Skip to content

Commit ce7d41f

Browse files
infinityabundanceinfinityabundance
committed
Prepare dsfb-gpu crate for publish
Add the copied dsfb-gpu crate as a standalone publish-ready crate. Retarget Colab packaging and metadata to the crate path. Verified locally with workspace checks, Colab packaging smoke, and cargo package.
1 parent e185db8 commit ce7d41f

767 files changed

Lines changed: 286862 additions & 0 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎crates/dsfb-gpu/.gitattributes‎

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,22 @@
1+
* text=auto eol=lf
2+
*.json text eol=lf
3+
*.toml text eol=lf
4+
*.rs text eol=lf
5+
*.cu text eol=lf
6+
*.cuh text eol=lf
7+
*.sh text eol=lf
8+
*.md text eol=lf
9+
*.tex text eol=lf
10+
11+
*.png binary
12+
*.pdf binary
13+
*.zip binary
14+
*.tar.gz binary
15+
16+
# SHA-256-pinned data fixtures + upstream archives MUST NOT have
17+
# line endings normalized by git. CRLF/LF mangling would break the
18+
# byte-pin chain (data/fixtures/MANIFEST.toml -> SHA-256 -> recipe
19+
# -> projected TSV -> audit case_file_hash).
20+
data/upstream/** binary
21+
data/fixtures/*.tsv binary
22+
data/recipes/*.py text eol=lf

‎crates/dsfb-gpu/.githooks/pre-commit‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
../scripts/pre_commit.sh

‎crates/dsfb-gpu/.gitignore‎

Lines changed: 378 additions & 0 deletions
Large diffs are not rendered by default.

‎crates/dsfb-gpu/.zenodo.json‎

Lines changed: 38 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,38 @@
1+
{
2+
"title": "DSFB-GPU - Clear-Box Pure Deterministic Inference CUDA Acceleration for Replayable Trace-Event Verdicts - A Prior-Art Architecture for non-probabilistic, non-stochastic, non-weighted, GPU-Accelerated Residual Signs, Detector Motifs, Bank-Governed Fusion, and Byte-Exact Case Files Without Probabilistic Models",
3+
"description": "DSFB-GPU-Debug is a bounded prior-art proof of clear-box pure deterministic inference accelerated on GPU hardware. The pipeline maps the seven canonical stages of DSFB debug inference (residual extraction, drift / slew sign construction, detector motif scoring, consensus grid formation, candidate collapse, bank-governed episode emission, and replayable case-file assembly) onto fixed-point CUDA kernels and a Rust-resident heuristics bank. The GPU accelerates only the evidence-production stages; the bank retains semantic authority. No neural network is used, no probability distribution is consulted, no learned weight is invoked, no stochastic sample is drawn. The result is not a prediction; it is a replayable verdict case file whose hash chain anchors every intermediate artifact back to the input catalog. The inference mode is named endoduction: deterministic adjudication of internal evidence-field relations into a replayable structural verdict, distinct from induction, deduction, and abduction. The package ships a 20-dataset S-REAL audit gauntlet (316 admitted episodes, byte-identical within-run replay), a 30-fixture saturation classification (10 saturation-class reaching 16.16 - 26.66 GB/s wide-arena throughput, 20 launch-bound honestly reported), and a public Colab replay surface for cross-hardware byte-identity verification.",
4+
"version": "1.0",
5+
"publication_date": "2026-05-22",
6+
"upload_type": "software",
7+
"access_right": "open",
8+
"license": "Apache-2.0",
9+
"doi": "10.5281/zenodo.20338027",
10+
"creators": [
11+
{
12+
"name": "de Beer, Riaan",
13+
"affiliation": "Invariant Forge LLC",
14+
"orcid": "0009-0006-1155-027X"
15+
}
16+
],
17+
"keywords": [
18+
"deterministic inference",
19+
"pure deterministic inference",
20+
"non-probabilistic",
21+
"non-stochastic",
22+
"non-weighted",
23+
"CUDA",
24+
"evidence court",
25+
"endoduction",
26+
"densor",
27+
"tekmeric inference",
28+
"replayable case file",
29+
"semantic non-bypass",
30+
"fixed-point Q16.16",
31+
"byte-exact replay",
32+
"prior art",
33+
"clear-box artificial inference"
34+
],
35+
"language": "eng",
36+
"communities": [],
37+
"notes": "v1.0 sealed to Zenodo on 2026-05-22 under DOI 10.5281/zenodo.20338027 (https://doi.org/10.5281/zenodo.20338027). The Zenodo deposit is the primary §102 public prior-art timestamp anchor; the human-readable timestamp receipt with cross-verification SHA-256 fingerprints lives at TIMESTAMP_RECEIPT.md."
38+
}

‎crates/dsfb-gpu/ARTIFACT_MANIFEST.v1.toml‎

Lines changed: 99 additions & 0 deletions
Large diffs are not rendered by default.

