File size: 2,545 Bytes
de15a9f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: mit
library_name: pytorch
tags:
  - verified-units
  - reed-solomon
  - archival
  - preservation
  - self-healing
---

# neural-cd-preserve — self-healing optical-disc archives

**Repositories:** [GitHub](https://github.com/quzi93/neural-cd-preserve) · [🤗 HuggingFace](https://huggingface.co/NeuralVerified/neural-cd-preserve)

Scan an optical disc into a self-healing `.pt` archive that **detects and repairs
bit-rot** and reconstructs the disc even from a damaged copy. Built on the same
neural-verified Reed-Solomon core as
[neural-storage](https://huggingface.co/NeuralVerified/neural-storage): each unique
chunk is RS-split into shards, every shard carries its own SHA-256, so silent
corruption is detected, marked as an erasure, and repaired from survivors (as long
as ≥ *k* of *n* shards remain).

> **Honest by design:** RS self-healing protects the *archive* against future
> bit-rot. It cannot recover disc sectors that were already unreadable at scan
> time — those are recorded, and re-reads are merged. No entropy is beaten.

## Use

```bash
pip install torch
python step_cd.py                                   # full self-healing demo

python cli.py archive \\.\D: mydisc.pt --label MY_CD   # drive or .iso
python cli.py verify  mydisc.pt
python cli.py heal    mydisc.pt
python cli.py restore mydisc.pt out.iso
```

The demo injects 149 silent bit-flips; `verify` detects all, `heal` repairs all,
`restore` returns bit-exact; corrupting beyond RS capacity is flagged **LOST**,
never silently wrong.

Weights: `GF256.pt` (verified GF(256) LOG/EXP).

**Create your own verified unit** (template: `cdpreserve/gf256.py`): write the
exact golden finite function → enumerate the domain (decompose big/linear ones
into bit/byte slices, see `cdpreserve/rs.py`) → `common.train` → `common.verify`
must be bit-exact on 100% of inputs → compose.

## Citation

```bibtex
@misc{byrne2026neuralcdpreserve,
  title  = {neural-cd-preserve: Self-Healing Optical-Disc Archival with Verified Erasure Coding},
  author = {Byrne, Dean (Quazim0t0)},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/NeuralVerified/neural-cd-preserve}}
}
```

**Dean Byrne (Quazim0t0)** · 2026


---

<!-- neuralverified-relocation-note -->
> **Now hosted by [NeuralVerified](https://huggingface.co/NeuralVerified).**
>
> This repo was moved into the NeuralVerified organization to help organize my profile.
> Originally published at [`Quazim0t0/neural-cd-preserve`](https://huggingface.co/Quazim0t0/neural-cd-preserve).