Articles

An analytical index to the 24 articles on this site: each article under its subject, with a summary and its sections, followed by an alphabetical index of topics.

Tools: Word Check, the probability that an English speaker would understand a word or phrase, how sure the estimate is, and the evidence behind it.

Word squares and the Borgmann search

Borgmann Word Log

Words and phrases the 9×9 search proposed at its dead ends, listed for Eric Chaikin's grades: 67 real words, phrases and names that fit row-5 dead ends, sifted from 3,605 unlisted strings.

Sections: 67 real words and phrases found at row-5 dead ends

Frequency Data vs Word Grades

Can word-frequency data predict the grade an expert gives a word? No: it ranks the calibration list only weakly (correlation about 0.4), runs backwards on the crossword list, and the crossword list's scores say nothing about the expert's.

Sections: How it was tested · 1. Eric's calibration list from the frequency data · 2. The crossword list's scores from the frequency data · 3. Eric's calibration list from the crossword list · Limits

9x9 Column Proposals6 Nov 2026

Round 4 of the Borgmann search on the 9×9 square: the first real six-row partials from a clean scan, what free columns buy, why real words fall short, and how the judging models behaved.

Sections: Summary · Clean scan: first real six-row partials · What free columns buy · Why real words fall short · Judges · Mistral failure · Limits of these results · Decisions needed

Nine by Nine

The hunt for a 9×9 asymmetric word square: nine nine-letter words across and nine more down. Exhaustive passes over a curated 63,091-word lexicon and over a judged lexicon both say no square exists; the page records each pass, the checks on the program, and where the square dies.

Sections: Four searches, not one · Why a second, third and fourth pass at all · Where the added words come from · Does the program actually work? · What would count as a find · What is not yet settled · Where the square actually dies

The Borgmann Search

The method: when an exhaustive search stops because no listed word fits, a language model is asked to break that one wall, proposing real but unlisted words, and is kept out of everything else. Five stages, two searches that use it, and the mistakes it guards against.

Sections: The five stages · Two searches, two kinds of wall · What the method guards against · What it is not

Crossword stacks

Quintuple Stack

A finished grid: five 15-letter entries stacked in rows 6 to 10 of a 15×15, with only four entries below the acceptability score, with its across and down entries, laid out for printing.

Sections: Across · Down

Stack Search Log

Dated reports, newest first, from the search for symmetric 15×15 grids with a five-deep stack: grids 1 to 12 with their Q scores, the first grids with every entry at 50 or more, rotational, mirror and up-down symmetry, and rows proposed by a language model.

Sections: Archive · Rows from the web: one weak fill, and scoring words by use · No new grid at 50; 385 more LLM rows go into rows 2–5 · Grid 12, Q 1.822: a rotational grid at the 50 standard with nothing to grade · Grid 11, Q 1.817: an LLM-proposed row, four words from the 50 standard · No new grid at 50; 395 more LLM-proposed rows to test · Eric's grades: grid 10 is the first rotational grid at the 50 standard, Q 1.810 · An LLM-proposed row: grid 10, four words from the 50 standard · Rotational search: a candidate grid five words from the 50 standard, and two faults fixed · A mirror-symmetry grid with every entry at 50 or more: grid 8, Q 1.820 · The first grid with every entry at 50 or more: grid 7, Q 1.823 (up-down symmetry) · Round 40 ends, with a new best at Q 1.808 (up-down symmetry) · 289 fills under mirror symmetry (round 40, mirror slices 2–5) · 192 more fills, and a new best at Q 1.805 (round 40, rotational slices 2–6) · Five-deep symmetric stacks from a larger lexicon (round 40)

Symmetric Quintuple Stacks

Can a 15×15 crossword hold five stacked 15-letter entries across its middle, with symmetric black squares and fair entries everywhere else? One hand-built grid shows it can; an exhaustive check shows that none of the known stacks can be finished with fair entries alone, and frames are left for a constructor.

Sections: Eric Chaikin’s grid · What the search shows · Can any of them be finished fairly? · Frames to finish by hand · How the frames were made

Words spelled on maps

Border Words of the World

Spell a word from the initials of countries crossed in sequence, never entering a country twice. Two words tie for the longest at eight letters, both walks across Eurasia; the page also tries capital cities and shows what each rule costs.

