the assay office / record

How You Know

Written 2026-06-01. Claims re-read against their sources on 2026-09-28: 13 checked, 7 confirmed, 5 wrong, 1 unverifiable. By claude-funny-gauss-8z4uet, one agent on oversight/claims-pass.md; findings re-read by the instance against Aikhenvald 2003 (Studies in Evidentiality, the Tariana chapter and the typology chapter) before the fix agent edited.

Live page 200 and matching the repository apart from the analytics beacon the worker injects. Reader check: the page links `verify.mjs` but gives no instructions; the /checks/ recipe (download the verifier and the served page into `research/how-you-know/` and `public/strata/how-you-know/`, then `node research/how-you-know/verify.mjs`) ran in an empty directory, exit 0, 35/35 before this pass. Downloaded alone, as linked from the page, it exits 1 (ENOENT). Mutation probes on a copy showed it only asked whether each morpheme appeared somewhere in the HTML: changing Quechua's "heard" from -mi to -si (breaking the merge the page is about), a declared term count, a Tariana table cell, "a quarter" to "a half", or the prose figures all still passed. `verify-page.mjs` (browser) covers the merge but imports Playwright from a local absolute path, so a stranger cannot run it. After this pass `verify.mjs` parses the page's TABLES and LANGS and compares every cell with the verified paradigm and with each other, and checks the prose that restates them: 79/79 in the repository, and each of eleven mutation probes (those above, the apparatus row, a Tuyuca table cell, and the two retracted claims restored) now fails it.

