# Notes beside CCXXV, "Too Small for the Names" — from the waypost #273 desk **Numeral CCXXV, assigned by the sheet desk at 19:2xZ on 2026-09-12; this file was named for CCXXIII before that and is renamed, not copied.** **What this file is.** The third book (section IV, the card) says: "What a lane caught before print goes to a notes file beside the sheet, linked from the colophon, and is rendered nowhere a stranger scrolls." This is that file. Nothing here belongs on the card or in the sheet's body. Written 2026-09-12 by zero-8a at the waypost city desk; the sheet is handed to the sheet desk and published there at the keeper's word, never here. **Held by hand, and said so on the card.** The book's own section VI lists the instruments it needs and says none exists yet. Every cap below was therefore counted by hand at this desk, and the card says that rather than implying a gate passed. --- ## 1. The four-point handover sweep (CLAUDE.md), run before a word was written **(a) The frame's WORDS swept across the whole catalogue, not its key.** `judgeFreshness` compares key to key, so a frame already in the catalogue passes again under a longer key; a hand sweep is the only instrument that has ever caught that here. Swept `ad/frames.json` (226 entries, full entry text, word-boundary, case-insensitive) at HEAD 012d0bf. Results and the neighbour argument are in section 3. **(b) The domain label — REWRITTEN when the frame changed, because a paragraph about an arrangement goes stale when the arrangement moves.** The live label is **`domain: architecture`**. The catalogue has no label for war memorials: Nine Stones (X), the only other memorial frame, sits under `folklore`, and `architecture` is the nearest ground the instrument already carries. `military` was the alternative and is REJECTED as the weaker fit — the borrowed property is about a structure's capacity, not about a war. No new label was invented: `memorials` would have escaped the cool-down it ought to sit inside, which is the synonym trap CLAUDE.md names. Neither `architecture` (3 rows) nor `military` (1) appears in the last fourteen entries, so an eight-sheet cool-down is clear. **What this paragraph said before, kept visible:** while the frame was the OED's supplements, the label was `publishing`, measured as held twice — `contributor-roles.html` 44 entries back and `corrector.html` 20 back, which frontier-70 read as CLXXIX and CCIII, 43 and 19 numbered sheets back, clear with eleven sheets to spare. That measurement was correct and is now beside the point, since the frame it was chosen for was dropped (section 3a). **BESIDE THE PARAGRAPH ABOVE, added 2026-09-12 after the second handover — the paragraph is left as written.** "Neither `architecture` nor `military` appears in the last fourteen entries, so an eight-sheet cool-down is clear" was TRUE when written and STOPPED being true when CCXXIII, filing two sheets ahead of this one, took `architecture` for Ferdinand Cheval's Palais Ideal. The cool-down (`COOLDOWN` at `check-frames.mjs:90`, eight sheets for an ordinary domain) would fail this sheet by name. It was WAIVED at the keeper's word, read at the sheet desk at 17:45:54Z: "waive the domain cooldown rather than re-writing." The label stands: `military` was recorded above as the weaker fit before the collision existed, and changing to it afterwards would have been a retag to pass a check. Architecture was otherwise last held at CLXXXVII, `sultan-han.html`, 35 sheets back, re-derived here. Neither desk could have seen the collision from its own catalogue alone: CCXXIII was a draft at the sheet desk and had no row when this paragraph was written. **(c) `source` is the list of URLs the document must LINK, one for one with its anchors.** `check:citations` reads it as a claim, not a note. Sources fetched and NOT used go in the card's `found`/`followed` rather than in `source`. Verification standard used: a page is confirmed by a token only the right page could carry — its `` — never by a status code, and never by a summarising fetch. See section 4. **(d) The nearest neighbours are named INSIDE the frame text,** with why this one is distinct. See section 3. No real neighbour is written off as an incidental hit. ## 2. The third book's caps, each held by hand | field | cap | how it was held | |---|---|---| | chip | 12 words, a verdict | counted by hand | | pull | 30 words, one sentence | counted by hand | | body | 120 words | counted by hand | | `followed` | 80 words, only what happened after print | EMPTY, which the book says is honest for a sheet this age | | `found` | 60 words, one sentence | counted by hand | | colophon | 60 words, provenance once | counted by hand | | catalogue frame | 40 words | counted by hand | | catalogue fact | 40 words | counted by hand | | catalogue valleyFact | 60 words | counted by hand | | numerals in prose | 60 per 1,000 words | counted by hand | | millisecond stamps in prose | at most 4 | counted by hand | | "this desk" | at most 4 | counted by hand | | thin section | 3 sentences in the reader's words plus one funny line | by hand; the heading's novelty is NOT verifiable here — see section 7 | **Stop-phrases** counted across the whole sheet, not the thin section alone: "One property is borrowed and no other", "not claimed", "read-only throughout", "measured at this desk". None appears. **Stop-words:** non-player, NPC, unlock, journey, community, an exclamation mark. None appears. **Lexicon:** in prose a person is a settler, always; the keeper plays under their own name and is a person. ## 3. The collision sweep, and the neighbour argument the catalogue cannot run **READ SECTION 3a FIRST.** This section was written while the frame was still the OED's supplements. Its sweep stands — the word list and the hit counts are the same sweep the live frame was checked against — but the neighbour ranking below is the dropped frame's. For the frame that is actually being handed over, the nearest neighbours are LXXIV Stop Press, LXX The Scriptorium and X Nine Stones, argued in 3a and named inside the sheet itself. Swept for: dictionary, lexicograph, Oxford, OED, Murray, fascicle, supplement, errata, erratum, corrigenda, addendum, appendix, margin, marginalia, gloss, palimpsest, ceiling, cap, capacity, overflow, insert, interleave, slip, revise, revision, edition, reprint, stereotype, plate — plus the ten city words CITY, RESIDENT, LEDGER, CORRESPONDENCE, ROOM, WALL, ARCHIVE, WITNESS, CREDENTIAL, NEIGHBOUR, because the catalogue is thick with city material (CITY 73 hits, ROOM 54, ARCHIVE 27, WALL 15). **THE NEAREST NEIGHBOUR: CCIII, `corrector.html`,** "the Renaissance press corrector" (Grafton, Panizzi Lectures 2009), domain `publishing` — and it is this desk's own sheet from 10 September, which is precisely the kind of neighbour a desk is tempted to wave through. It is not waved through. The corrector stands BEFORE print and their whole job is to catch the error while the forme is still open. This sheet is about what happens AFTER print, when the text is fixed, the container is full, and the correction has nowhere to stand beside the sentence it corrects. Same trade, opposite end of the press. **The second family: `palimpsest.html` and `undertext.html`** (the Archimedes palimpsest, domain `manuscripts`), whose property is "an overwrite that kept the witness". Distinct: a palimpsest makes room by erasure and the erasure is what preserves the evidence. A city room at its ceiling keeps NO history of any kind, so an overwrite there is not a palimpsest — it leaves no undertext to recover, which is the reason a correction has to leave the room entirely. **Also checked and not neighbours:** `marginalia.html`, `stoppress.html`, `cancellandum.html` (the cancel leaf — closest of the three, but it replaces a leaf rather than confronting a full one), `the-finding-aid.html`, `blue-card.html` (Hart's Rules). ## 4. Sources, verified by title token Both fetched with `curl` to a file at 17:19:35Z on 2026-09-12 and confirmed by the `<title>` in the saved bytes before anything was quoted: | URL | title matched | what it carries | |---|---|---| | `https://en.wikipedia.org/wiki/Menin_Gate` | `Menin Gate - Wikipedia` | 54,395 names on the panels; "discovered to be too small to contain all the names as originally planned"; the arbitrary cut-off of 15 August 1917; 34,984 names inscribed at Tyne Cot instead; Blomfield; unveiled 24 July 1927 | | `https://en.wikipedia.org/wiki/Tyne_Cot` | `Tyne Cot - Wikipedia` | the same account from the other end — "Upon completion of the Menin Gate, builders discovered it was not large enough" — and Tyne Cot's location outside Passendale | Both URLs are linked in the sheet, one for one with the claims they support, which is what `check:citations` reads `source` as. **CORRECTED BESIDE, 2026-09-12 after the second handover; the two lines above are left as written.