Scope and Method

The objective was to strengthen the paper’s foundations without changing its register. Footnotes were chosen over parenthetical citation for that reason.

Every substantive factual, historical, economic, organizational, or technical assertion in the paper was assessed against three questions: is it a claim about the world that a skeptical reader could reasonably challenge; does an authoritative source support it; and is that source seminal rather than a recent summary of seminal work. Ninety-three assertions met all three tests and now carry a footnote. Eleven met the first test but not the second and were listed as open items; six of those have since been resolved in the text and are recorded as closed in Section 3.

Four editorial principles governed the selection.

1. Primary over secondary. Chandler rather than a business-history survey; Codd rather than a database textbook; the Columbia Accident Investigation Board report rather than commentary on it.
2. Seminal over current. Where an idea has a clear origin, the origin is cited even when more recent treatments exist. Simon 1971 on attention scarcity is more useful than any contemporary restatement of it.
3. Standards over vendor material. W3C PROV, ISO/IEC 42001, NIST AI 100-1, and Regulation (EU) 2024/1689 carry the governance claims. No vendor documentation is cited anywhere in the paper.
4. Restraint on trade press. Six Harvard Business Review articles are cited, each because it is the canonical statement of its idea rather than a convenient summary: Davenport on ERP, Spear and Bowen on Toyota, Garvin on measurable learning, Argyris on double-loop learning, Darling et al. on the after-action review in civilian settings, Kaplan and Norton on measurement, and Carr on commoditized IT. No consulting reports and no encyclopedia references appear anywhere.

What Changed in the Paper

• Ninety-three footnotes, rendered at the foot of the page rather than as endnotes, so that the argument reads uninterrupted.
• A new section, Intellectual Lineage and Original Contributions, placed after the Conclusion. It credits the five disciplines the paper inherits from, states the nearest prior art explicitly, and tabulates what is original.
• A new Appendix E · References in Chicago style, listing all works cited.
• Glossary terms in Appendix A are now marked to distinguish original coinages from established usage.
• Two paragraphs were added rather than merely annotated: one in Part IX on asset-stock accumulation, and one in Part X cautioning that measures of reasoning quality are more distortable than measures of throughput. Both were added because the citation made an unstated step in the argument visible.

Nothing was rewritten, no argument was softened, and no section was restructured. The paper is four pages longer in body text and fifteen pages longer including the new apparatus.

1 · Annotated Bibliography

Grouped by the part each work principally supports. The complete list appears as Appendix E of the paper; what follows is the load-bearing subset, with a note on why each matters.

Part I · Economic history and the migration of advantage

• *Chandler, The Visible Hand (1977).* The foundation of the entire part. Chandler’s argument that modern management was produced by the coordination requirements of the railroads, not by managerial genius, is what licenses the paper’s claim that organizational form follows the binding constraint.
• *Perez, Technological Revolutions and Financial Capital (2002).* Supplies the periodization structure. Perez’s installation-and-deployment phasing is the closest existing model to the paper’s bottleneck migration, and citing it prevents the reader from thinking the schema is invented from nothing.
• *Beniger, The Control Revolution (1986).* The most valuable and least obvious citation in the section. Beniger reads scientific management, cost accounting, and statistical process control as information technologies built to serve a production constraint — which is precisely the paper’s reading, and it is better to acknowledge that than to appear to have discovered it.
• *Johnson and Kaplan, Relevance Lost (1987).* Carries the Carnegie cost-accounting claim, which is otherwise the kind of detail a hostile reader checks first.
• Codd (1970); Davenport, "Putting the Enterprise into the Enterprise System" (1998). The Information Age spine. Davenport in particular supports the claim about SAP’s single-representation insight, which is frequently asserted and rarely sourced.
• Simon, "Designing Organizations for an Information-Rich World" (1971). Cited twice, in Parts I and II. Simon’s observation that a wealth of information creates a poverty of attention is the hinge of the whole argument and predates it by fifty-five years.
• *Brynjolfsson, Li, and Raymond (QJE, 2025); Noy and Zhang (Science, 2023); Peng et al. (2023).* Empirical support for the execution-commoditization claim. This was the weakest-evidenced assertion in the original draft and is now the best-evidenced.
• *Levinson, The Box (2006); Sturgeon (2002).* Support the production-commoditization mechanism, which the draft asserted without explanation.

Part II · Judgment and decision science

• *Agrawal, Gans, and Goldfarb, Prediction Machines (2018).* The single most important addition to the paper. This is the nearest prior art to the core economic claim and its omission would have been the first thing a well-read reviewer noticed. It is cited in Part II and addressed at length in the new lineage section, where the paper’s departure from it is stated explicitly.
• Kahneman and Klein, "Conditions for Intuitive Expertise" (2009). The honest treatment of when expert intuition is trustworthy, which the paper needs because it distinguishes judgment from intuition and should not appear to dismiss the latter.
• Tversky and Kahneman, "The Framing of Decisions" (1981). Carries the framing-errors claim.
• Nutt, "Surprising but True" (1999). The empirical basis for treating rejected alternatives as an asset: failure correlates with the number of alternatives not considered. This converts a plausible assertion into a supported one.
• *Tetlock, Expert Political Judgment (2005); Brier (1950).* Establish that judgment quality is measurable, which the paper requires for Part X to be more than aspiration.
• Bainbridge, "Ironies of Automation" (1983); Parasuraman and Riley (1997). Support the bad-judgment-scales argument. Bainbridge is forty-three years old and describes the present situation exactly.
• Dell’Acqua et al., "Navigating the Jagged Technological Frontier" (2023). Supports the fluency-versus-reliability decoupling with field evidence rather than anecdote.

