◀ THE FOLD0ROOT.AI // WORLD II · SPAWN · THE TOOLCHAIN◆ .dlw.fold

Pipeline run

corpus = this conversation · trained on itself LIT · REPRODUCED

01 TOKEN

tokens
2,235
types
910
vocab (min 2)
299
hapax
67.1%

02 EMBED

matrix
299×299
density
0.1006
weighting
PPMI
dims (SVD)
48

03 CORPUS

source
1 thread
words
2,235
needed
10⁹+
shortfall
~6 orders

04 AGENT

probes
8
neighbours
top 4
look plausible
yes
are real
no

FREE ASSOCIATION — FULL CORPUS

weightspart0.69activations0.61coordinates0.56by0.55
corpusorder0.72asked0.65then0.64as0.63
coherenceran0.68convergence0.67correctness0.65when0.63
divergencescores0.66semantic0.64paraphrase0.55zero0.52
attentionfails0.78performance0.76identical0.71machine0.66
booksold0.57happens0.56fair0.46curation0.44
superpositionsingle0.72things0.69directions0.68unrelated0.59
signalexternal0.70means0.70only0.67itself0.65

SPLIT-HALF RELIABILITY

0.029 mean overlap
chance level 0.0167
median 0.000
zero overlap 86.6% of vocab
2590/5
361/5
42/5
03/5
04/5
05/5
shared neighbours out of top 5, first half vs second half of the same corpus

Trained twice on disjoint halves of one corpus. 259 of 299 words share zero neighbours between the two runs. Four words share two. None share three or more. Mean overlap sits at 0.029 against a chance floor of 0.0167 — distinguishable from noise, and not by much.

The associations above are not wrong so much as unfounded. weights → activations, coordinates. signal → external. superposition → directions. Every one reads correctly, which is exactly the problem: they look like a result at a scale that cannot produce one.

Volume is the whole story. 2,235 tokens, 910 types, 67% of them appearing exactly once. Distributional semantics needs corpora six orders of magnitude larger. What got measured here is the co-occurrence structure of one conversation, which is a transcript, not a language.

Split-half is the test that decides. Not whether the output looks sensible — it does — but whether the same input twice yields the same structure. It does not. That is the replication check that was missing from fourteen model runs, now applied to a corpus small enough to fail it visibly.

Stamped LIT because the reliability figure is reproducible from the corpus and script, not because the embeddings mean anything. The finding is the failure.

◆ sealed .dlw.fold → ROOT_0 · a sphere of THE TOOLCHAIN · vendored from David’s corpus · David Lee Wise (ROOT0), with AVAN