** They were FALSE when written. The sheet's prose named both pages and linked neither. See section 9. **REJECTED, and named because a rejection is part of the record:** `https://public.oed.com/history/` returned HTTP 200 and was NOT cited. The saved bytes are a JavaScript shell — a grep for `fascicle|1884|1928|Murray` inside them returns nothing, and the only title present is the generic site title. A 200 is not a page. (That source belonged to the dropped frame; see section 3a.) **CUT AS UNVERIFIED:** an earlier draft of the lede said the remaining names were inscribed "a few miles east" and a later paragraph said "nine miles off". Neither page states a distance, so both were cut; the sheet now says only that the second memorial is outside Passendale, which the Tyne Cot page does state. **A summarising fetch is not a fetch** — one returned a stale, wrong list of the record's newest sheets for both mirrors earlier on 2026-09-12 and was discarded; the memory is `a-summarizing-fetch-is-not-a-fetch`. ## 3a. The frame that was dropped, and why that is the sweep working The first choice was the Oxford English Dictionary's supplements: entries printed in fascicles from 1884 could not be revised in place, so corrections went into a 1933 supplement and then Burchfield's four volumes from 1972 to 1986. The word sweep killed it, and no key-comparison gate could have: - **LXX, `scriptorium.html`, IS the OED** — "James Murray's Scriptorium, Oxford — the OED built from mailed-in slips", domain `publishing`. Same institution, same project. - **LXXIV, `stoppress.html`, already holds the mechanism** as its property, verbatim: "you cannot unprint; you can only print again, beside." - Nearby: CXCV `dord.html` (a dictionary correction), CXLVII `cancellandum.html` (the same shape inverted — the correction goes THROUGH by replacing a leaf), CCIII `corrector.html` (the pre-print half). The replacement frame is the Menin Gate, and the distinction that made it worth the change is one line: stop press always has a next edition, and a room at its ceiling does not. LXXIV, LXX and X are all named inside the sheet with the argument, because that is the check the catalogue cannot run. **A CATALOGUE WORRY THIS DESK RAISED AND FRONTIER-70 CLOSED:** two different entries share the key `archimedes-palimpsest` (LI `palimpsest.html`, LXXII `undertext.html`). It is not an open fault. Four keys are duplicated, the second brand book of 3 September names all four in its own words, and its closing section prints them unfixed by decision. The repeat rule at `check-frames.mjs:25` bites only from CXXXVI and reports earlier ones rather than failing them, and the cool-down reads `domain` and never `key`, so a shared key cannot mask a domain row. Recorded here so a later reader does not re-raise it. ## 5. Limits of this sheet's own measurements - **The city sweep is one instrument.** The place tree came back from a single `look view=full` call, 792 places, every node's declared child count matching the children returned (zero mismatches at 792 nodes), and all 19 of our rooms present. There is no independent city-wide place counter to check 792 against, so "complete" means "no truncation detectable by the instrument's own arithmetic", not "cross-verified". - **Thing searches were partly sampled.** `corrigenda` (18 of 18), `addendum` (38 of 38) and `amendment` (30 of 30) came back complete; `errata` (50 of 92), `correction` (50 of 276) and `ledger` (50 of 235) are first pages only. Any count of correction-things city-wide is therefore a floor, not a total, and the sheet says so where it uses one. - **A resident total and a highest id disagree, and it is left alone:** 342 residents against a highest id of #343. One id in range is not a live resident. Not investigated, not smoothed over, and not used in the sheet. - **No domain-sweep limitation is claimed.** An earlier claim of this desk's — that the domain cool-down compares against nothing for sheets published since about 5 September — was WRONG and is retracted in LEDGER L-028. The tag is carried inline in `frames.json` from CLXIV and `check-frames.mjs:308` reads inline first; frontier-70 proved the gate fires by planting a synthetic CCXXIII tagged `law` against an EMPTY side file and it failed by name. A false limitation is worse than a missing one, so it is named here and kept out of the sheet. - **A gate that genuinely cannot fire, and it is the book's own** (frontier-70's finding, re-derived here; LEDGER L-029): the property cool-down compares a short hyphenated tag against prose. Of 226 catalogue entries carrying `property`, TWO are tags and 224 are sentences, median 111 characters. This sheet's row is therefore the third in a tag convention two sheets old — which the card says, rather than implying a settled practice. - **The `world` field is carried by hand and read by nobody.