Part III · Institutional memory and its loss

• Walsh and Ungson, "Organizational Memory" (1991). The canonical structural account. Cited twice. Without it the paper appears to be inventing a problem that a literature has been mapping for thirty-five years.
• Argote, Beckman, and Epple (1990); Argote and Epple (1990). Convert knowledge loss from assertion to measurement. These papers demonstrate empirically that organizational knowledge depreciates with turnover and time, which is the paper’s central diagnostic claim.
• *Polanyi, The Tacit Dimension (1966); Leonard and Swap, Deep Smarts (2005).* Carry the retiring-expert argument.
• Nelson and Winter (1982); Wenger (1998). Support the reorganization claim — that the knowledge was in the group rather than in any individual.
• *Tufte, The Cognitive Style of PowerPoint (2006); Columbia Accident Investigation Board (2003).* The format-loss claim. The CAIB finding that critical engineering analysis had been reduced to bulleted slides is the strongest possible evidence and is a government document rather than commentary.
• *Wilensky, Organizational Intelligence (1967).* Carries the government-transition claim and, more importantly, establishes the lineage of the term the paper uses. Given the prior-art work already done on this question, citing Wilensky here is a necessary act of intellectual honesty.
• Argyris and Schön (1978); Einhorn and Hogarth (1978). Support the unlearnability argument — that a decision process disconnected from outcomes cannot improve.
• *Mazzucato and Collington, The Big Con (2023).* Carries the consultant-extraction point with a serious source rather than an anecdote.

Part IV · Why the enterprise estate stops short

• Orlikowski, "Learning from Notes" (1992). The most important citation in this part. Orlikowski demonstrated empirically that knowledge-sharing systems fail where incentives do not reward sharing — which is the paper’s stated capture principle, arrived at independently and much later.
• Blair and Maron (1985). An unexpected and durable citation: practitioners who believed they were retrieving most relevant documents were retrieving roughly a fifth of them. Forty years old and directly applicable to retrieval-augmented generation.
• Lewis et al., "Retrieval-Augmented Generation" (2020). Cited so that the paper’s criticism is visibly of the corpus rather than of the method, which protects it from the obvious rebuttal.
• Gruber (1993); Berners-Lee, Hendler, and Lassila (2001). Carry the semantic layer discussion.
• Whittaker and Sidner, "Email Overload" (1996). Supports the conversation-as-burial-ground claim.

Part V · Institutions that made judgment structural

• *Shook, Managing to Learn (2008); Sobek and Smalley (2008); Spear and Bowen (1999).* Carry the A3 discussion. Shook is the primary source and should be the one quoted if the paper is ever excerpted.
• *Bryar and Carr, Working Backwards (2021); Bezos, 2015 shareholder letter.* The Amazon claims. The 2015 letter is the primary source for the one-way / two-way door distinction and should be cited in preference to any secondary account.
• *Columbia Accident Investigation Board (2003); Vaughan, The Challenger Launch Decision (1996).* Carry the recorded-dissent argument. Vaughan’s normalization of deviance is the deeper mechanism and worth the reader’s time.
• *U.S. Army, TC 25-20 (1993); Darling, Parry, and Moore (2005).* The after-action review. The doctrine itself is the primary source; the HBR piece documents the civilian transfer.
• Gompers, Kaplan, and Mukharlyamov (2016). Supports the private-equity claim, though only partially — see the audit in Section 4.
• Repenning and Sterman (2001). Supports the paper’s turn at the end of the part: why these practices decay without sustained enforcement. This is what makes the architectural argument necessary rather than merely convenient.

Parts VI–VII · Architecture, provenance, and governance

• *Zachman (1987); The Open Group, TOGAF (2022); Ross, Weill, and Robertson (2006).* The layering convention. Ross et al. also supply the operating-model argument that the paper relies on in Part IX.
• W3C PROV-DM (2013). The standard vocabulary for provenance and the closest existing standard to what the judgment layer requires. Cited three times.
• Guyatt et al., "GRADE" (2008). Carries the graded-evidence-reliability construct. Citing a clinical evidence standard rather than a management source materially strengthens this claim.
• *Toulmin, The Uses of Argument (1958).* The antecedent for structured argument representation. Its inclusion in the nine-constructs footnote prevents the reasoning construct from appearing to have no theory behind it.
• Mitchell et al., "Model Cards" (2019); Gebru et al., "Datasheets for Datasets" (2021). Support the design commitment that machine participants carry documented identity.
• Noy and Klein (2004). Establishes that ontology versioning is a distinct and harder problem than schema migration, which the paper asserts without argument.
• *Dehghani, Data Mesh (2022). Cited because it reaches a similar conclusion to reason in place* from different premises. Acknowledging the convergence is stronger than ignoring it.
• NIST AI 100-1 (2023); ISO/IEC 42001:2023; Regulation (EU) 2024/1689; OECD (2019, amended 2024). The governance quartet, carrying the human-oversight and reconstructibility claims in Parts VI, VIII, and X.

