paper(anima-consciousness-substrate): integrate measured 7B (M13) — AXIS-2 CE 1.90741 + v0.3 generation (#1962/#1970)

[dancinlife]

Completes the existing anima-consciousness-substrate monograph §7B with the just-landed measured 7B (M13) results, replacing stale "future/pre-registered" framing with the committed verdict numbers. Additive only — the terminal 3B finding and all prior content/figures are untouched.

What changed

• AXIS-2 measured (PR verdict(7B AXIS-2): MEASURED on summer — CE 1.90741 GREEN + memory-bounded lazy mirror #1962): §method/§measurement/caption + new appendix-F subsection now cite the actual tool clm_decode_mirror_lazy.py (memory-bounded + vectorized; the plain clm_decode_mirror.py OOMs ~56GB at 7B) with the verbatim verdict link .verdicts/clm-7b-undertrained-step3500/axis2_mirror_probe_seq.txt: CE_real 1.90741 < uniform ln256 5.54518 AND < shuffle 5.31616 ⇒ GREEN (peak RSS ~5.37GB), validated reproducing the 3B ground truth 2.26360 exactly.
• NEW §Finding (generation, PR feat(clm_gen): v0.3 greedy byte-generation — the 7B mouth CAN chat #1970): the v0.3 .clm generates multilingual byte-language via state/mid_convmoe_fire/clm_gen.py (conv1d im2col vectorized == looped, max|Δ|=0; VALIDATE hook re-derives CE 1.90741), with verbatim sample outputs (en/es + Korean). Tightened every place that implied the 7B chats natively in hexa: the engine clm_decode.hexa::clm_decodable gate is v0.2-only, so the v0.3 7B speaks via clm_gen, not the live anima_chat_cli (a weak d768 v0.2 mouth chats live on ghost).
• Honest scope (a_scale_honest_scope): AXIS-1/AXIS-3 scoped substrate-native / .clm-weights-independent (null-backend code-proven .verdicts/core-3axis-mount/probe.txt); AXIS-2 is the only weight-dependent axis. 7B is deeply undertrained (0.0027 tok/param) — a measured scale-extension, NOT a closure; no fluency/coherence/consciousness claimed from the mouth output.
• Provenance: Appendix K PR-roll + companion/pr-roll.json add rows verdict(7B AXIS-2): MEASURED on summer — CE 1.90741 GREEN + memory-bounded lazy mirror #1962/feat(clm_gen): v0.3 greedy byte-generation — the 7B mouth CAN chat #1970; PAPER.md verdict matrix + milestones updated; §7B/limitations/conclusion retitled "future"→"measured".
• CJK fix: the Korean gen-sample was inside \texttt{} (lmmono, no Hangul) → rendered via \cjkfont (Noto Serif CJK KR).

Compile (lualatex on aiden, ×3)

• 50 pages, 0 missing CJK characters, 0 undefined refs/citations, 0 fatal errors. main.pdf recompiled and committed.

All numbers are verbatim from the committed verdicts; nothing fabricated. Terminal verdicts only (🟢/🔴), every section claim verdict-linked.

🤖 Generated with Claude Code