Repository navigation
Expand file tree
/
Copy pathresearch-note.html
More file actions
745 lines (724 loc) · 30.2 KB
/
Copy pathresearch-note.html
File metadata and controls
745 lines (724 loc) · 30.2 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="A public research note documenting Project SLSSM-01's observable spectral dynamics evidence, metric definitions, limitations, and disclosure boundary.">
<link rel="canonical" href="https://trinity-research.com/research-note.html">
<meta name="author" content="Freddie Hunt, Joshua Hunt, PhD">
<meta name="robots" content="index, follow, max-image-preview:large">
<meta property="og:type" content="article">
<meta property="og:site_name" content="Trinity Research">
<meta property="og:title" content="Observable Spectral Dynamics in a Recurrent Sequence Model">
<meta property="og:description" content="Research note with evidence from Project SLSSM-01's v3 short-context run, including EDMD concordance, signed phase behavior, prompt manifolds, and limitations.">
<meta property="og:url" content="https://trinity-research.com/research-note.html">
<meta property="og:image" content="https://trinity-research.com/social-card.png">
<meta property="og:image:type" content="image/png">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta property="og:image:alt" content="Trinity Research social card">
<meta property="article:published_time" content="2026-06-28">
<meta property="article:author" content="Freddie Hunt">
<meta property="article:author" content="Joshua Hunt, PhD">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="Observable Spectral Dynamics in a Recurrent Sequence Model">
<meta name="twitter:description" content="A public research note with rounded evidence from Project SLSSM-01's observable spectral dynamics work.">
<meta name="twitter:image" content="https://trinity-research.com/social-card.png">
<title>Research Note | Project SLSSM-01 Observable Spectral Dynamics</title>
<link rel="icon" href="favicon.svg" type="image/svg+xml">
<link rel="alternate" type="application/rss+xml" title="Trinity Research Signals" href="https://trinity-research.com/feed.xml">
<link rel="stylesheet" href="styles.css?v=20260805a">
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "ScholarlyArticle",
"headline": "Observable Spectral Dynamics in a Recurrent Sequence Model",
"description": "A public research note documenting Project SLSSM-01's observable spectral dynamics evidence, metric definitions, limitations, and disclosure boundary.",
"datePublished": "2026-06-28",
"dateModified": "2026-07-10",
"url": "https://trinity-research.com/research-note.html",
"image": "https://trinity-research.com/social-card.png",
"author": [
{
"@type": "Person",
"name": "Freddie Hunt",
"jobTitle": "Founder / Lead Research Scientist",
"sameAs": [
"https://www.linkedin.com/in/freddie-hunt/",
"https://x.com/ExtraDeadFred"
]
},
{
"@type": "Person",
"name": "Joshua Hunt, PhD",
"jobTitle": "Research Collaborator / Academic Advisor",
"sameAs": [
"https://www.linkedin.com/in/joshua-hunt-936973255/",
"https://x.com/_Professor_Hunt",
"https://www.latech.edu/directory/joshua-hunt.php"
],
"affiliation": {
"@type": "CollegeOrUniversity",
"name": "Louisiana Tech University"
}
}
],
"keywords": [
"spectral state-space model",
"language model interpretability",
"EDMD",
"prompt manifold",
"recurrent sequence models"
]
}
</script>
</head>
<body>
<header class="site-header">
<a class="brand" href="index.html" aria-label="Trinity Research home">
<span class="brand-mark">T</span>
<span>Trinity Research</span>
</a>
<nav aria-label="Primary">
<a href="index.html#work">Work</a>
<a href="slssm-01.html">Research</a>
<a href="slssm-01.html#evidence">Evidence</a>
<a href="slssm-01.html#atlas">Atlas</a>
<a href="research-note.html" class="active" aria-current="page">Note</a>
<a href="signals.html">Signals</a>
</nav>
</header>
<main>
<article class="research-note">
<header class="note-hero">
<p class="eyebrow">Public research note</p>
<h1>Observable Spectral Dynamics in a Recurrent Sequence Model</h1>
<p class="lede">
This note describes what we can publicly disclose about Project SLSSM-01: the research
question, the evidence gathered from the v3 short-context run, and the
diagnostic vocabulary used to study the model. Implementation details,
weights, and the training recipe are outside the public disclosure boundary.