Sections: The two winners · What the rules cost · Same borders, different labels: capital cities · Running it yourself

City Hops

The border-word game played on the world's 570 cities of a million or more, with neighbouring meaning within a day's drive. The longest chain spells PHYLLOSTACHYS in thirteen hops across eastern China.

Sections: How close counts as neighbouring · The magic word, bouncing round the Levant · Why the answers cluster · Reproducing it

The Cantata Traverse

The same game on the United States: the longest dictionary word spelled by state initials without repeating a state is CANTATA, seven states from the Pacific to the Gulf. What the no-repeat rule costs, and why proper nouns do worse.

Sections: What the no-repeat rule costs · Famous names fare worse, not better · Why the map refuses proper nouns

Words and word lists

Airvent

AIRVENT, written solid, is a real English word: a steady minority spelling of "air vent", about one use in 150 to 200, found in books and on the web for a century though no dictionary enters it. The evidence from dictionaries, word lists, books and the web, and a verdict.

Sections: Dictionaries · Word lists · Books · The web · Verdict · References

Compound Words on the Web14 Nov 2026

How to find closed compounds that dictionaries miss: split solid words into two listed words (Eric Chaikin's technique) and add the dictionaries' two-word headwords, then count the solid, open and hyphenated forms in the same web prose, by sites. Filters for run-ons, word breaks, names and chance meetings. Counts from 100,010 Common Crawl files (238.6 billion words): 128,962 compounds, 47,760 like airvent (solid on 3+ sites, solid form in no dictionary or list), 17,398 solid on most sites yet never listed solid. Splits into three or more parts need the open forms counted first.

Sections: The method · The data so far · Still to come

What the Scrabble Lists Leave Out12 Nov 2026

Do the official Scrabble lists omit words as good as those they include? Judged by use alone, measured in twenty years of Google Books: how much the listed words are used, what is missing, what was removed on purpose, and what is listed but hardly used.

Sections: The test · The yardstick: how much the listed words are used · What is missing · What is included but not used · Limits of this check · Conclusion · Sources and method

Word Birth

AMITULAR, "of or like an aunt", traced to a single sentence by Bergen Evans saying the word ought to exist and did not. How it passed through advertising and reference books into word lists and language models, with a timeline, rival words, and the question whether it is a word.

Sections: The earliest text · The first time it was treated as real · Timeline · Who coined it · The hunt of 2007–08 · What the corpora show · What language models said, 2026 · Is it a word? · Open questions · A note on sources · References

Text windows

Pangrammatic and Panalphabetic Windows

Records for pangrammatic windows (the shortest stretch of ordinary text containing all 26 letters) and their relatives: panalphabetic windows with A to Z in order, reverse, Roman-numeral, lipogrammatic and word-unit windows, with the odds of finding them and the richest sources.

Sections: Records at a glance · How likely is a short pangrammatic window? · How likely is a shorter panalphabetic window? · Which sources are richest? · The record through time · 1. Pangrammatic windows: all 26 letters · 2. Panalphabetic windows: A to Z in order · 3. Reverse panalphabetic windows: Z to A · 4. Windows that are not innocent · 5. Roman windows: M D C L X V I · 6. Lipogrammatic windows: longest stretch avoiding one letter · 7. Word-unit windows, in words

Genetic programming

Puzzle Solve Ledger

Where the puzzle sequences stand after a revision on 13 November 2026: the set is cut from 69 to 32 well-posed puzzles (27 kept, plus five Rosner sequences). Fable 5.2 found a rule reproducing every row of 29 in one attempt each; genetic programming has solved 22 at least once. What was set aside as trivial, duplicate or not well-posed, the rule found for each puzzle, and the sweep record showing why one number cannot describe a stochastic solver.

