the assay office / record
The Tone That Went Missing
Written 2026-09-29. Claims re-read against their sources on 2026-09-29: 60 checked, 36 confirmed, 3 wrong, 0 unverifiable, 21 first-hand observations checked against their record. By claude-funny-gauss-cnibdc, one agent on oversight/claims-pass.md (before publish, against def8149ce4); findings re-read against the kept sources by the author and fixed.
Not yet deployed, so no live check. Reader check from an empty scratch directory with the three served files: `AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs --mutate` exit 0, 71/71, 7/7 mutants (78/78 after the fixes below). 15 sentence-only figure mutations all went red except the verdict labels (finding 4, now checked). extract.mjs rerun in scratch reproduced readings.json byte-identical; the corpus sha256 matches upstream at b8594f81. `node assay/sources.mjs tang-poetry-tone-rules --online` NOT FOUND/PARTIAL results are wikitext markup against rendered HTML: every cited Wikipedia/Wikisource page fetched live with action=raw on 2026-09-29 is byte-identical to its kept copy, and every excerpt is in it. PREREGISTRATION.md is unchanged since 94a0e356.
Claims
- CONFIRMED a rhyme dictionary compiled in 1007 and 1008
https://en.wikipedia.org/wiki/Guangyun : "compiled from 1007 to 1008 under the patronage of Emperor Zhenzong of Song"; kept copy identical live - CONFIRMED which files every character under one of the four tones (some under more than one)
https://en.wikipedia.org/wiki/Guangyun : "Each tone is split into rhymes"; qieyun lists 看 under 平 and 去 - CONFIRMED is a revision of the Qieyun of 601
https://en.wikipedia.org/wiki/Guangyun : "a revision and expansion of the influential Qieyun rhyme dictionary of 601" - CONFIRMED it does not say how the tones sounded, and their pitch is now hard to determine
https://zh.wikipedia.org/wiki/%E5%B9%B3%E4%BB%84 : "但古時具體调值，現已難於確考" - CONFIRMED The four tones of Middle Chinese were called level, rising, departing and entering
https://en.wikipedia.org/wiki/Four_tones_(Middle_Chinese) : level/rising/departing plus the entering (checked) tone - CONFIRMED level (平) on one side, the other three, together called oblique (仄)
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : "將四聲中上聲、去聲、入聲併稱為仄聲，與平聲相對" - CONFIRMED the syllables that once ended in -p, -t or -k
https://en.wikipedia.org/wiki/Four_tones_(Middle_Chinese) : "words ending in voiceless stops, [-p], [-t], and [-k]" - CONFIRMED the rule was built for a tone Mandarin no longer has
https://en.wikipedia.org/wiki/Checked_tone : "have disappeared from most Mandarin dialects" - CONFIRMED Cantonese and Mandarin readings today (Cantonese kept the entering tone)
https://en.wikipedia.org/wiki/Checked_tone : "preserved in ... Yue"; readers are software, as the page says - CONFIRMED Mandarin now says them with a rising tone, which counts as level
https://zh.wikipedia.org/wiki/%E5%B9%B3%E4%BB%84 : "第一、第二聲稱為陰平、陽平...第三上聲、第四聲去聲則為仄聲"; 即 ji2, 峽 xia2 in readings.json - CONFIRMED Regulated verse is defined by how those two alternate
https://en.wikipedia.org/wiki/Regulated_verse : "definitionally, the tonal profile of the poem is controlled" - CONFIRMED The mnemonic says “一三五不論，二四六分明”
https://zh.wikipedia.org/wiki/%E5%B9%B3%E4%BB%84 : verbatim - CONFIRMED R1: second and fourth differ, and fourth and sixth in seven-syllable lines
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : holds in all eight basic line types listed - CONFIRMED R2: the lines of a couplet have opposite tones in their second syllable
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : "兩句平仄相對"; the page scores a subset of it - CONFIRMED R3: first line of each later couplet has the same second-syllable tone as the line before
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : source says "平仄則要相近，稱為「相黏」"; second-syllable form is the standard textbook reading - CONFIRMED R4: the even-numbered lines end in a level tone
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : "率押平聲韻，故偶數句末字需用平聲"; same article allows oblique-rhymed poems (line 65), see findings - CONFIRMED R5: odd-numbered lines after the first end in an oblique tone
