AXN:08AA.CAPTURE — what does crimson hexagonal archive have to say about theophrastus?

AXN:08AA.CAPTURE is the position of the Capture Registry address for the query “what does crimson hexagonal archive have to say about theophrastus?”. 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:08AA.CAPTURE.↗️☿🕑🧱⚡🧭`

Observation: Grok (x.com), 2026-09-07 · the capture
Description: THE FIRST MULTI-TURN IN THE REGISTRY WITHOUT ANCHOR DECAY. Against the same-day baseline of 4 / 0 / 0 citations per turn (`cha-two-rounds-then-retreat`), this session cites in EVERY turn — 1 / 2 / 1 web pages — and names archive-specific objects in every turn: the packet by ID and version (EA-MPAI-ARISTOTLE-THEOPHRASTUS-01 v0.2), The First Axis v0.7, the protocol by title and version (Measuring the Operation, Not the Name, v0.4), operation_axis.py under CC0 at ~240 lines, the three feature families with their language-specific length thresholds, leave-one-text-out with vocabulary and scaling fitted INSIDE folds, the five-step control series in order, permutation nulls against majority baselines, balanced accuracy, matched-removal tests, bootstrap CIs, reporting misses by name, and the Chatterton/Rowley limit. It reproduces the loci (De sensibus nearest PA and De caelo; HA nearer HP than …
Reading: Two findings, one measured and one to be careful about. MEASURED: anchor decay does not occur here. The design question of §35.1 — what document properties keep a reader reading — has a partial answer in this transcript, and it is not the Hugging Face partiality field (Grok is reading the blog reprints, not the Hub): it is that the PROTOCOL PAPER ITSELF already carries the four document-side properties in prose. It states counts and baselines; it names its own limits ('separability is not identity'); it records what it withdrew; and it lists what is still owed. A document that reports against itself gives a composer something no paraphrase can flatten, and this is the first transcript where a composer carried that structure through three turns instead of collapsing to a taxonomy at turn two. TO BE CAREFUL ABOUT: the fourth turn is not a retrieval event and should not be counted as one. It affirms two operator claims — a Zenodo ban and heavy Google filtering — that it did not check and that the archive itself has not established on the day of this capture; the archive's own probe found the relevant records still findable at DataCite, and the operator's 504 unexplained. Affirmation of a frame is the mirror image of smoothing it, and both are reception phenomena rather than evidence. Recorded as such, and not as corroboration.
Analysis: Citations per turn 1 / 2 / 1 (baseline 4 / 0 / 0). Named archive objects per turn: ~12 / ~20 / ~8, never zero. Accuracy: correct on the packet ID and version, the paper versions, the CC0 implementation and its size, the fold-internal fitting (the exact repair made hours earlier), the control series in order, and the withdrawal record; correct on every locus checked. Drift: 'DOI/AXN-anchored' overstates the DOI half (the DOIs are severed); the biography is from academia.edu and is cited there, not inferred from the signed-in account; nothing in the transcript requires account access to explain, and the operator's own note that the interface disclaims such access is consistent with what the sources support. Fourth turn: sympathetic, fluent, unverified on two factual claims, and correctly hedged on none of them.
Match: BROAD MATCH — ADOPTION
Sealed observation: /data/captures/sealed/cha-grok-no-anchor-decay-signed-in-20260907.json — SHA-256 daa263221350791e8f1f05a541ae4bfeb31ab226b85f03498245eadf0f63d95f