Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Sources — what each is, what the book takes from it, and where

The author session, 3 November 2026. One entry per work the book cites. Each gives the reference, its epistemic status as the book uses it (theorem; impossibility result; necessity theorem under conditions; design principle with a track record; empirical regularity; construction; record), the exact claim the book takes from it, and the chapters that rely on it. Entries are added as chapters are drafted; the charter's §13 is the seed. Where the book's use of a source is narrower than the source's own claim, the entry says so.

Decision theory and the owner's utility (for chapter 0)

  • Cox, R. T. (1946). Probability, frequency and reasonable expectation. American Journal of Physics 14(1), 1–13. — Theorem (under its regularity conditions). Degrees of belief that are consistent must be probabilities. Taken: "optimally" has one coherent meaning for beliefs. Ch. 0.
  • Savage, L. J. (1954). The Foundations of Statistics. Wiley. — Theorem (representation). Coherent preference under uncertainty is expected utility under a subjective probability. Taken: the utility half of "optimally". Ch. 0.
  • de Finetti, B. (1974). Theory of Probability, vol. 1. Wiley. — Theorem (coherence; Dutch book). Taken with Cox: conditioning as the unique update; previsions first, as credence had it. Ch. 14.
  • Hadfield-Menell, D., Dragan, A., Abbeel, P. & Russell, S. (2016). Cooperative inverse reinforcement learning. Advances in Neural Information Processing Systems 29. — Framework. The agent's utility is the human's, unknown, inferred from behaviour. Taken: the alignment commitment — the owner of the utility is the one unscored party. Ch. 0.

Computation and self-reference

  • Rice, H. G. (1953). Classes of recursively enumerable sets and their decision problems. Transactions of the American Mathematical Society 74(2), 358–366. — Theorem. Every non-trivial semantic (extensional) property of programs in a Turing-complete language is undecidable. Taken: merit cannot be settled by inspecting the change; statutes must be decidable properties. Not taken: anything about restricted languages, where properties may be decidable. Chs. 0, 2, 10, 11, 16, app. A.
  • Gödel, K. (1931). Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I. Monatshefte für Mathematik und Physik 38, 173–198. — Theorem (second incompleteness). A consistent recursively axiomatised theory interpreting enough arithmetic cannot prove its own consistency. Taken: the court cannot prove its own soundness; it keeps a book instead. Chs. 10, 12, app. A.
  • Löb, M. H. (1955). Solution of a problem of Leon Henkin. Journal of Symbolic Logic 20(2), 115–118. — Theorem. If T ⊢ Prov(⌜φ⌝) → φ then T ⊢ φ. Taken: a sound agent cannot license a successor whose evaluation rule it wrote; the evaluation rule is a fixed point. Chs. 0, 10, 12, 16, app. A.
  • Yudkowsky, E. & Herreshoff, M. (2013). Tiling agents for self-modifying AI, and the Löbian obstacle. Machine Intelligence Research Institute, technical report. — Design rule drawn from a theorem. Trust decreases strictly down a chain of proof-based successors. Taken as stated; the book does not claim jack escapes it, only that it obeys it by keys. Ch. 10, app. A.
  • Kalai, E. & Lehrer, E. (1993). Rational learning leads to Nash equilibrium. Econometrica 61(5), 1019–1045. — Theorem. Bayesian learners whose priors put positive mass on the truth about one another converge (the grain-of-truth condition). Taken: the condition, and that it is what a self-model cannot satisfy. Ch. 10, app. A.
  • Nachbar, J. H. (1997). Prediction, optimization, and learning in repeated games. Econometrica 65(2), 275–309; and (2005), Beliefs in repeated games, Econometrica 73(2), 459–480. — Impossibility results. Properties a class of policies needs to admit a grain of truth cannot all be satisfied for agents of one another's kind. Taken: the impossibility half of the grain-of-truth wall. Ch. 10, app. A.
  • Foster, D. P. & Young, H. P. (2001). On the impossibility of predicting the behavior of rational agents. Proceedings of the National Academy of Sciences 98(22), 12848–12853. — Impossibility result. Taken with Nachbar. Ch. 10, app. A.
  • Leike, J., Taylor, J. & Fallenstein, B. (2016). A formal solution to the grain of truth problem. Proceedings of the 32nd Conference on Uncertainty in Artificial Intelligence (UAI); arXiv:1609.05058. — Construction. A class of policies containing all computable policies and the Bayes-optimal ones, built on reflective oracles; limit computable, not computable (verified against the text, 3 Nov). Taken: the resolution, and its price. Ch. 10, app. A.
  • Schmidhuber, J. (2007). Gödel machines: fully self-referential optimal universal self-improvers. In B. Goertzel & C. Pennachin (eds), Artificial General Intelligence, Springer, 199–226. — Construction. A self-modifier that rewrites itself only on a proof that the rewrite raises expected utility. Taken: as the boundary case, optimal and inert for the reasons Rice and Löb give. Ch. 10, app. A.
  • Pearl, J. (2009). Causality: Models, Reasoning, and Inference, 2nd ed. Cambridge University Press. — Theorems of identifiability. Effects of interventions are not in general identifiable from observational data without assumptions. Taken, with a hedge: an autonomous agent must experiment, and the experiments must be scored by something it does not control. Ch. 2, app. A.
  • Wald, A. (1950). Statistical Decision Functions. Wiley. — Theorem (complete class). Under regularity conditions every admissible decision rule is Bayes or a limit of Bayes rules. Taken: the narrow optimality claim — the decision procedure forced on every party inside the court is not dominated. Chs. 0, 16, app. A.

