The placement map says a piece of mathematics goes where its role is, and that XVI–XX stay empty until material arrives that fits nowhere else. This material has arrived, and it fits nowhere else in its register. Book XVII holds probability as a machine defines it — a variable with no second moment has variance zero, as a theorem (variance_of_not_memLp). Book XIII holds what a proof check establishes. Neither holds what a person may conclude from data. That is this volume, named after its material rather than the reverse.
Its shape is borrowed, and the debt is stated: Jordan Ellenberg's How Not to Be Wrong (Penguin, 2014) organises this ground into Linearity, Inference, Expectation, Regression and Existence. Those five parts are placed across the series, not all here:
| Ellenberg's part | where it lives | why there |
|---|---|---|
| Linearity | Book X (“the floor is not linear”) and Book XVIII | locally straight, globally curved is the chain rule; the non-linear floor is Book X's thesis |
| Inference | here, ch 1–2 | what a test or a surviving sample licenses |
| Expectation | here, ch 3 | the reasoning register; the definition is Book XVII's |
| Regression | here, ch 4–5 | the phenomenon, and a held dataset to test it on |
| Existence — geometry | Book XIII ch 9 | already written: Beltrami's 1868 model is the consistency question Bolyai's non-Euclidean geometry raised |
| Existence — public opinion | Book X, planned chapter | whether 55 states can have one negotiating voice is Condorcet and Arrow, and it is Book X's own question |
Ellenberg is a guide to the ground, not an authority on it. Nothing on these pages is cited to him that a primary text can carry, and no passage of his is reproduced.
| text | pp | what it carries, by address |
|---|---|---|
| Evans & Rosenthal, Probability and Statistics: The Science of Uncertainty | 774 | ch 3 Expectation, p. 129 · §5.1, p. 254 (a survivorship example, heart transplants) · §6.3.3 Testing Hypotheses and P-Values, p. 332 · §9.3 The Problem with Multiple Checks, p. 509 · ch 10 Relationships Among Variables, p. 511 · §10.3 regression, p. 538 |
| NOAA NHC, HURDAT2 Atlantic best-track file | — | 1,988 storms, 1851–2025, Atlantic basin only |
| Wald, A Method of Estimating Plane Vulnerability Based on Damage of Survivors (Statistical Research Group memoranda, 1943; CNA reprint CRC 432, 1980) | 100 | eight Parts; Part I, p. 1, first issued as SRG memo 85 · Part V, subdividing the plane, p. 56 · the engines / fuselage / fuel-system example, pp. 63–65 · “guides for locating protective armor”, p. 89 |
| Galton, “Regression towards Mediocrity in Hereditary Stature”, J. Anthropological Institute 15 (1886), pp. 246–263 | 21 | the law of regression, p. 252: the height-deviate of the offspring is, on the average, two-thirds of the height-deviate of its mid-parentage |
| Ellenberg, How Not to Be Wrong | 532 | structure only, as above |
Every address and count in this section is printed by book20/index-verify.py, which reads the ledger row (docs/floor-texts.tsv), re-hashes the file, reads the offset between PDF and printed pages off the book's own contents page, and locates each section by its heading. It caught one error before this page was written: §9.3 begins on p. 509, not 510. The Wald reprint is a 2003 scan with a rough text layer, so only passages the script found verbatim are addressed here.
| ch | title | state |
|---|---|---|
| 1 | The Planes That Came Back — survivorship: why the sample that reached you is not the population, from Wald's equations to the Stanford transplant study to the 66 pages this site no longer shows | written 2026-09-27 — ch01-verify.py, 0 FAIL |
| 2 | Many Tries, One Winner — multiple comparisons, and this corpus's own numerical coincidences, each audited by a script that counts the search space before it judges the match | written 2026-09-27 — ch02-verify.py, 0 FAIL |
| 3 | What to Expect — expected value as a decision rule, and why the best error rate is rarely zero | written 2026-09-27 — ch03-verify.py, 0 FAIL |
| 4 | Back Toward the Mean — regression to the mean as a phenomenon, before it is a model | written 2026-09-27 — ch04-verify.py, 0 FAIL |
| 5 | The Storms Nobody Counted — 175 years of Atlantic storms, and whether the early deficit is the climate or the counting | written 2026-09-27 — ch05-verify.py, 0 FAIL; verdict open |
Chapter 2's first cases are named in advance, so that the choice of cases cannot follow the results: the withdrawn g6_equals_schumann, and the hardback's 666-page limit, which CLAUDE.md calls resonant with 111 Hz. For each, the observable is fixed before any run: how many constants and combinations were available to match, and the chance of a match at the stated tolerance if nothing but that search were at work.
Chapter 5's observable is also fixed now: named storms per decade, 1851–2025, from the held file. The file alone cannot say whether a low early count is fewer storms or fewer ships to see them, and the chapter will not pretend it can. That separation needs a source on the history of observation that is not held. [OPEN]
This index was first written with chapters 1 and 4 blocked, because Wald's memoranda and Galton's paper were not held and the retellings are not the sources. Both arrived within hours, were hashed into the ledger, and are addressed above. The chapters are now writable. The earlier state is kept here because a volume that says what it lacks should also say when that changed.
What Wald actually contains is a method, not a slogan: equations for the probability that a plane is downed by i hits, estimated from the planes that returned. His worked example is labelled hypothetical. In it the engine area is the most vulnerable part (a single hit downs the plane with probability .39, against .05 for the fuselage), and armour is mentioned once, on p. 89, where the tables are offered as “guides for locating protective armor.” Chapter 1 is written from that, not from the one-line version of the story.
Galton states the law in one sentence on p. 252 and warns on the next page that its converse is not its numerical opposite. The held statistics text never names the phenomenon (the script searches it four ways and finds nothing), so chapter 4 is written from Galton, with Evans & Rosenthal for the model.
For the chapter placed in Book X (55 states, one voice), the social-choice sources also arrived. Condorcet's 1785 Essai is held: a Gallica scan with no text layer, read by OCR of the pages cited. The three-candidate cycle is on p. lvii of the Discours préliminaire (“de deux quelconques des trois propositions … résulte une conclusion contraire à la troisième”), with the sixty-voter example on p. lviii; the earlier example on restrictions of commerce, pp. lii–liv, which Suzumura calls Condorcet's first extended illustration, is a different inconsistency — the option carried by majorities on each proposition is the one that “seemed to have the fewest votes” — and the chapter will keep the two apart. Arrow's 1951 book is not held. Held in its place: Fey's A Straightforward Proof of Arrow's Theorem (2014), fully general (weak orders, any finite electorate, three or more alternatives); Suzumura's introduction to the Handbook of Social Choice and Welfare (2001); and Suppes's pre-history of Arrow's book (2005). A preprint claiming to prove social choice possible is also on the desk and is not used: an unrefereed claim to overturn a theorem is not a source for it. [OPEN] for Arrow 1951 only.
All five chapters can now be written. Chapter 5's verdict stays open until a source on the history of Atlantic observation is held.