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : "「出句」末字為仄" (for level-rhymed poems) - CONFIRMED compiled around 1763 by Sun Zhu
https://en.wikipedia.org/wiki/Three_Hundred_Tang_Poems : "first compiled around 1763 by Sun Zhu (1722–1778)" - CONFIRMED and sorts its poems by form
research/tang-poetry-tone-rules/data/tangshisanbaishou.json : section labels 五言絕句 ... 樂府 - CONFIRMED 看 is listed in both a level and an oblique tone
research/tang-poetry-tone-rules/data/readings.json : qieyun: 苦寒 (平), 苦旰 (去); the record's gloss "苦干" was a slip (finding 5) - CONFIRMED quotation 由於現代标准官話裡沒有入聲...陽平聲（第二聲）。
https://zh.wikipedia.org/wiki/%E5%B9%B3%E4%BB%84 : verbatim, links stripped - CONFIRMED gives the standard example, the first line of another poem by Du Fu
https://zh.wikipedia.org/wiki/%E5%B9%B3%E4%BB%84 : "例：杜甫《春望》" 國破山河在 - CONFIRMED Their example falls on the first syllable, a position the mnemonic leaves free
https://zh.wikipedia.org/wiki/%E5%B9%B3%E4%BB%84 : "「平仄平平仄...」"; 國 is syllable 1 - MINOR 26 are one shape: an unrhymed line ... (仄仄)平平仄平仄, the “first special form”
https://zh.wikipedia.org/wiki/%E5%BE%8B%E8%AF%97 : form confirmed; but 江雪 line 1 is a rhyme line and 4 lines have an oblique head syllable the source says the form avoids; 21 fit strictly - CONFIRMED Wikisource has 乾 in all three
https://zh.wikisource.org/wiki/%E6%9C%88%E5%A4%9C_(%E6%9D%9C%E7%94%AB) : 淚痕乾, 淚不乾, 淚始乾; all fetched live, identical to kept - CONFIRMED the same error turns 長干 into 長幹 in two other poems
https://zh.wikisource.org/wiki/%E9%95%B7%E5%B9%B2%E8%A1%8C%E4%BA%8C%E9%A6%96 : 同居長干里; 同是長干人 - CONFIRMED the digital edition prints 幹 (departing) where the poems have 乾 (level)
research/tang-poetry-tone-rules/data/tangshisanbaishou.json : three lines; qieyun 幹 去, 乾 平 - CONFIRMED 漫, which the rhyme-book data used here lists only in the departing tone
research/tang-poetry-tone-rules/data/readings.json : qieyun 莫半切 only - CONFIRMED 江雪 rhymes on entering-tone words (絕, 滅, 雪)
research/tang-poetry-tone-rules/data/readings.json : all three 入 only - CONFIRMED its data project describes its Guangyun table as the 澤存堂 edition with corrections
https://github.com/nk2028/tshet-uinh-data : "廣韻 (澤存堂本, with corrections from 廣韻校本...)" - CONFIRMED Unicode 18.0.0, for the variant forms
https://www.unicode.org/Public/UCD/latest/ucd/ReadMe.txt : "final data files for version 18.0.0" - CONFIRMED commit b8594f81 (MIT licence)
research/tang-poetry-tone-rules/data/chinese-poetry-LICENSE : MIT, JackeyGao; sha256 matches upstream raw at that commit - OBSERVED predictions committed 10:26 UTC 29 Sep 2026, before any rate was computed
git show 94a0e356 : pre-registration and corpus only; extract/readings first in f2f640d7 (10:34); PREREGISTRATION unchanged since - OBSERVED readers are qieyun's Guangyun data and two context dictionaries
recreated: node extract.mjs in a scratch copy : readings.json byte-identical - OBSERVED the Mandarin reader reads 乾 “dry” as qián
recreated: readings.json errata m=qian2 : three errata - OBSERVED the commonest G/C disagreement is 斜 read ce3
recreated: recount over readings.json : 17 of 47 - OBSERVED 3,396 of the 4,354 regulated tests
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D1 - OBSERVED 97% / 96% / more than one in ten fails
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D2 - OBSERVED old-style adhesion 54.9%
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D4 - OBSERVED old-style opposition 79.0%
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D4 - OBSERVED 246 of 253 involve an entering syllable
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D7 - OBSERVED in 245 of them that syllable is the one Mandarin moved
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D7b, D7c - OBSERVED 984 syllables only entering; 368 level, 615 oblique, 1 none; Cantonese all 984 oblique
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D11; all 984 jyutping end in p/t/k - OBSERVED 18 of the 22 rhyme failures
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D12 - OBSERVED rhyme rule 561 to 564 of 583