</p>
<div class="note-meta">
<span>Version 0.1</span>
<span>Public research summary</span>
<span>Prepared for credibility review</span>
</div>
</header>
<section>
<h2>Abstract</h2>
<p>
Project SLSSM-01 studies recurrent language models with inspectable spectral
dynamics, using state evolution rather than quadratic self-attention as the
primary sequence mechanism. The central idea is that a sequence model can be
studied through its learned dynamical structure as well as through loss curves
and generated text.
</p>
<p>
We report evidence from the v3 short-context run: stable
training across 250,000 steps, roughly 32.8 billion processed tokens,
improving loss, persistent signed phase populations, and close agreement
between measured trajectory behavior and learned spectral structure. We also
report a limitation: the short-context run did not build a large long-memory
mode population, motivating the next long-context experiment.
</p>
</section>
<section>
<h2>Model Class</h2>
<p>
Project SLSSM-01 is best described publicly as a stable linear spectral state-space
language model. In practical terms, that means the model is built around a
stable linear recurrent dynamics core whose internal behavior can be studied
as learned modes.
</p>
<p>
That does not mean the entire system is linear end to end. Language enters
and leaves the dynamics through learned nonlinear components. The important
distinction is that the core state evolution is stable and mode-structured,
which is what makes the Spectral Atlas, Extended Dynamic Mode Decomposition
(EDMD) Concordance, and Prompt Manifold analysis possible.
</p>
</section>
<section>
<h2>Public Spec Sheet</h2>
<p>
The following specifications are intentionally high-level. They are enough
to orient a technical reader without disclosing implementation details.
</p>
<div class="spec-grid">
<article>
<span class="finding-kicker">Class</span>
<h3>Stable linear spectral state-space language model</h3>
<p>Mode-structured recurrent dynamics are the analyzable core.</p>
</article>
<article>
<span class="finding-kicker">Scale</span>
<h3>360M-class research model</h3>
<p>Large enough to test training dynamics, small enough to iterate and inspect.</p>
</article>
<article>
<span class="finding-kicker">Sequence mechanism</span>
<h3>Recurrent/spectral core</h3>
<p>Designed around state evolution rather than a transformer self-attention stack as the primary mechanism.</p>
</article>
<article>
<span class="finding-kicker">Training run</span>
<h3>250k steps, 32.8B tokens</h3>
<p>Rounded summary from the v3 short-context run.</p>
</article>
<article>
<span class="finding-kicker">Hardware</span>
<h3>Single A100-class run</h3>
<p>The analyzed run was practical at research scale rather than requiring a large cluster.</p>
</article>
<article>
<span class="finding-kicker">Evidence</span>
<h3>Loss, EDMD, mode, prompt probes</h3>
<p>Analysis used multiple views instead of relying on one curve or one generation sample.</p>
</article>
</div>
</section>
<section>
<h2>What Is Being Claimed</h2>
<p>
The claim is deliberately narrow:
</p>
<ul>
<li>A stable linear spectral state-space language model was trained and analyzed.</li>
<li>Its internal dynamics were measurable enough to track across checkpoints.</li>
<li>Signed phase behavior remained present rather than collapsing into a single direction.</li>
<li>Trajectory-based analysis and learned spectral structure showed stable agreement after early training.</li>
<li>The analysis exposed a concrete next problem: long-memory capacity.</li>
</ul>
<p>
This is not a claim that Project SLSSM-01 replaces transformers, solves interpretability,
or discloses a reproducible open architecture.