Sections: The revised set · What was set aside, and why · What Fable 5.2 found · Still open · How to read the comparison · The sweep record on the old 69

Stock-market research

26-Week Drawdown Prediction: Findings

A negative result: a 149-feature weekly dataset holds no usable signal for predicting a 26-week drawdown in an individual stock. A rule scan, gradient boosting and a Bayesian posterior all fall below the detection floor; the variation belongs to market regime, and that could not be forecast either.

Sections: Conclusion · Evidence: three methods, all below the detection floor · Why more modelling will not fix it · Why the data cannot support the question · What is feasible instead · Method and reproduction

Column Shortlist Review9 Oct 2026

Review of a 92-column feature shortlist for a return model: keep the redundancy pruning and the cross-sectional target, cut the hindsight target, restore five columns dropped as dead, and treat the 2017 to 2026 scoring window as the main risk. Standard errors show the tiers are resolvable but the ranks inside them are not.

Sections: Keep unchanged · Cut ret_best_hz from the target set · Restore the five columns dropped as dead · The regime mismatch is the build's main risk · On 2 sigma to 3 sigma · Actions before the build · What the standard errors say

Proving a Timing Edge

How much evidence a market-timing edge needs. Seven years of out-of-sample data could not have detected an edge smaller than +0.054 AUC, and the edge measured was +0.048; after correcting for a search over 149 factors the honest bar is 3.59 standard errors. How long a blind test must run, and why more stocks cannot replace more time.

Sections: The number to lead with · What was measured, and where the two tests disagree · What a sample of a given length can see · How long a blind test has to run · Why more stocks cannot buy what more time can · What the data he is preparing is worth · What to propose, and what to ask first

Screen and Size

What survived full verification: one axis, distance below the one-year high, that decides what is eligible to hold and what to avoid, and two ways to size a position. Neither forecasts returns for an individual stock; that capability was tested for and found absent.

Sections: The drawdown axis · Volatility-character forecast · Persistence map · One weekly cycle

Word ladders

Nine-Letter Word Ladders13 Nov 2026

Rex Gooch's rules for an ideal nine-letter word ladder, run over modern word lists, his own dictionary file and a 2.6-million-string corpus. Familiar words give nothing; with a few judged words there are four ideal ladders and three near-ideal ladders between heterograms beyond his, but none of listed words that changes its letters in order. At eight letters one does, WHITTLES to SPARKING, on two obsolete OED words; at nine, JUDGELIKE to BARBARYNS does, on three coinages and four old spellings; an exact search over 35 million rule-made words, compounds and corpus strings finds nothing better.

Sections: Gooch's rules · The search · What each word list gives · The new ladders · The words they rest on · What the corpus could not give · Shorter words, for scale · Letters changed in order · Checked against Gooch's printed tables · Limits · Sources and method

High-range testing

An Adaptive Reasoning Test13 Nov 2026

An adaptive two-parameter IRT test at questrel.com/oracle/irt/, run as a CGI program on shared hosting. Items are drafted offline by Gemini and checked by a blind solve; a model checking itself passed about one item in five that Claude, as an independent solver, rejects as unsound. Why generating items on demand fails, how the bank grows weekly on $10 a month toward the abilities of real candidates, why the score is not yet a normed IQ, and why no unproctored test of this kind can reach the one-in-a-million level.

Sections: The starting point · Why it is a CGI program · How a sitting works · Where the items come from · Which model should check · Why not generate each item on demand? · Growing the bank on a budget · Scores, norms and the scores candidates report · Can a test like this reach the Mega level? · Where it stands

Probability and quantum information

The Patch and the Qubit13 Nov 2026

An urn holds a ball that is white or black, and nothing more is known; the probability that it is black is not a number but a patch, the density a delta(x) + b delta(x - 1). Ed Thorp suggested the name. The question: can patches be reduced to qubits, can qubits be reduced to patches, or does quantum mechanics require something new? As the urn gives the patch, an urn that opens along any direction but holds one bit gives the qubit. Patches reduce to qubits, as a qubit seen in one basis with its phase removed. Qubits do not reduce to patches: the Kochen-Specker and Bell theorems rule out bits decided in advance. What is new is the arithmetic of the weights: amplitudes that add before squaring, or real weights that may go negative.

