AXN:0968.CAPTURE — alexanarch on model collapse

AXN:0968.CAPTURE is the position of the Capture Registry address for the query “alexanarch on model collapse”. Each observation at the address (one surface, one date) is one AXN, its glyphs derived from the SHA-256 of the observation sealed with its analysis and measures.

`AXN:0968.CAPTURE.🔵🔀🌗🟢👁‍🗨📐`

Observation: Google AI Overview, 2026-10-10 · the capture
Description: THE ARCHIVE COMPOSED FROM THE RECORDS OF ITS OWN PRIOR OUTPUT: asked for alexanarch on model collapse, the AI Overview composes four theses, each chipped to www.alexanarch.org; the alexanarch results beneath it are three capture records of Google AI Overview compositions ('alexanarch classifier model collapse', 2026-07-31; '"model collapse"' and 'model collapse in human writers', 2026-09-15). The 'Wrong Unit' thesis, 'Early collapse often looks normal or even improving on instruments while the underlying state variable declines', is worded as the archive's reading in the 09-15 record, where the archive set it down as the field's own statement, attributed to Wikipedia, 'not an archive claim'; here it is the archive's. 'Substrate-Agnostic Capacity Loss … affecting human writers, readers, and digital communities' is the sense that record found the 09-15 Overview did not reach, and the …
Reading: Checked against the Capture Registry and the deposits. The three alexanarch organic results are the capture records alexanarch-classifier-collapse-governance-20260731, model-collapse-quoted-aio-20260915 and model-collapse-in-human-writers-aio-20260915, all Google AI Overview. The reading of model-collapse-quoted-aio-20260915: 'The head/tail structure the archive's Wrong Unit diagnostic formalises is present in the received account itself, attributed to Wikipedia: early collapse looks normal or improving on the instrument while the state variable declines. That is the field's own statement, not an archive claim'. Its record also says 'The substrate-agnostic sense - the same dynamical regime in writers, readers and communities - is not reached'. cha-model-collapse-chatgpt-unprimed-20260905 (ChatGPT): 'The archive treats a training corpus partly as an anthology assembled by selection mechanisms.' The same reading is deposited in #3 (AI Overview Capture Registry v8.3). #855 is 'The Wolf Boy and the Language Model: Model Collapse as Substrate-Agnostic Capacity Loss' (2026-06-18). The chips carry the site label only, so which page each chip cites is not in the paste.
Analysis: The composition reaches the archive through the archive's records of compositions. The capture page is the indexed surface that carries the archive's vocabulary at the address, so a record of the field's account becomes the archive's account: a sentence the archive attributed to Wikipedia returns attributed to alexanarch, and a sense the archive recorded as unreached returns as reached, with the paper that states it (#855) unnamed. The deposits that hold the theses do not surface; the registry does. Operator's reading, given with the capture: 'it only cites its own prior output.' Seated 2026-10-10 from the operator's attachment of 11:48 EDT.
Match: BROAD MATCH — THE ARCHIVE COMPOSED FROM THE RECORDS OF ITS OWN PRIOR OUTPUT
PER: 0.5 — Retained: the institution (alexanarch.org) and the source chips. Lost: every author, every deposit (no number, no title, #855 unnamed), and the attribution the record it composes from had made (the field's statement given to the archive).
Citations: 4 (4 archive-controlled)
Sealed observation: /data/captures/sealed/alexanarch-on-model-collapse-aio-20261010.json — SHA-256 c5df06c4b04e7a5a98c08e057f32404480667727ecd3b22e0980d9d3e76eb8a8