</p>
</section>
<section>
<h2>Diagnostic Vocabulary</h2>
<div class="note-grid">
<div>
<h3>Spectral Atlas</h3>
<p>
A map of learned mode families and their movement across training
checkpoints. The public visualization uses synthetic data to explain the
idea without exposing raw internal state.
</p>
</div>
<div>
<h3>EDMD Concordance</h3>
<p>
A comparison between measured trajectory dynamics and learned internal
structure. Public summaries show aggregate agreement, not the
implementation-specific transformation path.
</p>
</div>
<div>
<h3>Prompt Manifold</h3>
<p>
A prompt-conditioned view of mode usage. Public examples use labeled
synthetic token paths to show the concept without exposing actual
activations.
</p>
</div>
<div>
<h3>Spectral Convergence Signal</h3>
<p>
A training-time signal used to reason about changing mode pressure and
stabilization. Public discussion stays qualitative.
</p>
</div>
</div>
</section>
<section>
<h2>Public Evidence Summary</h2>
<div class="evidence-table-wrap note-table">
<table class="evidence-table">
<thead>
<tr>
<th>Checkpoint</th>
<th>Tokens</th>
<th>Best loss</th>
<th>Validation mean</th>
<th>EDMD mean distance</th>
<th>Interpretation</th>
</tr>
</thead>
<tbody>
<tr>
<td>30k</td>
<td>3.9B</td>
<td>4.03</td>
<td>3.91</td>
<td>0.0113</td>
<td>early organization</td>
</tr>
<tr>
<td>102k</td>
<td>13.4B</td>
<td>3.38</td>
<td>3.40</td>
<td>0.0079</td>
<td>agreement tightened</td>
</tr>
<tr>
<td>204k</td>
<td>26.7B</td>
<td>3.24</td>
<td>3.28</td>
<td>0.0077</td>
<td>stable dynamics</td>
</tr>
<tr>
<td>250k</td>
<td>32.8B</td>
<td>3.16</td>
<td>3.27</td>
<td>0.0078</td>
<td>short-context finish</td>
</tr>
</tbody>
</table>
</div>
<p>
The numbers are rounded public summaries. They are included to show that the
work is anchored in an actual training run, not only in diagrams or theory.
</p>
</section>
<section id="evidence-guide">
<h2>Evidence Guide</h2>
<p>
This is author-reported evidence from one analyzed training run. Values are
rounded, the public record does not include uncertainty intervals, and no
external replication is claimed. The CSV below mirrors the displayed tables
so readers can inspect or plot the released summary without transcribing it.
</p>
<p>
<a class="button primary" href="evidence/checkpoint-summary.csv" download>Download rounded checkpoint CSV</a>
</p>
<p>
Blank retention or rotation fields in the CSV mean that those summaries
are not included in the rounded public record for that checkpoint; they
should not be interpreted as zero.
</p>
<p>
The measurement method itself can be checked without trusting this
model: a reproducible EDMD demonstration on public Pythia-14m training
checkpoints, with a pre-specified protocol, dated post-run
amendments, pinned model commits, and shuffled-pairing and
persistence controls, runs on an ordinary CPU in seconds.