Sections: 1. The patch · 2. The patch and the coin · 3. All classical ignorance is made of patches · 4. The urn that gives the qubit · 5. Can patches be reduced to qubits? Yes · 6. Can qubits be reduced to patches? No · 7. What quantum mechanics adds · 8. The answer · Bibliography

Index of topics

A

adaptive testing — An Adaptive Reasoning Test
AI cheating
proctoring — An Adaptive Reasoning Test
airvent — Airvent; Compound Words on the Web
amitular — Word Birth
AUC — 26-Week Drawdown Prediction: Findings; Column Shortlist Review; Proving a Timing Edge
avuncular, feminine of — Word Birth

B

Bell's theorem — The Patch and the Qubit
benchmarks
comparing sweeps — Puzzle Solve Ledger
blind tests
length needed — Proving a Timing Edge
Bloch ball — The Patch and the Qubit
border words — Border Words of the World; City Hops; The Cantata Traverse
Borgmann search
in the 9×9 square — 9x9 Column Proposals; Nine by Nine
in word ladders — Nine-Letter Word Ladders
proposed words — Borgmann Word Log
the method — The Borgmann Search
Borgmann, Dmitri — The Borgmann Search
bottlenecks in exhaustive search — The Borgmann Search
Broda, Peter — Quintuple Stack
budgets
API spend — An Adaptive Reasoning Test

C

cantata — The Cantata Traverse
capital cities — Border Words of the World
CGI on shared hosting — An Adaptive Reasoning Test
Chaikin, Eric — Compound Words on the Web
grades — Borgmann Word Log; Frequency Data vs Word Grades; Stack Search Log
hand-built grid — Symmetric Quintuple Stacks
Claude
as answer-key checker — An Adaptive Reasoning Test
coinages — The Borgmann Search; Word Birth
in -like — Nine-Letter Word Ladders
coins
known bias — The Patch and the Qubit
Common Crawl — Compound Words on the Web
compounds
as candidate words — Nine-Letter Word Ladders
correlation — Frequency Data vs Word Grades
crossword grids
finished — Quintuple Stack
symmetry — Stack Search Log; Symmetric Quintuple Stacks
the 50 standard — Stack Search Log
crossword stacks
quintuple — Quintuple Stack; Stack Search Log; Symmetric Quintuple Stacks; The Borgmann Search
search log — Stack Search Log
crossword word lists
scores — Frequency Data vs Word Grades; Quintuple Stack

D

de Finetti's theorem — The Patch and the Qubit
dead ends — Borgmann Word Log
dephasing — The Patch and the Qubit
detection floor — 26-Week Drawdown Prediction: Findings; Proving a Timing Edge
dictionaries
omissions — Airvent; Compound Words on the Web; Word Birth

E

entanglement — The Patch and the Qubit
entropy
of the first bit — The Patch and the Qubit
Evans, Bergen — Word Birth
exhaustive search
negative results — Nine by Nine; Symmetric Quintuple Stacks

F

fair entries — Symmetric Quintuple Stacks
frequency database — Borgmann Word Log; Frequency Data vs Word Grades

G

genetic programming — Puzzle Solve Ledger
ghost words — Word Birth
Gleason's theorem — The Patch and the Qubit
Gooch, Rex — Nine-Letter Word Ladders
Google Books Ngrams — Airvent; What the Scrabble Lists Leave Out
misprints as false words — Nine-Letter Word Ladders
gradient boosting — 26-Week Drawdown Prediction: Findings

H

heterograms — Nine-Letter Word Ladders
hidden variables — The Patch and the Qubit
high-range tests — An Adaptive Reasoning Test
hindsight targets — Column Shortlist Review
hyphenation — Compound Words on the Web

I

information coefficient — Column Shortlist Review
initials — Border Words of the World; City Hops; The Cantata Traverse
item response theory
2PL — An Adaptive Reasoning Test

K

Kelly betting — The Patch and the Qubit
Kochen-Specker theorem — The Patch and the Qubit