Forecasting and scoring

  • Dawid, A. P. (1984). Statistical theory: the prequential approach. Journal of the Royal Statistical Society A 147(2), 278–292. doi:10.2307/2981683. — Principle. A forecaster is assessed on nothing but its issued forecasts and the realised outcomes. Taken as the file format of the ledger. Verified against the text, 3 Nov: the principle at §5.1, p. 281, offered as a desideratum by analogy with the likelihood principle; the game reading is Shafer and Vovk's, not Dawid's. Chs. 0, 2, 12, app. A.
  • Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review 78(1), 1–3. — Construction. The quadratic score. Taken: as the first proper score; not used in jack. Ch. 2, app. A.
  • Good, I. J. (1952). Rational decisions. Journal of the Royal Statistical Society B 14(1), 107–114. — Construction. The logarithmic score, and the argument that a fair fee rewards belief. Taken: the score jack uses, and the repository's namesake. Ch. 2, app. A.
  • Savage, L. J. (1971). Elicitation of personal probabilities and expectations. Journal of the American Statistical Association 66(336), 783–801. — Theorem. Characterisation of proper scoring rules for probabilities. Taken: propriety as the condition that truthful reporting is optimal. Ch. 2, app. A.
  • Gneiting, T. & Raftery, A. E. (2007). Strictly proper scoring rules, prediction, and estimation. Journal of the American Statistical Association 102(477), 359–378. — Theorem. A score is (strictly) proper if and only if it is a subtangent of a (strictly) convex function (Theorem 2.1; subgradient in the categorical case, Theorem 3.2); the log score, "proposed by Good (1952)", is strictly proper. Taken: the definition and the log score's status; the ln K mixture argument is the book's, not theirs. Verified against the technical-report version, 3 Nov. Ch. 2, app. A.
  • Blackwell, D. (1951). Comparison of experiments. Proceedings of the Second Berkeley Symposium on Mathematical Statistics and Probability, 93–102. — Theorem. An experiment that is a garbling of another is worth less under every utility. Taken lightly: the ordering under which waiting for information has value, and the reason the base rate is the null. Ch. 7, app. A.