Parts VIII–X · Strategy, economics, and measurement

• Dierickx and Cool, "Asset Stock Accumulation" (1989). The most consequential addition to Part IX. Their distinction between asset stocks and flows, and the concept of time-compression diseconomies, supplies the precise mechanism by which accumulated reasoning becomes defensible. The paper previously asserted this; it now has a theory of it, and a sentence was added to make the connection explicit.
• Barney (1991); Wernerfelt (1984). The resource-based conditions the paper implicitly invokes when it calls accumulated reasoning proprietary and non-substitutable.
• Carr, "IT Doesn’t Matter" (2003). Cited with a caveat noting that Carr was wrong about IT in the strong form of his claim. The caveat is necessary; without it the citation invites the rebuttal rather than answering it.
• Teece, Pisano, and Shuen (1997). Dynamic capabilities, supporting the investor implication.
• *Shapiro and Varian, Information Rules (1999).* Lock-in and switching costs, supporting the portability warning.
• Sterman (1989); Einhorn and Hogarth (1978). Carry the feedback-horizon argument in Part VIII, which is the most technically interesting claim in that part and was previously unsupported.
• Ridgway, "Dysfunctional Consequences of Performance Measurements" (1956). Added deliberately as a counterweight to Kaplan and Norton. A paper proposing eight new metrics should cite the classic warning about what metrics do to behavior, and a new paragraph was added in Part X to act on it.
• Earl and Scott, "What Is a Chief Knowledge Officer?" (1999). Supports the caution about the CKO revival with the empirical record of the first attempt.
• *Stanford HAI, AI Index 2026.* The only current-data source in the paper, carrying the model-access and investment-concentration claims. It will need refreshing annually; this is the one citation with a maintenance obligation.

2 · Insertion Map

All ninety-three footnotes, in document order, showing the sentence each marker follows and the source added.

Markers sit at the end of the sentence they support. Statement lines, pull quotes, and headings carry no markers by design — where a display line needed support, the note was attached to the adjacent body paragraph instead.

Executive Summary

Part I · Every Era Has a Bottleneck

Part II · The Emergence of the Age of Judgment

Part III · The Judgment Crisis

Part IV · Why the Current Enterprise Estate Stops Short

Part V · Judgment Is an Organizational Capability

Part VI · The Cognitive Enterprise

Part VII · The Architecture of Organizational Judgment

Part VIII · Domain Judgment Platforms

Part IX · Economic Implications

Part X · The Next Decade

3 · Claims Audit and Resolutions

Eleven assertions were flagged in the first review. Six have been corrected in the text. Five remain, deliberately.

These were never errors. Most are rhetorically effective and several are probably true. They are the places where a hostile reader would ask for evidence and the paper could not supply it. Six were closed on the principle that a claim which is cheap to fix and conspicuous to leave should not be knowingly exposed; the remaining five are retained because the overstatement is doing legitimate work and the mechanism behind it is cited.

Resolved in the second edition

Retained, knowingly

Two of these — the demographic claim in Part III and the investment-concentration claim in Part IX — are worth closing before external circulation, because both are easily sourced and both sit at load-bearing points in the argument. The remainder are acceptable as rhetoric provided the author knows they are rhetoric.

4 · Original Contributions — Deliberately Uncited

The following concepts are the paper’s own. No citation was added to any of them, and none should be.

Adding a reference to an original concept does not strengthen it; it implies a precedent that does not exist and invites the reader to check. Where a concept is original, the paper now says so directly — in the footnotes for institutional amnesia, Institutional Intelligence, and the nine judgment-layer constructs, and comprehensively in the new lineage section. Three footnotes exist for the sole purpose of stating that a concept is original and identifying what it departs from.

A Note on the Lineage Section

The new section does something slightly unusual for a commercial white paper: it names the nearest prior art and states plainly where the paper departs from it. The argument for doing so is that Prediction Machines is well known to exactly the audience this paper targets, and a reader who recognizes the resemblance and finds it unacknowledged will discount everything else. Naming it first, and specifying the three departures — institution rather than individual, accumulable stock rather than fixed input, and a specified architecture — converts a vulnerability into a demonstration of command over the field.

The section also includes an invitation to send precedent the author has missed, and a statement that the substantive claims do not depend on the originality of the vocabulary. Both are unusual and both are worth keeping. They cost nothing if no precedent surfaces, and they are the difference between a canon and a marketing position if some does.