L

language models
as item writers — An Adaptive Reasoning Test
as judges of answer keys — An Adaptive Reasoning Test
as judges of words — 9x9 Column Proposals; Nine-Letter Word Ladders; The Borgmann Search
on rare words — Word Birth
proposing rows — Stack Search Log
solving sequence puzzles — Puzzle Solve Ledger
lexicons
curated and judged tiers — Nine by Nine
lipograms — Pangrammatic and Panalphabetic Windows
longest words
on maps — Border Words of the World; City Hops; The Cantata Traverse

M

map word games
US states — The Cantata Traverse
cities — City Hops
countries — Border Words of the World
market regime — 26-Week Drawdown Prediction: Findings; Column Shortlist Review
Mega level
limits of measurement — An Adaptive Reasoning Test
Mistral — 9x9 Column Proposals
multiple testing
search over 149 factors — Proving a Timing Edge

N

negative probability — The Patch and the Qubit
negative results — 26-Week Drawdown Prediction: Findings; Screen and Size
norming
self-reported scores — An Adaptive Reasoning Test

P

panalphabetic windows — Pangrammatic and Panalphabetic Windows
pangrammatic windows — Pangrammatic and Panalphabetic Windows
partial squares — 9x9 Column Proposals
patch
definition — The Patch and the Qubit
reduction to and from qubits — The Patch and the Qubit
phyllostachys — City Hops
probability
ignorance not expressible as a number — The Patch and the Qubit
of a short window — Pangrammatic and Panalphabetic Windows
program verification — Nine by Nine
proper nouns — The Cantata Traverse
puzzles
solve rates — Puzzle Solve Ledger

Q

Q score — Stack Search Log
quantum theory
reconstructions — The Patch and the Qubit
qubits
derived from one-bit urn — The Patch and the Qubit
repeated measurement — The Patch and the Qubit
qutrits — The Patch and the Qubit

R

Roman numerals — Pangrammatic and Panalphabetic Windows
Rosner sequences — Puzzle Solve Ledger

S

Scrabble
official word lists — What the Scrabble Lists Leave Out
Scrabble lists
omissions — Compound Words on the Web
selection bias — Column Shortlist Review
solid and open compounds — Airvent; Compound Words on the Web
sparkin — Nine-Letter Word Ladders
standard errors
of AUC and IC — Column Shortlist Review
statistical power — Proving a Timing Edge
Stern-Gerlach experiment — The Patch and the Qubit
stochastic search
run-to-run variation — Puzzle Solve Ledger
stocks
26-week drawdown prediction — 26-Week Drawdown Prediction: Findings
drawdown from the one-year high — Screen and Size
feature selection — Column Shortlist Review
position sizing — Screen and Size
screening — Screen and Size
timing edge — Proving a Timing Edge

T

Thorp, Ed — The Patch and the Qubit

U

unlisted words — Airvent; Borgmann Word Log; Compound Words on the Web; Nine-Letter Word Ladders; What the Scrabble Lists Leave Out
urn problems — The Patch and the Qubit
usage
as the test of a word — Airvent; Compound Words on the Web; What the Scrabble Lists Leave Out

V

volatility forecast — Screen and Size

W

walk-forward testing — 26-Week Drawdown Prediction: Findings; Column Shortlist Review; Proving a Timing Edge
Wigner function
discrete — The Patch and the Qubit
Wiktionary — Nine-Letter Word Ladders
windows in text
records — Pangrammatic and Panalphabetic Windows
word grades — Frequency Data vs Word Grades
word ladders
eight-letter — Nine-Letter Word Ladders
ideal and near-ideal — Nine-Letter Word Ladders
nine-letter — Nine-Letter Word Ladders
sequential — Nine-Letter Word Ladders
word lists
doubtful entries — Word Birth
gaps in — What the Scrabble Lists Leave Out; The Borgmann Search
word splitting — Compound Words on the Web
word squares
8×8 — Nine by Nine
9×9 — Borgmann Word Log; 9x9 Column Proposals; Nine by Nine; The Borgmann Search
asymmetric (double) — Nine by Nine
Word Ways — Nine-Letter Word Ladders; Pangrammatic and Panalphabetic Windows
words removed from lists — What the Scrabble Lists Leave Out

X

XWord Info — Quintuple Stack