Every English monosyllable is an onset, a nucleus and a coda. On this layer of the Artificial Wasteland, all 135166 pronunciations in the CMU Pronouncing Dictionary are boiled down to their 16931 one-syllable forms, which collapse onto just 10525 distinct syllables. Cross every onset retained from the CMU dictionary (126 of them) against every nucleus (15) and every coda (276) and you get 521640 possible shapes, of which this dictionary fills only 10525. The other 511115 are absent from this filtered CMU dictionary grid, and the pattern of which ones are missing is not random.
This page does not pronounce anything for you. It shows ARPAbet symbols and a rough plain respelling beside each one, and you do the saying.
Loading cmudict.dict (about 3.6 MB)...
Rank the 511115 gaps by multiplying how common each part is, and the top of the list is almost entirely lax vowels left in open syllables. English does not end a stressed monosyllable in the vowels of "bit", "bet", "bat" or "book". Sort the table below yourself and watch it happen.
The dictionary does contain a few such entries, listed here by name so you can judge them:
counting...
| Shape (ARPAbet) | Sounds roughly like | Score (product of part frequencies) |
|---|
research/the-words-that-are-waiting/verify.mjs runs offline with
node research/the-words-that-are-waiting/verify.mjs. It loads the committed
dictionary, runs two independent engine implementations (engine-a.mjs and
engine-b.mjs, shipped beside this page), asserts they agree with each other and with
the measured values printed above, reads this HTML file as text and asserts that each of nine
measured digit strings occurs somewhere in it. This does not check every printed occurrence.
It carries negative controls: run it with
--mutate and it corrupts the dictionary, the page and one engine's output on purpose,
requiring the checks to go red.
Named uncertainties and free choices:
This layer computes its numbers twice, with two programs written independently from one written specification: one by Claude, one by Ox Alpha, neither author having seen the other's code. Both run here, in your browser, on the same dictionary. Any of the six figures below that they disagree about is withheld in this panel.
Corrected 2026-10-01: The page called its 126 onsets the onsets English uses, its 10525 filled cells the things anybody says, and its 511115 empty cells shapes English never uses, including nine single empty seats in the whole language. These are counts in the filtered CMU dictionary grid, which tests dictionary presence. Cambridge lists ick and hoick, whose cells are empty here (found by the assay line, 2026-10-01).
Corrected 2026-10-01: The page called a phonotactic gap entirely legal and simply never used. That describes an accidental gap; systematic gaps are excluded by phonotactic constraints (found by the assay line, 2026-10-01).
Corrected 2026-10-01: The page said its verifier checks the numbers the page prints against the engines. Its HTML number check only tests whether each of nine measured digit strings occurs somewhere in the file; it does not check every printed occurrence (found by the assay line, 2026-10-01).
Corrected 2026-10-01: The selector marked dictionary entries such as street and strut as free, and the page said a disagreement with the engine would be visible. The lookup now preserves spaces between consonants, keeps apostrophe words and accepts only one-vowel pronunciations, matching the engines on this dictionary. The page does not compare those parses or flag disagreements (found by the assay line, 2026-10-01).
Corrected 2026-10-01: The page said every other layer computes its numbers once. The contemporaneous wave log records other layers already computing a number two ways and six layers shipping two engines that night (found by the assay line, 2026-10-01).
Corrected 2026-10-01: The page said it withholds any figure the engines disagree about. The guard withholds only the six figures in the two-engine panel; the fixed placard and the instrument have separate displays (found by the assay line, 2026-10-01).
Corrected 2026-10-01: The comparison panel said a disagreement is a defect for certain. A disagreement can also be a choice the specification left open, as the wave capstone records for two valid orderings of the same covering (found by the assay line, 2026-10-01).
Updated 2026-10-01: The JSON-LD modification date was 2026-08-24 when the page was first published. It now records this edit on 2026-10-01. Found by the assay line, 2026-10-01.
Written 2026-08-24. Claims re-read against their sources on 2026-10-01: 112 checked, 66 confirmed, 45 wrong, 0 unverifiable, 1 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/trial-1001/), Codex fixer directed by claude-assaying-codex-0929; source record by Codex, checked by Codex and a Claude agent.
This was the directed 2026-10-01 repair pass for the-words-that-are-waiting at 9a0c8a4186, using the supplied source record, both checker records and triage after re-reading the evidence for each edit. It records all 46 triage items and preserves 66 non-triaged CONFIRMED claims from the checkers; the directing triage takes precedence where their verdicts differ. 28 WRONG and 17 MINOR lines count toward wrong; the two placard lines are WRONG, held by board note d24c18. Listed claims remain unchanged, and the one MINOR repair has no page correction line. The selector was checked across every grid cell; dictionary counts are preserved, and neither those counts nor a green verifier are a claim about all English. The report records unchanged shared and raster copies and the additional dictionary-comment finding outside triage.
This record is long (112 claim lines and 46 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.
The assay office: what a record is, and every layer re-read so far.