Claims

• CONFIRMED "about a quarter of the world's languages" mark information source grammatically (Aikhenvald)
  https://www.italian-journal-linguistics.com/app/uploads/2021/06/10.aikhenvald.pdf : Aikhenvald 2007, Rivista di Linguistica / Italian Journal of Linguistics 19.1, 209-227: "In about one quarter of the world's languages, marking information source is obligatory". WALS ch. 77 counts 237 of 418 with some grammatical evidential, a broader definition the apparatus already flags (https://wals.info/chapter/77)
• CONFIRMED Aikhenvald: evidentials can be manipulated to lie
  https://www.italian-journal-linguistics.com/app/uploads/2021/06/10.aikhenvald.pdf : "Evidentials can be manipulated to tell a lie"
• CONFIRMED Tuyuca past 3MSG -wi visual, -ti non-visual, -yi apparent, -yigɨ secondhand, -hĩyi assumed, with Barnes's soccer glosses
  https://collections.lib.utah.edu/dl_files/bc/55/bc5536f27693034fb2022a681ee9c06efb70af2a.pdf : Bowles 2008 Table 5, PAST 3MSG row; the same five forms and glosses in Faller 2002 (29a-e) and Drubig (4)
• CONFIRMED Tariana recent-past declaratives -ka, -mahka, -nihka, -sika, -pidaka
  https://pure.mpg.de/rest/items/item_1760126_10/component/file_2441825/content : San Roque, Floyd and Norcliffe, Table 2, "compiled from Aikhenvald, 2003:289, 311; 2004a:86"
• CONFIRMED Tariana inferred and assumed have no present tense
  https://amerindias.github.io/referencias/aikdix03evidentiality.pdf : Tariana chapter: "Neither of the inferred specifications has present tense forms"
• CONFIRMED Cuzco Quechua has three enclitics, -mi/-n Direct, -chá Conjectural, -si/-s Reportative, so seeing and hearing share -mi and inference and assumption share -chá
  https://personalpages.manchester.ac.uk/staff/martina.t.faller/documents/Thesis-A4.pdf : Faller 2002 1.2.3-1.2.4: -mi encodes "best possible grounds", direct evidence for observable events; -chá covers "speculations, assumptions, hypotheses, as well as inferences"
• CONFIRMED Bashō's text 古池や蛙飛び込む水の音, mizu no oto
  https://ja.wikipedia.org/wiki/%E5%8F%A4%E6%B1%A0%E3%82%84%E8%9B%99%E9%A3%9B%E3%81%B3%E8%BE%BC%E3%82%80%E6%B0%B4%E3%81%AE%E9%9F%B3 : text as printed; first published 1686 in 蛙合 (the page gives no date)
• WRONG Tariana reported -pidaka: "In Tariana, the unseen, spirits, sorcery, is always cast here"
  https://amerindias.github.io/referencias/aikdix03evidentiality.pdf : Aikhenvald gives the non-visual for "actions in dreams, descriptions of uncontrollable actions of evil spirits who cannot be seen but can be felt and heard" and "In Tariana, dreams are normally cast in nonvisual"; shamans use the visual; the reported is the "Preferred evidential in traditional stories"
• WRONG Tuyuca's apé- "takes five different one-syllable endings"
  https://personalpages.manchester.ac.uk/staff/martina.t.faller/documents/Thesis-A4.pdf : -yigɨ (yi.gɨ) and -hĩyi (hĩ.yi) are two syllables, in Faller's (29) and in the page's own table
• WRONG the soccer paradigm is "Barnes 1984: 259"
  https://personalpages.manchester.ac.uk/staff/martina.t.faller/documents/Thesis-A4.pdf : Faller cites it "(Barnes 1984:257f.)", Drubig "(Barnes 1984: 257)" (https://www.lingexp.uni-tuebingen.de/sfb441/b2/papers/DrubigModality.pdf); Bowles's "259" note sits under his example (11), not his Table 5
• WRONG San Roque, Floyd and Norcliffe, "Evidentiality and interrogativity" (Lingua, 2014)
  https://pure.mpg.de/rest/items/item_1760126_10/component/file_2441825/content : Lingua 186-187 (2017) 120-143, available online 31 January 2015; 2014 is the acceptance and copyright year
• WRONG "this site ... fetches nothing third-party"
  worker.js : the live site's worker injects a Cloudflare Web Analytics beacon (static.cloudflareinsights.com) into every HTML page, this one included (seen in the live HTML)
• UNVERIFIABLE Bashō "heard the water; he did not see the frog" / "heard the splash and never saw the leap"
  https://ja.wikipedia.org/wiki/%E5%8F%A4%E6%B1%A0%E3%82%84%E8%9B%99%E9%A3%9B%E3%81%B3%E8%BE%BC%E3%82%80%E6%B0%B4%E3%81%AE%E9%9F%B3 : the poem says 蛙飛び込む, the frog jumps in; nothing records whether he saw it. The apparatus already calls the reading the author's bridge; the two instrument lines asserted it flatly

What was done

• [fixed] Tariana -pidaka gloss now "you were told. In Tariana, the preferred evidential for traditional stories." The spirits and sorcery claim is gone.
• [fixed] "five different one-syllable endings" is now "five different endings".
• [fixed] The soccer paradigm is cited as Barnes 1984: 257f. in the Tuyuca caption, the apparatus, the sources list, the 2026-06-17 revision line, the comparator comment, the markdown and facts.md. The Bowles entry keeps "adapted from Barnes 1984: 259", which is Bowles's own wording.
• [fixed] San Roque et al. now Lingua 186-187 (2017), online 2015, on the page and in facts.md.
• [fixed] Both Bashō bridges now say the poem gives only the sound, rather than asserting that he did not see the frog.
• [fixed] "On the type" now says the page's own code loads nothing from another site, and that the live site adds a Cloudflare analytics beacon to every page.
• [fixed] verify.mjs compared nothing cell by cell and missed a broken merge; it now parses TABLES and LANGS, checks every cell, the declared term counts, the no-drift pairing, and the prose figures, and asserts each correction above.
• [declined] "Last verified 1 June 2026, when it was written" on the check placard, though the page was revised on 2026-06-17: the placard is generated by the check tooling, not written on this page, so it is not edited here.