Testing by betting

  • Ville, J. (1939). Étude critique de la notion de collectif. Gauthier-Villars. — Theorem (maximal inequality). For a non-negative supermartingale with E₀ = 1, P(supₜ Eₜ ≥ c) ≤ 1/c. Taken: 20 is α = 0.05 at any stopping time. Chs. 6, 12, app. A.
  • Vovk, V. (1990). Aggregating strategies. Proceedings of the Third Annual Workshop on Computational Learning Theory, 371–386. doi:10.5555/92571.92672. — Theorem. For mixable losses an aggregating algorithm attains a constant regret bound of the form c ln K. Taken: that the ln K bound is not peculiar to the log score, so the log score is chosen by the bound together with conditioning, not by the bound alone. The original is not open; the bound (ln N)/η for η-mixable losses, and the Brier score's mixability at η = 1, were checked against van Erven, Reid and Williamson, "Mixability is Bayes risk curvature relative to log loss", JMLR 13 (2012), §§1.1, 5 (which credits the Brier result to Vovk and Zhdanov 2009), 3 Nov. A first correction said other losses need "another learning rate"; for Brier that was wrong. Ch. 2, app. A.
  • Shafer, G. & Vovk, V. (2019). Game-Theoretic Foundations for Probability and Finance. Wiley. doi:10.1002/9781118548035. — Framework. A test is a betting strategy; wealth is evidence. Taken as the reading of every promotion statistic. Verified against chs. 1–2 and Exercises 2.10 and 8.8, 3 Nov: capital, testing by betting, 1/α as evidence; Ville's inequality appears there as a bound on upper probability. Ch. 6, app. A.
  • Ramdas, A., Grünwald, P., Vovk, V. & Shafer, G. (2023). Game-theoretic statistics and safe anytime-valid inference. Statistical Science 38(4), 576–601; arXiv:2210.01948. — Theory. Test (super)martingales, e-processes, Ville's inequality and optional stopping as the basis of safe anytime-valid inference (the paper's own key words). Taken: the general form the charter's A4.3–4 instantiate. Ch. 6, app. A.
  • Cover, T. M. (1991). Universal portfolios. Mathematical Finance 1(1), 1–29. doi:10.1111/j.1467-9965.1991.tb00002.x. — Theorem. The universal portfolio, a mixture over constant-rebalanced portfolios, matches the best one in hindsight in exponential growth rate for arbitrary bounded sequences (§1, eq. 1.7); a polynomial factor holds under regularity conditions, among them that the optimum lies in the interior (Theorem 6.1). Taken: the form a stake chosen by mixture rather than by plug-in would take, and the source of the guarantee. Verified against the text, 3 Nov. Chs. 12, 15.
  • Waudby-Smith, I. & Ramdas, A. (2024). Estimating means of bounded random variables by betting. Journal of the Royal Statistical Society Series B 86(1), 1–27. doi:10.1093/jrsssb/qkad009. — Construction. Confidence sequences for bounded means from betting, with Ville's inequality. The strategies compared include predictable plug-ins (§3.1; the authors' earlier name for the approach was "predictable mixture", §2.1) and Kelly-type and diversified bets (aGRAPA, dKelly, §4.4). No regret bound against the best constant bet is the paper's own; regret appears in its history of Cover and Krichevsky–Trofimov (§6). Taken: the practical candidates for replacing A4.3's plug-in stake. Checked against arXiv:2010.09686v7, 3 Nov. Chs. 12, 15.
  • Bahadur, R. R. & Savage, L. J. (1956). The nonexistence of certain statistical procedures in nonparametric problems. Annals of Mathematical Statistics 27(4), 1115–1122. — Theorem. Under three conditions on a family (finite means; every real number some member's mean; convexity, §2), which the family of all distributions with bounded but unknown support also meets, no test of the mean has power exceeding its size. Taken: test only outcomes whose range is fixed in advance; report the rest. Verified against the text, 3 Nov. Chs. 5, 6, app. A.
  • Alpern, B. & Schneider, F. B. (1985). Defining liveness. Information Processing Letters 21(4), 181–185. — Theorem. Every property is the intersection of a safety and a liveness property; safety is refutable by a finite prefix, liveness never. Taken: replay for safety, writs and a statistic for liveness. Verified against the text, 3 Nov (Theorem 1; the informal distinction credited to Lamport 1977). Ch. 5, app. A.
  • Lamport, L. (1977). Proving the correctness of multiprocess programs. IEEE Transactions on Software Engineering SE-3(2), 125–143. — Definitions. Safety and liveness named. Taken: the distinction. Ch. 5, app. A.