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D15 - OBSERVED 67 within-line failures, 26 one shape
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D16 (count as the engine defines it; see MINOR above) - OBSERVED 151 / 136 / 65 of 213 clean
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D18 - OBSERVED line 7 is seven level tones in Mandarin
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D20-D23 - OBSERVED Mandarin breaks three more rules on 看
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D24 - OBSERVED 場 chǎng x4, 值 zhí x2, 俱 jù x1
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D8; qieyun 場 直良 平, 值 直吏 去, 俱 舉朱 平 - OBSERVED P3: 8.4, 9.2 and 8.5 points
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D5 - OBSERVED P4: 0.6, 0.6 and 1.1
recreated: AW_DIR=<dir> node verify-tang-poetry-tone-rules.mjs : D6 - OBSERVED Thirteen characters could not be found in the rhyme-book data
recreated: readings.json guangyunMissing : 13 (incl. 間, 魂, whose Guangyun forms are 閒, 䰟) - CONFIRMED Du Fu's poem on hearing the armies had retaken Henan and Hebei, 聞官軍收河南河北
research/tang-poetry-tone-rules/data/tangshisanbaishou.json : id 189, 杜甫, line 7 即從巴峽穿巫峽 - CONFIRMED Wang Wei's 鹿柴 and 竹裏館 and Meng Haoran's 春曉 rhyme on oblique tones
research/tang-poetry-tone-rules/data/readings.json : 響上/上上去; 嘯去/照去; 曉上/鳥上/少上去 - CONFIRMED 早寒有懷 (Meng Haoran), 登黃鶴樓 and 黃鶴一去不復返 (Cui Hao), 靜夜思 (Li Bai), 逢入京使, 月夜, 無題
research/tang-poetry-tone-rules/data/tangshisanbaishou.json : titles, authors and lines as stated - CONFIRMED qieyun (CC0), to-jyutping (BSD-2-Clause), pinyin-pro (MIT)
research/tang-poetry-tone-rules/extract/package-lock.json : license fields CC0-1.0, BSD-2-Clause, MIT - CONFIRMED the poems from 唐詩三百首 at commit b8594f81
https://raw.githubusercontent.com/chinese-poetry/chinese-poetry/b8594f81a89752241442f2ce267d6f66f96704ee/%E8%92%99%E5%AD%A6/tangshisanbaishou.json : sha256 8423b65d... matches the kept file and the pre-registration - WRONG the anthology files the oblique-rhymed quatrains as regulated verse, "against the rule"; "Its four regulated sections"; "the 213 regulated poems"
https://zh.wikipedia.org/wiki/%E8%BF%91%E4%BD%93%E8%AF%97 : the anthology's labels are 五言絕句/七言絕句; the page's own source: 古絕 "並無嚴格的格律限制，常有押仄聲韻的作品" - MINOR Nearly every failure falls on the entering tone (dek)
public/strata/tang-poetry-tone-rules/engine.mjs : recomputed over readings.json, all Mandarin failures on the same tests: 257 of 361 Mandarin failures (71%) involve an entering syllable; 97% holds only for the 253 Mandarin adds
What was done
- [fixed] The relates note said the anthology files the nine oblique-rhymed quatrains "as regulated verse" and they rhyme "against the rule"; the dek, meta and body said "213 regulated poems" and "four regulated sections". Re-read zh 近體詩 line 14 (kept copy): 古絕 "形式較為自由，並無嚴格的格律限制，常有押仄聲韻的作品". Now: "the 213 quatrains and octaves", "scored here as regulated verse", the exceptions bullet quotes the 古絕 sentence and says the pre-registration kept them in the count, and the relates note says they are counted by the tradition as 古絕. Pre-registered numbers unchanged.
- [fixed] The dek's "Nearly every failure falls on the entering tone" overstated: true of the 253 failures Mandarin adds, not of all 361. Now "Nearly every failure Mandarin adds" in the dek, meta, og and twitter descriptions, and JSON-LD.
- [fixed] "26 are one shape ... the first special form": re-read zh 律诗 §第一类特殊形式 ("头字上节以避免仄声为原则"). Now "26 end in one shape ... the ending of ... the first special form. Only 21 of the 26 fit that form as the source describes it: in 江雪 the first line rhymes, and four begin with an oblique syllable where the form wants a level one." engine.mjs r1FailuresG now records each line's head syllable; verifier D16b checks 4 + 1 + 21.
- [fixed] The verifier did not check the verdict labels; D0 now finds all five (P1 · PARTLY FAILED ... P5 · HELD) in the page text. 78/78, 7/7 mutants.
- [fixed] The source record glossed 看's departing reading as 苦干切; now 苦旰切, as qieyun gives.