</p>
<p>
<a class="button secondary" href="https://github.com/trinity-research/pythia-edmd-demo">Run the public demo on GitHub</a>
</p>
<div class="definition-grid">
<article>
<h3>Best loss</h3>
<p>The lowest rounded training objective reported by the stated checkpoint. It is not a cross-model benchmark.</p>
</article>
<article>
<h3>Validation mean</h3>
<p>The rounded mean objective from the checkpoint validation probe; the final probe used twenty batches.</p>
</article>
<article>
<h3>EDMD distance</h3>
<p>Nearest-distance summary between trajectory-derived modes and learned spectral structure under a fixed analysis transform. Lower is closer.</p>
</article>
<article>
<h3>Retention counts</h3>
<p>Mode counts in fixed internal high- and very-high-retention bands. The public record releases counts, not raw modes or threshold mechanics.</p>
</article>
<article>
<h3>Signed phase counts</h3>
<p>Final-checkpoint mode populations grouped by negative or positive rotation direction.</p>
</article>
<article>
<h3>Prompt scores</h3>
<p>Normalized internal descriptors for comparing probes within this analysis. They are not standardized external evaluation scores.</p>
</article>
</div>
<h3>What would weaken this interpretation?</h3>
<ul>
<li>A checkpoint audit that fails to reproduce the released rounded loss, validation, or population summaries.</li>
<li>EDMD proximity that disappears under held-out trajectories, repeat runs, or appropriate control comparisons.</li>
<li>Signed phase balance that collapses on replication or proves unrelated to measurable behavior.</li>
<li>A targeted long-context run that increases neither long-memory mode population nor held-out long-context performance.</li>
</ul>
<p>
Stronger support would require repeat seeds, controlled ablations, external
evaluation, and a reproducible public demonstration of the analysis method.
Those tests are validation targets, not completed claims.
</p>
</section>
<section>
<h2>Observed Analysis Record</h2>
<p>
The v3 short-context run was examined from several independent angles. The
purpose was not only to ask whether loss improved, but whether the model's
internal dynamics could be measured and used to make a concrete next
training decision.
</p>
<div class="finding-grid">
<article>
<span class="finding-kicker">Training</span>
<h3>Loss moved while validation stayed coherent</h3>
<p>
Best observed loss moved from about 4.03 at 30k steps to about 3.16 at
250k steps. The final validation probe averaged about 3.27 across twenty
batches.
</p>
</article>
<article>
<span class="finding-kicker">Concordance</span>
<h3>Trajectory dynamics stayed close to learned structure</h3>
<p>
Mean nearest EDMD distance improved from about 0.0113 at 30k to about
0.0078 at the final checkpoint, with the median staying near 0.0071 late
in training.
</p>
</article>
<article>
<span class="finding-kicker">Phase</span>
<h3>Signed phase did not collapse</h3>
<p>
The final checkpoint retained both negative and positive rotation
populations. Public counts are rounded in interpretation, but the two
directions remained broadly balanced.
</p>
</article>
<article>
<span class="finding-kicker">Prompt use</span>
<h3>Prompts produced different spectral fingerprints</h3>
<p>
Math, code, story, and long-memory probes produced distinct aggregate
profiles across amplitude, rotation, spread, slow-mode use, and
coherence.
</p>
</article>
</div>
</section>
<section>
<h2>EDMD Concordance Details</h2>
<p>
EDMD-style analysis was used as an external measurement of trajectory
behavior. The important observation is that the measured trajectory structure
stayed close to the learned spectral structure after the early checkpoint.
</p>
<div class="evidence-table-wrap note-table">
<table class="evidence-table">
<thead>
<tr>
<th>Checkpoint</th>
<th>Mean distance</th>
<th>Median distance</th>
<th>90th percentile</th>
<th>Public read</th>
</tr>
</thead>
<tbody>
<tr>
<td>30k</td>
<td>0.0113</td>
<td>0.0103</td>
<td>0.0187</td>
<td>early alignment</td>
</tr>
<tr>
<td>102k</td>
<td>0.0079</td>
<td>0.0069</td>
<td>0.0138</td>
<td>alignment tightened</td>
</tr>
<tr>
<td>204k</td>
<td>0.0077</td>
<td>0.0071</td>
<td>0.0131</td>
<td>stable agreement</td>
</tr>
<tr>
<td>250k</td>
<td>0.0078</td>
<td>0.0071</td>
<td>0.0134</td>
<td>agreement held</td>
</tr>
</tbody>
</table>
</div>
<p>
The public record does not include the analysis transform or model internals.
It reports the observed relationship: measured dynamics and learned spectral
structure were close enough to compare over training.
