechanism.

The comparison is therefore methodological rather than genealogical. Both a material procedure and a modern technical system require conditions, standards, trained or organised action, evaluation, and a response to error. Their materials, purposes, histories, and risks remain different.

The conclusion is primarily authorial synthesis. It brings together the book's central distinction between resemblance and demonstrated connection, and between a useful result and a trustworthy procedure. Its historical claims remain dependent on the sources recorded above; its final argument is my own interpretation of what those materials make possible to say.

I end by acknowledging that the Śulba Sūtras belong to a living religious and cultural history. Treating them with care does not mean suspending criticism. It means using accurate terminology, distinguishing evidence from interpretation, acknowledging uncertainty, and refusing to make the past carry claims it cannot bear.

The book's final position is simple. A procedure becomes worthy of reliance when its purpose, conditions, operations, limitations, errors, standards, and routes of correction are sufficiently visible for people to judge what should follow. That principle does not make an ancient ritual construction and a modern AI system the same thing. It gives us a disciplined way to ask questions of both.

A reader wishing to retrace the research can begin with the references below and follow the same order I used. First, establish the primary or direct record for the claim. Second, consult scholarly interpretation in the relevant field. Third, record the exact locator and the limits of the source. Fourth, separate the source's reported result from any calculation or interpretation added by the researcher. Fifth, write the caveat next to the claim rather than adding it after the argument has already become too strong.

For the mathematical material, reproduce each operation from its stated inputs and distinguish the historical expression from the modern calculation. For the Agnicayana material, distinguish the performance from the film record, the archive catalogue, and later scholarship. For the Aadhaar case, keep the study phases, denominators, populations, outcomes, and identification strategies separate. For the AI material, distinguish the model from the system, the explanation from the mechanism, the benchmark from deployment, and transparency from accountability.

This method does not remove interpretation. It makes interpretation visible and therefore open to criticism.

This research record is provided for educational, historical, analytical, and general informational purposes. It is not legal, medical, financial, tax, investment, insurance, engineering, policy, or other professional advice, and it does not create a professional-client relationship. Historical interpretations, technical explanations, case-study summaries, and policy discussions are bounded by the sources cited here and may require further checking against primary materials, current law, current technical documentation, or local professional advice.

Readers should not make consequential decisions affecting rights, benefits, safety, privacy, health, livelihood, or compliance solely on the basis of this book or research record. For such decisions, readers should consult a suitably qualified professional or the relevant primary authority. External links, policies, databases, and technical systems may change after publication.

Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., Suh, J., Iqbal, S., Bennett, P. N., Inkpen, K., Teevan, J., Kikin-Gil, R., & Horvitz, E. (2019). Guidelines for human–AI interaction. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1–13. https://doi.org/10.1145/3290605.3300233

Bailey, D. H., & Borwein, J. M. (2012). Ancient Indian square roots: An exercise in forensic paleo-mathematics. The American Mathematical Monthly, 119(8), 646–657. https://doi.org/10.4169/amer.math.monthly.119.08.646

Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. (2009). Introduction to algorithms (3rd ed.). MIT Press. https://mitpress.mit.edu/9780262033848/introduction-to-algorithms/

Dani, S. G. (2023). Ancient Indian mathematics: Śulbasūtras—A mathematical review. In Handbook of the history and philosophy of mathematical practice. Springer. https://doi.org/10.1007/978-3-030-19071-2_114-1

Datta, B. (1993). Ancient Hindu geometry. Cosmo Publications. (Original work published 1932.)

Doshi-Velez, F., & Kim, B. (2017). Towards a rigorous science of interpretable machine learning (arXiv:1702.08608). arXiv. https://arxiv.org/abs/1702.08608

Drèze, J., Khalid, N., Khera, R., & Somanchi, A. (2017). Aadhaar and food security in Jharkhand: Pain without gain? Economic and Political Weekly, 52(50), 50–60. https://www.jstor.org/stable/45132599

Fowler, D., & Robson, E. (1998). Square root approximations in Old Babylonian mathematics: YBC 7289 in context. Historia Mathematica, 25(4), 366–378. https://doi.org/10.1006/hmat.1998.2209

Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J. W., Wallach, H., Daumé III, H., & Crawford, K. (2021). Datasheets for datasets. Communications of the ACM, 64(12), 86–92. https://doi.org/10.1145/3458723

Gerety, F. M. M. (2018). Digital guru: Embodiment, technology, and the transmission of traditional knowledge in Kerala. Asian Ethnology, 77(1/2), 3–32. https://www.jstor.org/stable/26604831

Gupta, R. C. (1988). New Indian values of π from the Mānava Śulba Sūtras. Centaurus, 31(2), 114–125. https://doi.org/10.1111/j.1600-0498.1988.tb00682.x

Henderson, D. W. (2000). Square roots in the Śulba Sūtra. In C. A. Gorini (Ed.), Geometry at work: Papers in applied geometry (MAA Notes 53, pp. 39–45). Mathematical Association of America. https://books.google.com/books?hl=en&lr=&id=Eb6uSLa2k6IC&oi=fnd&pg=PA39

