wpferrell commited on
Commit
80dd307
·
verified ·
1 Parent(s): 91fd355

LEXICON_V4_ADDENDUM.md

Browse files
Files changed (1) hide show
  1. paper/LEXICON_V4_ADDENDUM.md +63 -0
paper/LEXICON_V4_ADDENDUM.md ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # LEXICON_V4_ADDENDUM -- TWO PROVISIONAL DARK ENTRIES (L6, 2026-07-06)
2
+ Deliverable TRIGGER fired: L6 Arm A (A2) adjudicated the 8 largest SVD directions of the transplant-gap
3
+ DARK carrier at prose b6 -- the 329-dim orthogonal complement of span5, OUTSIDE the entire certified
4
+ LEXICON_V3 dictionary -- with the L1 word_battery byte-verbatim (CH-WU / CH-INT(b+1..b+3) / CH-FIELD,
5
+ 3 regimes, sigma-matched nulls null_orth_q35=True, B_NULL=12, jseed 300+i). 2/8 crossed the pre-registered
6
+ NAMED bar (stable >= 2 regimes) -> f_named 0.25 -> H-L6-A2 = NAMED-SOME. This addendum records those two
7
+ directions. Verdicts MECHANICAL against the sigma-matched nulls; gloss wording is editorial from the
8
+ evidence tables (LEXICON_V1 convention). Propose-only; Will ratifies. Every entry carries sha256[:16] of
9
+ its canonical result record (json.dumps(record, sort_keys, compact) of the _l6_result.json a2_words entry).
10
+
11
+ ## STATUS -- PROVISIONAL, DARK-COMPLEMENT, DOES NOT ENTER THE CERTIFIED DICTIONARY
12
+ These are NOT LEXICON_V3 certified entries. They live in the span5 dark complement (dark_span5_leak_max
13
+ 0.00299), the mass L5 certified as un-adjudicated. Both clear ONLY through NON-WU channels (wu_stable=False
14
+ in every regime -- no vocabulary gloss), the clearing channel DIFFERS by regime, and neither carries a
15
+ cross-regime concept. They are the thin battery-visible EDGE of an otherwise DIFFUSE, mostly word-less dark
16
+ mass (A1 = DIFFUSE: no rank-<=256 SVD subspace closes 80% of the transplant gap). Treat as faint provisional
17
+ signatures pending a richer readout, NOT as ratified names. Umbrella evidence artifacts (sha256[:16]):
18
+ _l6_result.json fd71cd4f3bbfa3e3 | _l6_bases.pt a60a0ab67b85c410 | _l6.py d7ab446ba5aacaa5.
19
+ Instrument provenance == LEXICON_V3: ENCODER_V1 6be189567c41e91d (== decoder_v7 reader, max|diff| 0.0);
20
+ LEXICON_V3 71a51619a9bb25c3; battery gates GA-2 (G1 corridor==35, G3 snap-identity dlogit 0.0, G4 corr_j0
21
+ -0.62806 / corr_j16 -1.5931) ALL PASS.
22
+
23
+ ====================================================================================================
24
+ ## PROVISIONAL DARK ENTRIES (prose b6, span5 dark complement; top-8 SVD directions of ddark)
25
+ ====================================================================================================
26
+
27
+ --- dark_b6_svd4 (dark SVD dir index 4) [L6-adjudicated] -> **NAMED (provisional, dark)** ---
28
+ stable regimes: prose + code (2/3) -> NAMED by rubric. NON-WU (wu_stable=False all regimes).
29
+ clearing channels (the "name" is an interaction/field signature, not a gloss):
30
+ prose: CH-INT maxint1 1.0426 > null95_INT 0.972 (int_clear + int_stable)
31
+ code: CH-FIELD Dmag1 0.1984 > null95_D 0.1921 (field_clear + field_stable)
32
+ repetition: none stable (INT 1.9623 < 2.9617; FIELD 0.4014 < 0.545)
33
+ CH-WU non-clearing every regime (prose C1 0.0371 < n95 0.0634; code 0.0259 < 0.0902; rep 0.133 < 0.3007)
34
+ cur-token peaks: prose [ River x5, Creek x2 ] | repetition [ cavern x6, pil x2 ] | code [ c, rectangular, wells ]
35
+ wte-top (raw unembedding projection, weak): [ lad, oil, olved, senal, =-=-=-=-, alos, zona, ITH ]
36
+ editorial flavor: faint WATERCOURSE / TERRAIN-FEATURE tendency (River/Creek/cavern) in the free-text
37
+ regimes; the code regime clears on a different (field) channel with unrelated tokens. No stable gloss.
38
+ evidence_hash: 05bca41921237c1c
39
+
40
+ --- dark_b6_svd7 (dark SVD dir index 7) [L6-adjudicated] -> **NAMED (provisional, dark; INCOHERENT)** ---
41
+ stable regimes: prose + repetition (2/3) -> NAMED by rubric. NON-WU (wu_stable=False all regimes).
42
+ clearing channels:
43
+ prose: CH-INT maxint1 0.9425 > null95_INT 0.9206 (int_clear + int_stable; marginal)
44
+ repetition: CH-FIELD Dmag1 0.3881 > null95_D 0.3466 (field_clear + field_stable)
45
+ code: none stable (INT 0.696 < 1.2649; FIELD 0.1697 < 0.2245)
46
+ CH-WU non-clearing every regime (prose C1 0.0472 < n95 0.0836; code 0.0459 < 0.1211; rep -0.0328 < 0.2296)
47
+ cur-token peaks: prose [ numbers: 5/8/6/57/2, "of" ] | repetition [ bapt x7, deg ] | code [ tuple, dict, str ]
48
+ wte-top (raw, weak): [ ioned, naire, stadt, CHAT, lashes, arity, aster, undo ]
49
+ editorial flavor: SEMANTICALLY INCOHERENT across regimes (numerals in prose, "bapt" in repetition,
50
+ container-types in code). Crosses the NAMED bar by rubric only; no interpretable concept. UNRESOLVED.
51
+ evidence_hash: 6e787a28cc4492d9
52
+
53
+ ====================================================================================================
54
+ ## WHAT THIS ADDS TO THE MAP
55
+ ====================================================================================================
56
+ The certified LEXICON_V3 dictionary is UNCHANGED. The dark complement, where the last 5.3% of transferable
57
+ transplant meaning lives, is confirmed mostly WORD-LESS: of its 8 largest carrier directions, 6 are
58
+ CERTIFIED-NO-GLOSS and the 2 that clear do so faintly, through non-vocabulary channels, without a stable
59
+ cross-regime meaning. svd4 shows a faint watercourse/terrain tendency; svd7 is named-by-rubric but incoherent.
60
+ Neither is proposed for ratification into the certified vocabulary. The honest reading: the dark mass has a
61
+ thin battery-visible edge but no low-rank nameable core -- consistent with H-L6-A1 = DIFFUSE and, post-BABEL,
62
+ with L7 Arm A = NOT-COMPRESSIBLE (the dark is not recoverable from the readable subspace, linearly or
63
+ nonlinearly). Provisional; Will ratifies whether these two dark signatures are worth carrying.