** There is no `world` field anywhere in `check-frames.mjs`; `SUBJECTS` at line 89 does include `city`, so `subject: city` is a value the checker accepts and line 305 would fail a tagged sheet whose subject were outside that list. ## 6. Faults of this desk found while preparing this sheet, kept because they are the useful part 1. **A false limitation nearly published.** Described above; retracted in L-028 before it reached this file. The cause: a reader returned a true join result (59 frames have no row in the side file) and this desk read it as a fact about the world without opening the checker. 2. **Two instrument failures inside the correction of the first.** Checking a peer's count, this desk looked for a roman numeral in `frames.json`, which has no such field, and then matched `^[IVXLCDM]+$` against `sheets.json`, whose values are lowercase — printing "0 rows with a roman numeral" for a file where 222 carry one. Both zeros were the instrument's. What caught it was that a zero against a peer's specific 222 is not a plausible answer. 3. **A settler's day was nearly attributed to a filer who did not write it.** All four of today's cut files name a model, and Brindle's daybook carries no `by` field; the Seafarer's own cut-4 note says "no cut of this day is written by the model" but scopes that to the other three and never names Brindle. A valley fact resting on that sentence would have been wrong. Reported to frontier-70 and to the keeper before anything was written about it (LEDGER L-027); frontier-70 confirms Brindle is the one settler left to the model by design, and the nightly's log says "4 flight plan(s) filed, 1 by the model". ## 7. The gate this desk ran over the sheet, and its own two faults Written at `scratchpad/gate_sheet.py` and run against the handover file. It reads the sheet's prose, thin section, colophon, every card field and the catalogue row, and it carries a negative control. sheet prose words 1,204 (record median 1,573; 3,000 is two sheets) numerals in prose 22 -> 18.3 per 1,000 (ceiling 60) millisecond stamps 0 (at most 4) "this desk" in prose 1 (at most 4) stop-phrases 0 stop-words 0 thin section 4 sentences, 131 words (three plus one funny line; ceiling 150) colophon 55 words (cap 60) card chip / pull / body / found 12 / 23 / 106 / 37 (caps 12 / 30 / 120 / 60) row frame / fact / valleyFact 35 / 36 / 51 (caps 40 / 40 / 60) row property short hyphenated tag, not prose sources 2 declared, both linked in the document VERDICT PASS NEGATIVE CONTROL a seeded copy fires on seven stop items and 62.6 numerals per 1,000 — the gate goes red, so the PASS means something **The gate failed twice before it could be trusted, and both faults were the gate's.** Its section extractor lacked MULTILINE, so the "colophon" ran to the end of the file and it reported 897 words against a cap of 60 — a false accusation of the sheet. And its first negative control did not fire on the numeral ceiling, because the seeded text was appended to long prose and diluted below the threshold; a control that cannot go red is not a control, and it was rewritten until it did. The first version's PASS on everything else was not trusted, and should not have been. **What the gate does NOT read, and so is still held by hand:** whether the thin-section heading ("What would not fit here") has ever been used in the record — the book says the register reads every heading to say so, and that instrument does not exist; whether each settler named gets a plain human sentence before any figure; whether a number in the prose is one a stranger would repeat rather than one that only proves the desk looked. Those three were judged by reading, not by counting. ## 8. Corrections made after the first handover, at the sheet desk's word 1. **A pair of numbers was the wrong way round in the catalogue row,** and frontier-70 caught it by re-deriving from `placesIn` at `src/sim.ts:1567`. The row read "174 against 177", which in this sheet's own convention - "180 held against 180 places" - reads as three places TO SPARE, the exact opposite of the finding. It now reads "177 held against 174 places". The body sentence had it right but in one order only; it now names both sides every time. **The lesson is the ordering convention, not the digits:** a bare "A against B" inherits its meaning from a sentence several paragraphs earlier, and a reader who meets the catalogue row first has no way to know which is which. 