Kashikar, C. G. (Ed. & Trans.). (2003). The Baudhāyana Śrautasūtra (4 vols.). Indira Gandhi National Centre for the Arts in association with Motilal Banarsidass. https://ignca.gov.in/the-baudhayana-srauta-sutra/

Kichenassamy, S. (2006). Baudhāyana's rule for the quadrature of the circle. Historia Mathematica, 33(2), 149–183. https://doi.org/10.1016/j.hm.2005.05.001

MacTutor History of Mathematics Archive. (n.d.-a). Indian Sulbasutras. University of St Andrews. Retrieved August 23, 2026, from https://mathshistory.st-andrews.ac.uk/Projects/Pearce/chapter-5/

MacTutor History of Mathematics Archive. (n.d.-b). Mānava. University of St Andrews. Retrieved August 23, 2026, from https://mathshistory.st-andrews.ac.uk/Biographies/Manava/

Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model cards for model reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220–229. https://doi.org/10.1145/3287560.3287596

Muralidharan, K., Niehaus, P., & Sukhtankar, S. (2020). Identity verification standards in welfare programs: Experimental evidence from India (NBER Working Paper No. 26744). National Bureau of Economic Research. https://doi.org/10.3386/w26744

Muralidharan, K., Niehaus, P., & Sukhtankar, S. (2025). Identity verification standards in welfare programs: Experimental evidence from India. The Review of Economics and Statistics, 107(2), 372–392. https://doi.org/10.1162/rest_a_01296

National Academies of Sciences, Engineering, and Medicine. (2019). Reproducibility and replicability in science. The National Academies Press. https://doi.org/10.17226/25303

National Institute of Standards and Technology. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). U.S. Department of Commerce. https://doi.org/10.6028/NIST.AI.100-1

Organisation for Economic Co-operation and Development. (2019, amended 2024). Recommendation of the Council on Artificial Intelligence (OECD/LEGAL/0449). https://legalinstruments.oecd.org/en/instruments/oecd-legal-0449

Olah, C. (2022). Mechanistic interpretability, variables, and the importance of interpretable bases. Transformer Circuits Thread. https://www.transformer-circuits.pub/2022/mech-interp-essay

Pingree, D. (1981). Jyotiḥśāstra: Astral and mathematical literature (A history of Indian literature, Vol. VI, Fasc. 4). Otto Harrassowitz.

Pineau, J., Vincent-Lamarre, P., Sinha, K., Larivière, V., Beygelzimer, A., d'Alché-Buc, F., Fox, E., & Larochelle, H. (2021). Improving reproducibility in machine learning research: A report from the NeurIPS 2019 reproducibility program. Journal of Machine Learning Research, 22(164), 1–20. https://jmlr.org/papers/v22/20-303.html

Plofker, K. (2009). Mathematics in India. Princeton University Press. https://press.princeton.edu/books/hardcover/9780691120676/mathematics-in-india

Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). "Why should I trust you?": Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1135–1144). Association for Computing Machinery. https://doi.org/10.1145/2939672.2939778

Schechner, R. (1986). Wrestling against time: The performance aspects of Agni. The Journal of Asian Studies, 45(2), 359–363. https://doi.org/10.2307/2055848

Selbst, A. D., Boyd, D., Friedler, S. A., Venkatasubramanian, S., & Vertesi, J. (2019). Fairness and abstraction in sociotechnical systems. Proceedings of the Conference on Fairness, Accountability, and Transparency, 59–68. https://doi.org/10.1145/3287560.3287598

Sen, S. N., & Bag, A. K. (1983). The Śulbasūtras of Baudhāyana, Āpastamba, Kātyāyana and Mānava, with text, English translation and commentary. Indian National Science Academy. https://bruknow.library.brown.edu/discovery/fulldisplay/alma991011837279706966/01BU_INST:BROWN

Staal, F. (Ed.). (1983). Agni: The Vedic ritual of the fire altar (2 vols.). Asian Humanities Press. https://search.worldcat.org/title/Agni-the-vedic-ritual-of-the-fire-altar/oclc/10018427

Sundararajan, M., Taly, A., & Yan, Q. (2017). Axiomatic attribution for deep networks. In Proceedings of the 34th International Conference on Machine Learning (Vol. 70, pp. 3319–3328). PMLR. http://proceedings.mlr.press/v70/sundararajan17a.html

UNESCO. (2008). Tradition of Vedic chanting. Intangible Cultural Heritage. https://ich.unesco.org/en/RL/tradition-of-vedic-chanting-00062

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30. https://papers.neurips.cc/paper/7181-attention-is-all-you-need

I have written this research record to make the distinction between evidence and interpretation visible. I remain responsible for the selection of sources, the wording of the claims, the calculations presented as modern calculations, the boundaries of the comparisons, and the final interpretation of the book's argument.

The Śulba traditions deserve to be read in their own historical and religious setting. Contemporary AI systems deserve scrutiny in their own technical and institutional setting. The value of the comparison lies not in collapsing those worlds into one another, but in asking—carefully and openly—what procedures require when people are expected to trust their results.