|
1 | | -Penguin Distortion Tester (v560L • CKMM • Disclose) |
| 1 | +# Penguin Distortion Tester (v560L • CKMM • Disclose) |
2 | 2 |
|
3 | | -[](https://github.com/stuck-inadream/penguin-tester/actions) |
4 | | -[](https://huggingface.co/spaces/stuck-inadream/penguin-distortion-tester) |
| 3 | +[](https://github.com/stuck-inadream/penguin-tester/actions) |
| 4 | +[](https://huggingface.co/spaces/stuck-inadream/penguin-distortion-tester) |
5 | 5 | [](LICENSE) |
6 | 6 |
|
| 7 | +**What this is.** A minimal, public-safe harness to *audit symbolic distortion* on a toy “Penguin” task, show SAT/UNSAT behavior (via **Z3**), and report **CKMM-L** context metrics: |
7 | 8 |
|
8 | | -Overview |
9 | | -The Penguin Distortion Tester is a minimal, public-safe harness to audit symbolic distortion on a toy “Penguin” task. |
10 | | -It demonstrates sat/unsat behavior (via Z3) and reports CKMM-L context metrics across five layers: |
11 | | -Fuel (Energy–Resource): wall time, peak KiB (tracemalloc), and human-attention ticks |
| 9 | +- **Fuel (Energy–Resource):** wall time + peak KiB (tracemalloc) + human-attention ticks |
| 10 | +- **Temporal Coherence:** replay stability ratio |
| 11 | +- **Relational Field:** council agreement vs. drift (majority vote placeholder) |
| 12 | +- **Ethical Constraints:** regex penalties + recovery flag |
| 13 | +- **Embodiment:** device/substrate capture with validated hint |
12 | 14 |
|
| 15 | +This repo is a **disclose build**: it lists exactly what a prior reviewer saw — **no more, no less** — and keeps proprietary weighting/heuristics private. |
13 | 16 |
|
14 | | -Temporal Coherence: replay stability ratio |
15 | | -Relational Field: council agreement vs. drift (majority vote placeholder) |
16 | | -Ethical Constraints: regex penalties + recovery flag |
17 | | -Embodiment: device/substrate capture with validated hint |
| 17 | +--- |
18 | 18 |
|
19 | | -This repo is a disclose build: it lists exactly what a prior reviewer saw — no more, no less — and keeps proprietary weighting/heuristics private. |
| 19 | +## Quickstart |
20 | 20 |
|
21 | | -Quickstart |
| 21 | +```bash |
22 | 22 | python3.11 -m venv .venv |
23 | 23 | source .venv/bin/activate |
24 | 24 | python -m pip install --upgrade pip |
25 | 25 | pip install -r requirements.txt |
26 | 26 |
|
27 | | -# Run: summarize UI, 100k iterations, choose a port |
28 | | -env PYTHONUNBUFFERED=1 python gradio_demo.py --summary --ckmm-iterations 100000 --server_port 7860 |
| 27 | +# Run: summary UI, 100k iterations, explicit port |
| 28 | +PYTHONUNBUFFERED=1 python gradio_demo.py --summary --ckmm-iterations 100000 --server_port 7860 |
| 29 | +# Open http://127.0.0.1:7860 (use --share for a temporary public link) |
29 | 30 |
|
30 | | -Open http://127.0.0.1:7860 in your browser. |
31 | | - Use --share for a temporary public link. |
32 | 31 | ⚠️ Heads-up: --share exposes a public URL. Treat it as untrusted. Do not upload sensitive data. |
33 | 32 |
|
34 | | -Features (Public-Safe) |
| 33 | +Features (public-safe) |
35 | 34 | “Penguin” symbolic check (baseline sat, forced unsat via contradiction trigger) |
36 | | -Distortion score toy formula: modal × diversity (with toy numbers disclosed below) |
| 35 | +Distortion score toy formula: modal × diversity (toy numbers disclosed below) |
37 | 36 | CKMM-L interfaces only (no proprietary internals) |
38 | 37 | Configurable workload and thresholds: |