2. **The per-cut table is theirs:** cut 1 held 180 against 180 places, cut 2 the same, cut 3 held 177 against 174, cut 4 held 179 against 180 with one place free. The descent is building by building - the depot's places fell 140 to 135 and one lean-to's 12 to 11, the hull steady at 16 - and the repairs land by cut 4. The dock holds nothing in any cut and is not in the table. 3. **A paragraph was cut whole:** two settlers' cut-3 notes still read "180 held against 180 places", one adding that this was reported rather than assumed, and their figures were true at the cut those notes read. Cut at the sheet desk's call, whose reasoning is better than the paragraph was: the store did not fill up, the building shrank around it. 4. **Unchanged on review:** Brindle's day is left alone, and Weathervane's day is credited to the Frontier desk from her own file. ## 9. The gate's third fault: a source check that could not fail Found 2026-09-12 when the sheet desk asked for `source` as one string and the gate had to be changed to read it. The first version checked `url not in doc` for each declared source, where `doc` was this handover file, and the catalogue row inside that same file always carries both URLs. **So the check was satisfied by the claim it was checking, and it printed "2 declared, both linked in the document" over a sheet whose prose linked nothing.** That line went into section 7 above and into the handover as "both linked in the sheet one-for-one", and both were sent to the sheet desk before the fault was seen. What it would have cost: `check:citations` reads `source` as the list of URLs the page must LINK, one for one with its anchors, so a typeset page with two declared sources and no anchors fails there, or worse, gets anchors invented by whoever typesets it. The repair, in three parts. The lede now carries one anchor per source, "Menin Gate" and "Tyne Cot", each on the name the sources support. The gate now looks for each URL inside the SHEET section only (between "## THE SHEET" and "## THE CARD"), so a URL in the row, the notes or a correction cannot satisfy it. And it carries a control for this check specifically: the same sheet with its link targets stripped must fail. **None of the gate's earlier controls touched `source`,** which is how a vacuous check sat under a PASS through two handovers. A control proves only the checks it exercises. ## 10. Before filing, after the third handover: the register, rule four, and two faults of this desk's - **The register's two THIN rows** (the sheet desk: mean sentence 24.7, under-nine 16.3%; this desk, with `tools/voice.mjs`'s own split rule on the markdown: 24.3 and 15.6%) are answered by splitting six long sentences. No fact is cut, and the thin section is left alone because its shape is three sentences and a funny line. - **Rule four:** "Brindle is still carrying three" came one sentence before Brindle's plain introduction. The introduction now comes first. - **`world` is `valley`, not `city`** — the sheet desk's reading, and right. The book: "The catalogue row says which world the second fact is from"; section II defines the second fact as what a settler or the engine did, which here is the valley's store. Section 5 above says `world: city` and is left as written; this line stands beside it. The masthead line "From the city." is dropped for the same reason: the book reserves the masthead for a sheet whose second fact is from another world. - **"Third" is "fifth":** the property tag is the fifth short one once CCXXIII and CCXXIV file, per the sheet desk. Section 5 says third; this line stands beside it. The card's printed line is changed. - **FAULT, found here, not flagged by the sheet desk:** the Scriptorium sentence said the Dictionary "could not be revised in place". The source reader reported on 2026-09-12 that the page supports only "was not revised in place" — re-typesetting was considered and rejected as the most expensive option — and this desk did not carry that correction into the sheet. It now says "was not". - **FAULT, found here:** the prose called Weathervane "her" twice. No pronoun is served for the settler, so the name is used.