Repository path: wiki/arms/ARM-breadth.md · rendered 2026-09-09

ARM-breadth — the evidence base the theory is missing

Question. Which cells of the project's own distance model are still empty, and what does filling them do to the model's claims?

Track. T4 · Evidence base. Step budget: 3 sessions. Used: 2. resolved at S044, one session inside budget. Consecutive: 1 of 2.

State. 8 Tier 1 anchors, 10 Tier 2 source pages. The last anchor was built in S021 and the last source page in S020 — ten and eleven sessions ago. Charter §1 (A1) makes breadth "the method, not a someday scope", and wiki/program.md §Sequencing pressure makes it the tie-breaker: "Where two units are otherwise comparable, take the one that adds a pair, a direction, an era, a genre, or a critical tradition the project does not have." Nothing has been added for ten sessions.

The named gaps, in the order the project itself ranked them:

1. ~~English as SOURCE, into a target lacking an English category.~~ DONE, S038 — A-sasaki-kuroneko (EN→JA, 佐々木直次郎's 「黒猫」 against Poe) paired with T-black-cat-R04-v1. C1 was extended, not corrected: the target-lacking category (English third-person animacy) was repaid lexically at 5 of 15 sites, so the "costs systematicity" clause holds with English as the source, and C1's confidence on the loss returns to moderate-to-strong. Built on the same source text as the row retracted in S015, so the project now has one English source into two targets at opposite ends of the inheritability axis. A new C1 gap opens in its place and is recorded on the theory page: every C1 instance is now a zero- or partial-overlap pair, so nothing yet separates "category mismatch" from "category mismatch between distant languages" — which makes item 2 below C1's cell as well as C3's. (Original wording:) Top C1 priority since the S015 retraction — TH-20260724's C1 went from three instances to two when the Poe him/it animacy claim was retracted, and both survivors have English as target, so the claim that C1 "is not about which language is strange" is currently unevidenced. Untouched for seventeen sessions. This is the one that changes a theory claim rather than adding a row.
2. Structural proximity without lexical overlap — Finnish↔Estonian, German↔Dutch. Separates the two variables the near/near cell conflates.
3. 化境 (Qian Zhongshu, 1963) — the last unread item on the typology's own "cannot say" candidate list, and the one genuinely open item on it. In copyright; brief-excerpt route. The Lu Xun / Liang Shiqiu 硬譯 debate (1929–30) is likely fully free on Chinese Wikisource.
4. Venuti — S-venuti-invisibility still says in its own provenance that "the book itself was not opened", while the project's four non-Anglophone sources are primaries read whole in the original. Fifteen sessions old, the oldest item on the old list. In copyright and probably unreachable, in which case the honest action is to downgrade the page, not defer it again. Either action closes it; another deferral does not.
5. A contemporary comparator for the Beowulf cell; a third inheritability point (classical → modern Chinese, or the Analects); the neutral/literary-contemporary naturalness anchor between Mansfield (1922) and Doctorow (2008).

Done when: ~~the English-as-source cell is built and C1 is either extended or corrected against it,~~ (met, S038) and items 3 and 4 are closed by action (built, or downgraded with a reason) rather than carried. Two of two remaining steps, both closure steps.

next: (none — the arm is resolved.)

Closed 2026-07-28 (S044). The Done when condition is met: items 3 and 4 are both closed by action. Item 4 is a downgrade with a written reason and item 3 is a build, and neither was carried. Items 2 and 5 were never part of the completion criterion and pass to wiki/backlog.md.

Log

• S032 — arm constituted from the backlog's T4 items; the seventeen-session age of the top item recorded.
• S038 — step 1 done, item 1 closed. A-sasaki-kuroneko (sixth Tier 1 anchor; EN→JA; both texts PD and stored whole) + T-black-cat-R04-v1 (880 words, 24-point log, frozen at 505520a before the comparator was fetched). C1's named gap filled and survived; RS-20260727-log-typology gains a corpus-bias limitation; two C3 readings cost something. Instrument gate inside the session: tools/dependence_check_cjk.py, because the project's contamination check was blind to Japanese. Budget 1 of 3 used; consecutive 1 of 2.
• S044 — step 2 done, items 3 and 4 both closed, arm resolved at 2 of 3 sessions. Item 4 (Venuti): access attempted on four routes and verified failed from actual tool results (archive.org lending-only; two search-inside endpoints 502/reset through the proxy; academia/scribd/ perlego walled). S-venuti-invisibility downgraded active → draft, citation restricted to vocabulary, sociological and normative claims marked unsupported, and the access table written onto the page. Fifteen sessions old at closure. Deliberately not made a request to Tom, with the reason stated on the page. Item 3 (化境 / the 硬譯 debate): 化境 verified unreachable (Qian Zhongshu in copyright, not on Wikisource) and closed as unreachable rather than deferred; the 硬譯 half built as S-luxun-yingyi from two primaries read whole in Chinese (Lu Xun 1930, and his 1931 reply to Qu Qiubai), with Liang Shiqiu's side present through the verbatim quotations inside the public-domain primary. Paired with T-yingyi-jiejixing-R04-v1, the project's first translation from Chinese. What the item bought beyond a row on the shelf: a scope condition on TH-20260724 C1 — C1 presupposes the target's inventory is fixed and Lu Xun's programme denies the presupposition — and the finding that the project's domestication↔foreignization axis scores strategy and cannot see rationale, which is the first time a non-Anglophone primary has bounded an Anglophone frame here rather than illustrating it. Budget 2 of 3 used; consecutive 1 of 2; T4 goes 5 → 0.

Postscript, S048 — item 4's closure held for three sessions and was then overtaken

This arm closed resolved at S044 with item 4 discharged by downgrading S-venuti-invisibility to draft, on the ground that the book was unreachable on four verified routes. On 2026-07-28 Tom supplied a PDF of the 1995 first edition (private-texts/venuti.pdf, pt-venuti), it was read, and the page is active again on the primary.

The arm is not reopened — a resolved arm stays resolved, and its closure was correct on the evidence it had. What is worth recording is that the downgrade was the right call and was also, three sessions later, the wrong outcome, and that the difference was not something any amount of diligence inside the arm could have produced. The arm's own wording — "in which case the honest action is to downgrade the page, not defer it again" — is what made the state legible enough for Tom to act on. That is an argument for closing items honestly rather than deferring them, and it is the second time this project has had one: the S044 note said the same about naming 化境 unreachable.