</p>
</section>
<section>
<h2>Mode Population Observations</h2>
<p>
The mode population did not behave like an unstructured black box. The final
checkpoint showed a strong middle-retention population, active rotation, and
a smaller slow-mode population. That finding matters because it explains why
the next experiment should focus on long-context memory rather than simply
continuing to celebrate short-context loss.
</p>
<div class="evidence-table-wrap note-table">
<table class="evidence-table">
<thead>
<tr>
<th>Measure</th>
<th>30k</th>
<th>204k</th>
<th>250k</th>
<th>Interpretation</th>
</tr>
</thead>
<tbody>
<tr>
<td>High-retention modes</td>
<td>2</td>
<td>2</td>
<td>6</td>
<td>small but measurable slow population</td>
</tr>
<tr>
<td>Very-high-retention modes</td>
<td>0</td>
<td>1</td>
<td>1</td>
<td>not enough long-memory mass yet</td>
</tr>
<tr>
<td>High-rotation modes</td>
<td>470</td>
<td>733</td>
<td>638</td>
<td>phase-active dynamics remained present</td>
</tr>
<tr>
<td>Max measured half-life</td>
<td>12.7</td>
<td>16.8</td>
<td>16.2</td>
<td>short-context behavior dominated</td>
</tr>
</tbody>
</table>
</div>
</section>
<section>
<h2>Prompt Manifold Observations</h2>
<p>
Prompt-manifold analysis tested whether different prompts left different
spectral signatures. The public version reports only aggregate descriptors,
not prompt activations, mode IDs, or raw token traces.
</p>
<div class="evidence-table-wrap note-table">
<table class="evidence-table">
<thead>
<tr>
<th>Probe</th>
<th>Tokens</th>
<th>Amplitude</th>
<th>Rotation</th>
<th>Spread</th>
<th>Public read</th>
</tr>
</thead>
<tbody>
<tr>
<td>math explanation</td>
<td>22</td>
<td>0.83</td>
<td>0.97</td>
<td>0.88</td>
<td>broad analytical trace</td>
</tr>
<tr>
<td>code probe</td>
<td>19</td>
<td>0.87</td>
<td>0.98</td>
<td>0.86</td>
<td>high-amplitude structured trace</td>
</tr>
<tr>
<td>story probe</td>
<td>20</td>
<td>0.82</td>
<td>0.89</td>
<td>0.85</td>
<td>lower-rotation narrative trace</td>
</tr>
<tr>
<td>long-memory probe</td>
<td>40</td>
<td>0.87</td>
<td>0.98</td>
<td>0.88</td>
<td>strong activity, limited true long-memory use</td>
</tr>
</tbody>
</table>
</div>
<p>
The useful conclusion is not that one prompt "wins." The useful conclusion is
that prompts can be compared as paths through spectral behavior, giving a
practical way to inspect generation beyond reading text output alone.
</p>
</section>
<section>
<h2>Interpretation</h2>
<p>
The v3 short-context run provides evidence that the model can train stably
while preserving signed spectral structure. The EDMD agreement is important
because it gives an independent check against the model's learned internal
organization. The prompt-manifold work is important because it turns model
behavior into a path through mode neighborhoods rather than a single opaque
output.
</p>
<p>
The most useful result is not that any one metric is final. The useful result
is that the model produced measurable internal structure that can guide the
next experiment.
</p>
</section>
<section>
<h2>Limitation Found</h2>
<p>
The short-context run did not produce a large population of very long-memory
modes. That is not hidden in this note, because it is one of the strongest
reasons the analysis matters. The diagnostics did not merely celebrate a
result; they identified what the next run needs to target.
</p>
<p>
The next research direction is a long-context version that preserves the
useful short-context behavior while deliberately training stronger long-memory
dynamics.
</p>
</section>
<section>
<h2>Skeptic FAQ</h2>
<div class="faq-list">
<details open>
<summary>Is this just a visualization of a model, not evidence?</summary>
<p>
The public atlas visualization is synthetic, but the evidence tables are
rounded summaries from the analyzed v3 short-context run. The distinction
is intentional: public visuals explain the method; public evidence
reports aggregate observations from the real run.
