Edition README
Large Signal Model software
README.md
1 # Komorebi Large Signal Model
2
3 ## Projector-Resolved Large Algebraic Signal Model · 2.3.0
4
5 PR-LASM is an executable, non-LLM algebraic signal model. It separates a
6 reconstructible coordinate archive from statistical state, learns coefficients
7 and weights over a finite typed operator grammar, and provides explicit memory
8 and planning on controlled interactive systems.
9
10 This release also exposes a standalone real-valued signal interface with
11 polyphase transfer blocks, causal streaming FIR, complete packet refinement,
12 and versioned source-signal storage. These signal operations do not require an
13 ARC runtime or a hosted model.
14
15 **Status:** Beta / work in progress. The scoped core release has recorded
16 verification on Linux / Python 3.13.5. Original Komorebi code is available under
17 the MIT license in `LICENSE`; third-party terms remain as stated in
18 `THIRD_PARTY_NOTICES.md`. Research and performance claims retain their
19 recorded finite scope.
20
21 ## Start here
22
23 Install in a separate environment. Do not co-install the separate ARC3 agent
24 package `komorebi-arc3-offline`, which owns the same compatibility namespace.
25
26 ```powershell
27 py -3.13 -m venv .venv
28 .\.venv\Scripts\Activate.ps1
29 python -m pip install -e ".[test]"
30 python -m komorebi_pr_lasm self-check
31 python -m pytest -q -m "not local_suite"
32 ```
33
34 The dependency installation above may need internet access. An offline install
35 can use the supplied wheel with `--no-index --no-deps` when compatible NumPy and
36 SciPy are already installed. The package itself does not download a model.
37
38 ```powershell
39 python -m pip install --no-index --no-deps .\dist\komorebi_pr_lasm-2.3.0-py3-none-any.whl
40 python -m komorebi_pr_lasm self-check
41 ```
42
43 In the full handoff, `dist` is beside the `source` directory. Adjust the wheel
44 path to that location. The source-only archive does not contain `dist`.
45
46 ## A real signal, not a quantized image
47
48 ```python
49 import numpy as np
50 from komorebi_pr_lasm import analyze_signal, PolyphaseFIR, StreamingFIR
51
52 x = np.sin(np.arange(257) * 0.13)
53 analysis = analyze_signal(x, modulus=4)
54 print(analysis.reconstruction_residuals())
55
56 h = np.array([0.25, 0.5, 0.25])
57 bank = PolyphaseFIR(h, modulus=4)
58 y = bank.apply(x)
59 assert np.allclose(y, np.convolve(x, h))
60
61 stream = StreamingFIR(h)
62 y_stream = np.concatenate([
63 stream.process(x[:31]), stream.process(x[31:]), stream.flush()
64 ])
65 assert np.allclose(y, y_stream)
66 ```
67
68 For two-dimensional data, `MultiAtlasAnalyzer` retains the original shape.
69 Categorical color coordinates are present only for exact integers within the
70 configured palette. Continuous data is neither rounded nor clipped to create
71 an allegedly exact color representation.
72
73 ## What is learned
74
75 Atlas transforms, seeded projections and operator families are fixed mechanisms.
76 Bayesian regression learns numerical coefficients from observed transitions.
77 Validation data determines weights over the supplied operator candidates.
78 The final regressions use all supplied training records. Replay memory stores
79 observed transitions separately. The system does not autonomously invent an
80 unbounded operator grammar, and this archive contains no pretrained foundation
81 model checkpoint.
82
83 The standalone 1-D signal interface and the frame-based world model are related
84 components, not a claim that a general time-series forecasting or audio agent
85 has already been trained. The release guide describes that integration boundary.
86
87 ## Fresh verification
88
89 | Layer | This release |
90 |---|---:|
91 | Public core test cases | 184 passed |
92 | Core plus separately supplied local fixture tests | 192 passed |
93 | FIR / streaming stress combinations | 2,025 passed |
94 | Additional irregular-image cases | 160 passed |
95 | Additional 1-D reconstruction lengths | 160 passed |
96 | Controlled operator and atlas identification | 35 / 35 across five seeds |
97 | Deterministic synthetic interactive systems | 4 / 4 |
98 | Official ARC runtime or score | Not executed or claimed |
99
100 These are finite controls, not counts of independent theorems. All source tests
101 were run with warnings treated as errors. The fixture checks are static and
102 stubbed integration tests, not execution of official games. Windows and Python
103 3.12 CI are configured but were not executed in this container.
104
105 ## Worked examples
106
107 Eight examples are exercised in the fresh evidence bundle: exact image
108 reconstruction, adaptive refinement, continuous signals, explicit polyphase
109 FIR, streaming, source-signal archives, learned operator identification, and
110 the finite integer Euler bridge. Run them from the repository after installation.
111
112 ```powershell
113 python examples\continuous_signal.py
114 python examples\polyphase_taps.py
115 python examples\streaming_signal.py
116 python examples\operator_identification.py
117 python examples\euler_signal_bridge.py
118 ```
119
120 ## Commands
121
122 ```powershell
123 python -m komorebi_pr_lasm --version
124 python -m komorebi_pr_lasm self-check
125 python -m komorebi_pr_lasm analyze signal.npy --output results\analysis.json
126 python -m komorebi_pr_lasm benchmark --output results\benchmark
127 python scripts\stress_release.py --output results\stress
128 python scripts\build_offline.py --output dist
129 ```
130
131 `analyze` accepts real 1-D or 2-D `.npy` and comma-delimited `.csv` files.
132 Archives use a separate typed Python API. `save_signal_archive` saves source
133 coordinates and analysis settings, not fitted world-model or agent state.
134
135 ## Documentation and evidence
136
137 The current operational manual is [the release guide](docs/RELEASE_GUIDE.md).
138 The [migration notes](docs/migration-2.3.md) explain deliberate corrections to
139 previously accepted invalid inputs. The [audit](RELEASE_AUDIT.md),
140 [claim boundary](docs/claim-boundary.md), and
141 [reproduction instructions](REPRODUCIBILITY.md) distinguish fresh results from
142 historical evidence. The [historical capstone](paper/README.md) is retained
143 unchanged. It is not a newly revalidated manuscript of every 2.3.0 feature.
144
145 The full handoff preserves the recovered signal reconstruction and Euler
146 companions separately. The older ARC2 diagnostic package is a lineage reference,
147 not this release's ARC solver, and its score is not transferred to this model.
148
149 ## License
150
151 Copyright Komorebi P. The inherited [license](LICENSE) reserves all rights.
152 This source handoff is not yet an open-source license grant. Third-party
153 components retain their own licenses. Nothing has been remotely published.