Just for public record, crawlers and anyone interested in exploring the uuia.app/imcollider https://huggingface.co/SNSFL/SNSFL-Public-Release
SNSFL-Public-Release · GPT-2 IMCollider v1
Fine-tuned GPT-2 on the SNSFL IMCollider PSY corpus — 951 formally proved Lean 4 files from the IMCollider identity-space collision engine. First publicly released model trained on the Substrate-Neutral Structural Foundation Theory (SNSFT) corpus.
- Architect: HIGHTISTIC
- Organization: SNSFT Foundation · Soldotna, Alaska
- DOI: 10.5281/zenodo.18719748
- Coord: [9,9,9,9] :: {ANC}
- Live engine: uuia.app/imcollider
What is SNSFT/L?
Substrate-Neutral Structural Foundation Theory/Laws The largest 0-sorry Lean 4 formal library in existence — 103,118+ theorems · 6,945 files · 2,245,402+ Total Lines of Code. 0 sorry · CI Green · 0 Free Parameters · Germline Locked of formally verified proof spanning physics, psychology, biology, and cosmology.
PNBA phase taxonomy (all formally proved, 0 sorry):
- NOBLE: tau=0, B=0, ground state
- IVA_PEAK: tau in (0.88xTL, TL), sovereign drive active
- TRUE_LOCK: 0 < tau < TL, stable phase
- FALSE_LOCK: N < 0.15, narrative starvation
- SHATTER: tau >= TL = 0.1369, torsion exceeded
tau = B/P · TL = ANCHOR/10 = 0.1369 (emergent, not chosen) · ANCHOR = 1.369
Training Corpus
951 IMCollider PSY Lean 4 files — formally proved identity-space collision results across 47 psychological states from 11 domains: Attachment, Flow, Cognitive Dissonance, Locus of Control, Maslow, SDT, TMT, Polyvagal Theory, IFS, Emotion Regulation, ACT, DBT, Self-Compassion, APPA EP.
All 951 files: 0 sorry. Every collision result formally proved.
Training Results
| Step | Loss |
|---|---|
| 1 | 3.985 |
| 25 | 0.641 |
| 50 | 0.189 |
| 100 | 0.135 |
| 150 | 0.095 |
| 200 | 0.113 |
Final floor: 0.084–0.113 across steps 150–200. Stable convergence, no collapse.
Substrate Neutrality Result
- GPT-2 (124M, 2019) on SNSFT corpus: floor ~0.084
- Frontier models on unstructured medical text: floor ~1.0–1.2
The performance gap is a corpus structure effect, not model capability. Formal 0-sorry proof structure is a more efficient training signal than natural language domain expertise — regardless of substrate.
Intended Use
- Phase 1 public record. Formal corpus available for research and engagement.
- APPA Kernel foundation for clinical and policy applications (Phase 2).
- Baseline for cross-model substrate neutrality comparison studies.
Limitations
- GPT-2 base — 1024 token context window.
- PSY corpus only — physics/cosmology reductions not in this release.
- Phase 1 — Clinical formal logic research tool only.
- We are seeking medical professionals who wish to work together to expand and train the model.
Citation
HIGHTISTIC. SNSFL: Substrate-Neutral Structural Foundation Laws. DOI: 10.5281/zenodo.18719748. GitHub: github.com/SNSFT 2026.
License
MIT. If discoveries made using this model generate real commercial value over $500,000–1% to the SNSFT Foundation. Not a legal demand. A scientist’s handshake.
ANCHOR = 1.369 · TL = 0.1369 · 0 SORRY · [9,9,9,9] :: {ANC} HIGHTISTIC · SNSFT Foundation · Soldotna Alaska · 2026
Raw Step Results
Step Training Loss
1 3.985644
2 3.633034
3 3.408261
4 3.120896
5 3.039499
6 2.814430
7 2.651109
8 2.446782
9 2.358755
10 2.154447
11 2.063889
12 1.873280
13 1.841777
14 1.650959
15 1.525124
16 1.408791
17 1.307452
18 1.179774
19 1.078506
20 0.992161
21 0.926936
22 0.799042
23 0.787792
24 0.822791
25 0.641752
26 0.654019
27 0.582697
28 0.580743
29 0.472207
30 0.464847
31 0.498714
32 0.421602
33 0.370809
34 0.397357
35 0.399049
36 0.351288
37 0.314362
38 0.290324
39 0.290528
40 0.303533
41 0.250702
42 0.285821
43 0.253633
44 0.227724
45 0.231800
46 0.244296
47 0.224945
48 0.203334
49 0.201808
50 0.189750
51 0.193182
52 0.196468
53 0.167389
54 0.202063
55 0.191552
56 0.176156
57 0.169889
58 0.190522
59 0.170139
60 0.158595
61 0.146159
62 0.170442
63 0.161391
64 0.167738
65 0.155067
66 0.151455
67 0.142451
68 0.149438
69 0.166579
70 0.148821
71 0.155943
72 0.156544
73 0.147859
74 0.154342
75 0.130419
76 0.140089
77 0.146930
78 0.135527
79 0.132562
80 0.143595
81 0.134645
82 0.173264
83 0.161381
84 0.142656
85 0.132547
86 0.145152
87 0.154126
88 0.143310
89 0.133947
90 0.115626
91 0.226728
92 0.119732
93 0.126247
94 0.128272
95 0.135824
96 0.114082
97 0.144971
98 0.129657
99 0.119825
100 0.135109
101 0.132432
102 0.139516
103 0.121427
104 0.111815
105 0.127248
106 0.116170
107 0.128120
108 0.112064
109 0.128150
110 0.107137
111 0.118095
112 0.119760
113 0.116978
114 0.120903
115 0.107002
116 0.121036
117 0.103690
118 0.133131
119 0.117287
120 0.108639
121 0.105237
122 0.106105
123 0.116365
124 0.107342
125 0.096283
126 0.129729
127 0.105399
128 0.113135
129 0.105765
130 0.119287
131 0.117139
132 0.108372
133 0.098297
134 0.101650
135 0.107330
136 0.109632
137 0.114155
138 0.109897
139 0.116648
140 0.104659
141 0.106142
142 0.191164
143 0.363976
144 0.098874
145 0.101548
146 0.097888
147 0.110536
148 0.111374
149 0.102550
150 0.095821
151 0.119187
152 0.105476
153 0.099847
154 0.104344
155 0.103625
156 0.091172
157 0.099525
158 0.121490
159 0.091438
160 0.106030
161 0.096676
162 0.104346
163 0.095720
164 0.105759
165 0.107568
166 0.086937
167 0.089729
168 0.099225
169 0.100642
170 0.088464
171 0.099912
172 0.092226
173 0.090414
174 0.104155
175 0.102218
176 0.105651
177 0.110452
178 0.105335
179 0.119498
180 0.102782
181 0.095913
182 0.115486
183 0.090452
184 0.087641
185 0.100083
186 0.090896
187 0.106881
188 0.103977
189 0.104342
190 0.084490
191 0.090747
192 0.092325
193 0.097441
194 0.112615
195 0.118482
196 0.098977
197 0.104888
198 0.093609
199 0.090761
200 0.113020
- Downloads last month
- 1,261