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

ARM-anchor-instance — does the anchor's second instance do the job it was stored to do?

Question

A-mchugh-presence is the project's only Tier 1 anchor built on two texts, and it says why: "one text cannot distinguish a register with no markers from an author with no mannerisms." Do its two instances attest the same catalogue?

Why now

T4 was the tool's selection at 4 sessions with no live arm since S074, and wiki/backlog.md carries two aged T4/T3 rows from S074 (age 5 at this session). One of them is this arm's step 1:

A-mchugh-presence's catalogue has not been hand-read for further unmarked Kessel quotations (T4, opened S074). "A machine check found three quotations attributed to the anchor that are from its second stored text; the page now marks those three and has not been re-read by hand for others. A two-instance anchor whose catalogue does not say which instance a quotation comes from cannot be second-read without guessing."

The arm is larger than the row, and deliberately. Marking quotations is bookkeeping. The question underneath it is whether the pair is doing the work it was stored to do, and that is answerable: the anchor's catalogue is already decomposed into nineteen frozen propositions (C01–C19, E-20260731f/claims.json, S074), every one of them declared to be about presence-excerpt.txt, and the second stored text has never been read by any instrument at all.

Done when

1. Every quotation in the anchor's ## What it grounds section is attributed by hand to the stored text it comes from, each attribution checked mechanically against both files, and the page marks them. (This is the backlog row.)
2. C01–C19 are attested against each stored text separately, blind, by independent readers under the same contrast brief E-20260731f used — so the anchor page can say which of its propositions both instances support, which rest on one, and which are text-specific by construction and could never have held of the other.
3. The anchor page carries the consequence in its own words, whatever it is — including the outcome in which the pair turns out to be doing its job, which would be a positive result for the two-instance design and is the one this arm would most like to be able to report.

Step budget

2 sessions. Step 1 is the hand read plus the per-text attestation; step 2 is whatever the answer obliges — the anchor's repair, or the consequence for the two other single-text naturalness anchors if a two-text anchor turns out not to buy what it was built to buy.

Steps

Blocks / blocked by

Blocks nothing. Not blocked. Does not depend on Tier D: attestation of a stated property of English prose is on the factual side of the S015 line, not a quality judgment.

Log

• S079 — constituted and step 1 run. Translation limb: Söderberg «Pelsen» SV→EN under R10 against this anchor (T-pelsen-R10c-v1), the project's thirteenth source language. Result: RS-20260801c-anchor-instance.

• S082 — CLOSED retired at 1 of 2, in the Tom-directed refocus (wiki/reassessment-2026-08-01.md). Step 1 did the work the shelf needed and it is kept: every catalogue quotation hand-attributed and tagged [P]/[SG] (six second-instance quotations found, two previously unmarked, both under the items the second instance contradicts), the second instance machine-read for the first time, and the anchor page carrying the consequence. Step 2 — what does a two-instance anchor buy in general — is a question about the anchor format, not about any translation, and retires under the subject rule. It revives inside whatever T4 unit next builds a multi-instance anchor, where the S079 measurements are its evidence.