‎crates/dsfb-gpu/CITATION.cff‎

Lines changed: 68 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,68 @@
1+
cff-version: 1.2.0
2+
message: "If you use DSFB-GPU, please cite the v1.0 Zenodo software record (DOI 10.5281/zenodo.20338027)."
3+
title: "DSFB-GPU - Clear-Box Pure Deterministic Inference CUDA Acceleration for Replayable Trace-Event Verdicts - A Prior-Art Architecture for non-probabilistic, non-stochastic, non-weighted, GPU-Accelerated Residual Signs, Detector Motifs, Bank-Governed Fusion, and Byte-Exact Case Files Without Probabilistic Models"
4+
type: software
5+
version: "1.0"
6+
date-released: "2026-05-22"
7+
license: Apache-2.0
8+
doi: "10.5281/zenodo.20338027"
9+
identifiers:
10+
- type: doi
11+
value: "10.5281/zenodo.20338027"
12+
description: "Zenodo concept DOI for the v1.0 deposit of the DSFB-GPU-Debug prior-art paper."
13+
authors:
14+
- family-names: "de Beer"
15+
given-names: "Riaan"
16+
orcid: "https://orcid.org/0009-0006-1155-027X"
17+
affiliation: "Invariant Forge LLC"
18+
email: "riaan@invariantforge.net"
19+
repository-code: "https://github.com/infinityabundance/dsfb"
20+
url: "https://doi.org/10.5281/zenodo.20338027"
21+
abstract: >
22+
DSFB-GPU is a bounded prior-art proof of clear-box
23+
deterministic inference accelerated on GPU hardware. The pipeline
24+
maps the seven canonical stages of DSFB debug inference --- residual
25+
extraction, drift / slew sign construction, detector motif scoring,
26+
consensus grid formation, candidate collapse, bank-governed episode
27+
emission, and replayable case-file assembly --- onto fixed-point
28+
CUDA kernels and a Rust-resident heuristics bank. The GPU
29+
accelerates only the evidence-production stages; the bank retains
30+
semantic authority. No neural network is used, no probability
31+
distribution is consulted, no learned weight is invoked, no
32+
stochastic sample is drawn. The result is not a prediction; it is a
33+
replayable verdict case file whose hash chain anchors every
34+
intermediate artifact back to the input catalog. The inference mode
35+
is named endoduction --- deterministic adjudication of internal
36+
evidence-field relations into a replayable structural verdict,
37+
distinct from induction, deduction, and abduction.
38+
keywords:
39+
- "deterministic inference"
40+
- "pure deterministic inference"
41+
- "non-probabilistic"
42+
- "non-stochastic"
43+
- "non-weighted"
44+
- "CUDA"
45+
- "evidence court"
46+
- "endoduction"
47+
- "densor"
48+
- "tekmeric inference"
49+
- "replayable case file"
50+
- "semantic non-bypass"
51+
- "fixed-point Q16.16"
52+
- "byte-exact replay"
53+
- "prior art"
54+
- "clear-box artificial inference"
55+
preferred-citation:
56+
type: software
57+
title: "DSFB-GPU - Clear-Box Pure Deterministic Inference CUDA Acceleration for Replayable Trace-Event Verdicts - A Prior-Art Architecture for non-probabilistic, non-stochastic, non-weighted, GPU-Accelerated Residual Signs, Detector Motifs, Bank-Governed Fusion, and Byte-Exact Case Files Without Probabilistic Models"
58+
authors:
59+
- family-names: "de Beer"
60+
given-names: "Riaan"
61+
orcid: "https://orcid.org/0009-0006-1155-027X"
62+
affiliation: "Invariant Forge LLC"
63+
year: 2026
64+
version: "1.0"
65+
publisher:
66+
name: "Zenodo"
67+
doi: "10.5281/zenodo.20338027"
68+
url: "https://doi.org/10.5281/zenodo.20338027"

0 commit comments

Comments
 (0)