</p>
</details>
<details>
<summary>Are you claiming state-of-the-art language modeling performance?</summary>
<p>
No. The public claim is not benchmark dominance. The claim is that a
stable linear spectral state-space language model was trained and that
its internal dynamics were observable enough to compare across
checkpoints and guide the next experiment.
</p>
</details>
<details>
<summary>Is the model really linear?</summary>
<p>
The full system is not linear end to end. The public description is
precise: Project SLSSM-01 has a stable linear recurrent dynamics core,
surrounded by learned nonlinear token drive, gating, and readout. The
linear core is what makes the mode analysis meaningful.
</p>
</details>
<details>
<summary>Could the EDMD agreement be accidental?</summary>
<p>
It could be overstated if treated alone. That is why the note pairs it
with other evidence: loss movement, validation probes, signed phase
preservation, mode population tracking, and prompt-manifold behavior.
The stronger claim is convergence across multiple diagnostics, not one
isolated plot.
</p>
</details>
<details>
<summary>Why not publish the code and weights?</summary>
<p>
The public release is the evidence, vocabulary, and analysis framing.
Implementation, weights, initialization mechanics, kernels, and the
training recipe are not part of this release.
</p>
</details>
<details>
<summary>Why use the name Project SLSSM-01?</summary>
<p>
It is a provisional public codename based on the model class: stable
linear spectral state-space model. It avoids relying on a potentially
crowded public mark while the research identity, legal naming, and any
future company or product branding remain unresolved.
</p>
</details>
<details>
<summary>What would count as stronger proof?</summary>
<p>
Stronger public proof would include controlled ablations, external
evaluation, reproducible toy demonstrations of the analysis method, and
a long-context run showing that the identified memory limitation can be
deliberately improved.
</p>
<p>
One item on that list now exists: a
<a href="https://github.com/trinity-research/pythia-edmd-demo">reproducible EDMD demonstration
on public Pythia-14m checkpoints</a>, with an author-reported,
pre-specified protocol with dated post-run amendments,
shuffled-pairing and persistence controls, and pinned model commits.
It validates the measurement method on a public model; it does not
validate this model's claims.
</p>
</details>
</div>
</section>
<section class="disclosure-box">
<h2>Disclosure Boundary</h2>
<p>
This public note does not disclose:
</p>
<ul>
<li>source code</li>
<li>model weights or checkpoints</li>
<li>training notebooks</li>
<li>architecture equations</li>
<li>initialization mechanics</li>
<li>optimizer internals</li>
<li>raw activations or raw mode tables</li>
<li>kernel or implementation details</li>
</ul>
</section>
<section>
<h2>How To Cite</h2>
<p>
Suggested citation for public discussion:
</p>
<pre class="citation-block">Hunt, F., & Hunt, J. (2026). Observable Spectral Dynamics in a Recurrent Sequence Model. Project SLSSM-01. <span data-page-url>research-note.html</span></pre>
<p>
BibTeX:
</p>
<pre class="citation-block">@misc{hunt2026observable,
title = {Observable Spectral Dynamics in a Recurrent Sequence Model},
author = {Hunt, Freddie and Hunt, Joshua},
year = {2026},
url = {<span data-page-url>research-note.html</span>},
note = {Public research note, version 0.1}
}</pre>
</section>
<footer class="note-footer">
<a class="button secondary" href="slssm-01.html">Back to public dossier</a>
<a class="button secondary" href="mailto:contact@trinity-research.com">contact@trinity-research.com</a>
</footer>
</article>
</main>
<script>
const citationUrl = window.location.href.split(/[?#]/)[0];
document.querySelectorAll("[data-page-url]").forEach((element) => {
element.textContent = citationUrl;
});
</script>
</body>
</html>