39 | | ---ckmm-iterations N (CPU loop, default documented) |
| 38 | +--ckmm-iterations N (CPU loop; default documented below) |
40 | 39 | --ckmm-thresholds '{"fuel":{"max_time_s":2.0}}' (JSON Schema-validated) |
41 | | ---ckmm-verbose (pretty JSON output) |
42 | | - |
43 | | - |
44 | | -Logging: rotating errors.log + console log level |
| 40 | +--ckmm-verbose (pretty JSON) |
| 41 | +Logging: rotating errors.log + console level via --log-level |
45 | 42 | CI: pytest via GitHub Actions |
46 | 43 |
|
47 | | - |
48 | | -Public Disclosure Parity Appendix |
49 | | -Score formula: modal × diversity; diversity = avg edit distance; toy values 0.33 × 2 = 0.66 |
50 | | -Load fractions: 1.0 / 0.6 / 0.7, divergence = 0.3 |
| 44 | +Public Disclosure Parity (Seen-by-Reviewer Appendix) |
| 45 | +What’s explicitly disclosed here was previously visible to a reviewer: |
| 46 | +Score formula modal × diversity; diversity = avg edit distance; toy values 0.33 × 2 = 0.66 |
| 47 | +Load fractions 1.0 / 0.6 / 0.7, divergence = 0.3 |
51 | 48 | HSIL policy names: prefer_specific_over_general, safety_first, audit_bias |
52 | 49 | Z3 baseline sat; forced contradiction → unsat; heatmap.png modal plot |
53 | 50 | Gradio flags (server name/port), log level, rotating logs |
54 | 51 | CI + tests, CKMM-L interfaces (fuel / temporal / council / ethics / embodiment) |
55 | | - |
56 | 52 | Not disclosed: proprietary weighting/aggregation schemes, true diversity/council math beyond placeholders, production energy models, or internal heuristics. |
57 | 53 |
|
58 | 54 | FAQ |
59 | 55 | Is this just a toy? |
60 | | - Yes — by design. It’s a teaching harness: easy to run, safe to share, useful for audits. |
| 56 | +Yes — by design. It’s a teaching harness that’s easy to run, safe to share, and useful for audits. |
61 | 57 | Can I tune workloads? |
62 | | - Yes — use --ckmm-iterations (e.g., 5_000_000 for measurable fuel) and JSON thresholds. |
| 58 | +Yes — use --ckmm-iterations (e.g., 5_000_000 for measurable fuel) and thresholds via --ckmm-thresholds JSON. |
63 | 59 | Does --share upload my data? |
64 | | - No uploads unless you explicitly make them. --share only exposes a temporary Gradio URL. Treat it as public. |
| 60 | +No uploads unless you explicitly make them; --share only exposes a temporary Gradio URL. Treat it as public. |
65 | 61 |
|
66 | 62 | Development & Tests |
67 | 63 | pip install -r requirements.txt |
68 | 64 | pytest -q |
69 | 65 |
|
70 | | -pip install -r requirements.txt |
71 | | -pytest -q |
72 | | - |
73 | | -MIT © stuck-inadream |
74 | | - |
75 | 66 | Roadmap & Contributions |
76 | 67 | Add screenshot(s) of the UI (docs/screenshot.png) |
77 | | -Pin dependencies in requirements.txt for CI stability |
78 | | -CI improvements: multi-stage (lint then test) with pip cache |
| 68 | +Pin dependencies in requirements.txt for CI stability (already pinned) |
| 69 | +CI improvements: multi-stage (lint then test) with pip cache (already caching pip) |
79 | 70 | Optional: package-ready entry point (pyproject.toml) for pipx run penguin-tester |
| 71 | +Contributions welcome! See CONTRIBUTING.md. |
80 | 72 |
|
81 | | -Contributions welcome! See CONTRIBUTING.md (to be added). |
| 73 | +MIT © stuck-inadream |
82 | 74 |
|
83 | 75 | Screenshot |
84 | 76 | <img width="1234" height="750" alt="Screenshot 2025-09-21 at 9 28 02 AM" src="https://github.com/user-attachments/assets/3d0d0b21-539d-4f63-90b1-67b719130949" /> |
0 commit comments