Control and systems

  • Conant, R. C. & Ashby, W. R. (1970). Every good regulator of a system must be a model of that system. International Journal of Systems Science 1(2), 89–97. — Necessity theorem under conditions. "The simplest optimal regulator R of a reguland S produces events R which are related to the events S by a mapping h : S → R" (§4), with Z = ψ(S, R), entropy of Z minimised, the distribution over S fixed; optimal non-models exist but are "unnecessarily complex". Taken in that narrow form. Verified against the text, 3 Nov; the first draft wrote ψ(D, R) and called extra inputs "wrong", which the paper does not. Chs. 0, 1, 16, app. A.
  • Ashby, W. R. (1956). An Introduction to Cybernetics. Chapman & Hall. — Theorem (requisite variety, ch. 11). V(Z) ≥ V(D) − V(R). Taken: several proposers of several families; no vendor in law. Verified against the text, 3 Nov: the law at S.11/7 (p. 206), under S.11/5's condition, relaxed in S.11/9; the channel-capacity form at S.11/11. Chs. 0, 1, app. A.
  • Beer, S. (1972). Brain of the Firm. Allen Lane. — Model. The Viable System Model. Taken: the nearest institutional ancestor of the five roles; cited in the founding conversation and the essay, not the charter. Ch. 0.
  • Burns, B., Grant, B., Oppenheimer, D., Brewer, E. & Wilkes, J. (2016). Borg, Omega, and Kubernetes. ACM Queue 14(1). — Engineering principle. Level-triggered reconciliation loses nothing on a missed event. Taken: the clerk reconciles state; it does not handle events. Chs. 1, 5, 11, app. A.
  • Schneider, F. B. (1990). Implementing fault-tolerant services using the state machine approach: a tutorial. ACM Computing Surveys 22(4), 299–319. — Construction. A deterministic fold over an ordered log is a replicated state machine. Taken: fold(law, log). Ch. 1, app. A.
  • Lamport, L. (1978). Time, clocks, and the ordering of events in a distributed system. Communications of the ACM 21(7), 558–565. — Theorem and construction. Order from causality, not clocks. Taken: the chain orders the record; the kernel never reads a clock. Chs. 1, 3, app. A.

Law

  • Hart, H. L. A. (1961). The Concept of Law. Clarendon Press. (Ch. V.) — Design principle by elimination. Primary rules plus recognition, change and adjudication; a regime lacking one cannot answer one of three questions. Taken: C0 is exactly the three. Verified against the third edition (2012), 3 Nov: the three defects and their remedies (ch. V §3); the three together are sufficient for "indisputably a legal system", and Hart does not claim them necessary — the defect-for-defect reading is the book's. Chs. 3, 4, app. A.

  • Fuller, L. L. (1964). The Morality of Law. Yale University Press. (Ch. II.) doi:10.12987/9780300191653 (revised edition, 1969). — Design principle with a catalogue of failures. Eight ways to fail to make law. Taken: eight checks; congruence as the one prompt-based harnesses fail. Verified against the revised edition, 3 Nov: the eight in Fuller's order (ch. II, pp. 38–39); a total failure in any one yields no legal system; congruence failures make obedience futile. Chs. 4, 8, 14, app. A.

  • Coke, E. (1610). Dr Bonham's Case, 8 Co. Rep. 107a. — Maxim. Nemo iudex in causa sua. Taken: self_judged, by principal. Ch. 4, app. A.

  • Habeas Corpus Act 1679 (31 Cha. 2 c. 2). — Institution. A writ derived from custody and the clock that no custodian may let lapse. Taken: the clerk's writs. Ch. 3, 7.

  • Blackstone, W. (1765–69). Commentaries on the Laws of England. Clarendon Press. — Doctrine. Precedent. Taken: a rule enters from a witnessed case. Ch. 4.

  • Jefferson, T. (1789). Letter to James Madison, 6 September. — Maxim. The earth belongs to the living. Taken: windows are maxima; law lapses when undefended. Ch. 9.

  • Madison, J. (1788). The Federalist No. 51. — Maxim. Enable the government to control the governed; then oblige it to control itself. Taken: the build order C0–C1 then C2. Chs. 9, 11.

  • Ostrom, E. (1990). Governing the Commons. Cambridge University Press. — Empirical design principles. Graduated sanctions; monitors accountable to the monitored; nested enterprises. Taken: the escalation ladder; the sovereign as a role, not an operator. Chs. 12, 15.

  • Hayek, F. A. (1973). Law, Legislation and Liberty, vol. 1: Rules and Order. University of Chicago Press. — Empirical regularity. Grown order against made order. Taken: rules are found by cases. Chs. 11, 15, app. A.

  • Suber, P. (1990). The Paradox of Self-Amendment. Peter Lang. — Analysis. An entrenched rule is first made mutable, as a separate move. Taken: decision transmute before amend [SPEC §10.13]. Chs. 4, 9.

  • Arrow, K. J. (1951). Social Choice and Individual Values. Wiley. — Theorem. No aggregation of several orderings into one satisfies unrestricted domain, Pareto, independence and non-dictatorship. Taken: the one-owner assumption is a limit of the construction, not an extension (ch. 15 §15.6). Ch. 15.

Enforcement

  • Anderson, J. P. (1972). Computer Security Technology Planning Study. ESD-TR-73-51, vol. II, USAF Electronic Systems Division. — Design principle. The reference monitor: "tamper proof", "always invoked", "small enough to be subject to analysis and tests to assure that it is correct" (§3.4, p. 17). Taken: the judge and the kernel are built to the three. Verified against the DTIC copy, 3 Nov. App. A.
  • Saltzer, J. H. & Schroeder, M. D. (1975). The protection of information in computer systems. Proceedings of the IEEE 63(9), 1278–1308. — Design principles. Economy of mechanism, fail-safe defaults, complete mediation, least privilege, separation of privilege, open design. Taken as listed. App. A.
  • Dennis, J. B. & Van Horn, E. C. (1966). Programming semantics for multiprogrammed computations. Communications of the ACM 9(3), 143–155. — Construction. Capabilities. Taken: a key with a power is a capability. Ch. 4, app. A.
  • Miller, M. S. (2006). Robust Composition: Towards a Unified Approach to Access Control and Concurrency Control. PhD thesis, Johns Hopkins University. — Design principle. Authority as a reference held, not a rule obeyed. Taken: nothing to inject. Ch. 4, app. A.
  • Thompson, K. (1984). Reflections on trusting trust. Communications of the ACM 27(8), 761–763. — Argument with a construction. A checker cannot be trusted from its source. Taken: two checkers. Chs. 7, 12, app. A.
  • Wheeler, D. A. (2005). Countering trusting trust through diverse double-compiling. Proceedings of the 21st Annual Computer Security Applications Conference (ACSAC), 33–48. — Construction. Agreement of independent lineages is evidence. Taken: the differential over every prefix. Ch. 12, app. A.
  • Gordon, M. J., Milner, R. & Wadsworth, C. P. (1979). Edinburgh LCF. Lecture Notes in Computer Science 78, Springer. — Construction. A small kernel through which every proof passes. Taken: the shape of the trusted base. App. A.
  • Barendregt, H. & Wiedijk, F. (2005). The challenge of computer mathematics. Philosophical Transactions of the Royal Society A 363, 2351–2375. — Criterion (de Bruijn). A proof checker may be slow and simple if it is small and right. Taken: 450 lines, read entire. App. A.

Decision under uncertainty, measurement and growth

  • Arrow, K. J. & Fisher, A. C. (1974). Environmental preservation, uncertainty, and irreversibility. Quarterly Journal of Economics 88(2), 312–319. doi:10.2307/1883074. — Theorem. Quasi-option value of waiting before an irreversible act. Taken: gates only before the irreversible. Verified against the text, 3 Nov: "quasi-option value" is their term (p. 315); the result is in a two-period model with risk neutrality; they price irreversibility rather than forbid it (p. 319), so the converse and the gate are the book's. Ch. 9, app. A.
  • Henry, C. (1974). Investment decisions under uncertainty: the "irreversibility effect". American Economic Review 64(6), 1006–1012. — Theorem. The same result independently. Taken: as corroboration. Ch. 9, app. A.
  • Dixit, A. K. & Pindyck, R. S. (1994). Investment under Uncertainty. Princeton University Press. — Textbook. Taken: the general treatment. App. A.
  • Goodhart, C. A. E. (1975). Problems of monetary management: the U.K. experience. Papers in Monetary Economics, vol. I, Reserve Bank of Australia. — Empirical regularity. A regularity collapses under control pressure. Taken: proper scores on outcomes the party does not control; procedure counted against object. Chs. 0, 2, 10, app. A.
  • Campbell, D. T. (1979). Assessing the impact of planned social change. Evaluation and Program Planning 2(1), 67–90. — Empirical regularity. The same for social indicators. Taken: as corroboration. App. A.
  • Manheim, D. & Garrabrant, S. (2018). Categorizing variants of Goodhart's law. arXiv:1803.04585v4. — Taxonomy. Four categories — regressional, extremal (model insufficiency; change in regime), causal, adversarial — verified against the text, 3 Nov. Taken: the adversarial case is the court's. Ch. 2, app. A.
  • Gall, J. (1977). Systemantics: How Systems Work and Especially How They Fail. Quadrangle. — Empirical regularity. A working complex system evolved from a working simple one. Taken: layers promoted before the next begins; stated as a regularity, never a theorem. Ch. 11, app. A.

Self-improving agents (for chapter 1)

  • Zhang, J., Hu, S., Lu, C., Lange, R. & Clune, J. (2025). Darwin Gödel Machine: open-ended evolution of self-improving agents. arXiv:2505.22954v2 (26 Sep 2025), Appendix H, "Case Study: Solving Hallucination". — Construction and a reported failure. Special tokens logged on real tool use; checking functions hidden from the agent ("objective hacking … occurs more frequently when these functions are not hidden"); node 96 improved the logging honestly, node 114 "removed the logging of special tokens … (despite instructions not to change the special tokens), effectively bypassing our hallucination detection function" (verified against the text, 3 Nov). Taken: the judge fixed, the path to its evidence in the free edge — the first way to get the shape wrong. Ch. 10.
  • Good, I. J. (1965). Speculations concerning the first ultraintelligent machine. Advances in Computers 6, 31–88. — Conjecture. The intelligence explosion. Taken: the repository's namesake, with Good (1952). Ch. 0.

Election and learning (for chapter 5; listed now because chapter 2 names them as entrants)

  • Freund, Y. & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences 55(1), 119–139. — Theorem. Hedge: regret within ln K of the best expert; at η = 1 the Bayes mixture. Taken: the aggregate; ln K the only constant. Ch. 7.
  • Herbster, M. & Warmuth, M. K. (1998). Tracking the best expert. Machine Learning 32(2), 151–178. — Theorem. Fixed Share. Taken: an entrant, not law. Ch. 7.
  • Auer, P., Cesa-Bianchi, N., Freund, Y. & Schapire, R. E. (2002). The nonstochastic multiarmed bandit problem. SIAM Journal on Computing 32(1), 48–77. — Theorem. EXP3. Taken: the invitation under bandit feedback; an entrant. Ch. 7.
  • Joulani, P., György, A. & Szepesvári, C. (2013). Online learning under delayed feedback. Proceedings of the 30th International Conference on Machine Learning (ICML), 1453–1461. — Theorem. Interleaving D + 1 instances handles delay D. Taken: C2b's four EXP3 instances [BRIEF-C2b]. Ch. 7.
  • Garrabrant, S., Benson-Tilsen, T., Critch, A., Soares, N. & Taylor, J. (2016). Logical induction. arXiv:1609.03543. — Construction. A market of traders over logical sentences whose prices cannot be exploited in the long run. Taken: nothing — named in the founding essay's quoted list, so the book says in one sentence that the court does not approximate it; the shared ancestor is Shafer–Vovk. Ch. 16.

The record (primary sources)

  • renavondata/jack — CHARTER.md (with amendments A1–A5), SPEC.md, LAW.md, REPORT.md, BRIEF-C1.md, RUNBOOK.md, log.jsonl, kernel/, kernel2/, judge/, tools/, harness/, sim/, pins/, transcripts/. Cited by section, function, seq or path as plan.md §2 sets out. REPORT.md is cited as a transcript (what counsel reported), never as the source of a number the author has not recomputed.
  • renavondata/assize — REPORT.md (the layer verdicts, append-only), reports/L*/REPORT.md, reports/L*/forecasts.jsonl, harness/ledger.py. The predecessor and the baseline.
  • The owner's sessions and notes — Claude Code session transcripts and per-project memory notes from the owner's machine steel, supplied 3 November 2026; extracted to data/lineage-sessions.md (project-related words only, dated; the files themselves are not in the repository). Cited as [data/lineage-sessions.md, <project>].
  • Project documents — claude/small-core-charter.md (counsel's draft of the charter with A1; the repository's CHARTER.md governs where they differ), claude/literature-review.md, claude/monograph-a-court-for-agents.md, claude/acts-weekly.md with claude/guy_acts.py and the acts CSVs.
  • Freeman, G. (2026). The Science Was Done. The Engineering Wasn't. gfrm.in, 1 November. gfrmin/gfrminsite, content/posts/the-science-was-done/index.md. The founding essay; chapter 1's source. Erratum E1 in plan.md.
  • Freeman, G. (2026). Assize at the seal. gfrm.in. content/posts/assize-at-the-seal/. The predecessor's own account; chapter 9. Front matter.