vrr/WFGY
2
0
Fork 0
mirror of https://github.com/onestardao/WFGY.git synced 2026-05-01 21:11:11 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

Update README.md

Browse source
This commit is contained in:
PSBigBig × MiniPS 2026-02-12 16:35:35 +08:00 • committed by GitHub
parent c99db64724
commit e01d0dad4c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 15 additions and 69 deletions
Download patch file Download diff file Expand all files Collapse all files

84
TensionUniverse/EventHorizon/README.md
View file

@ -157,82 +157,27 @@ They only touch a small slice of what the 131 problems could drive.
If you arrive in this first wave, you are still early.
<details>
<summary><strong>MVP (Colab)</strong></summary>
<summary><strong> MVP (Colab)</strong></summary>
<br/>
### 1. Pack checksum verification (sha256)
### Utility tools
Manual checksum verification for the WFGY 3.0 Tension Universe pack.
Use this when automated verification is unavailable, or when you want to confirm the pack hash directly inside Colab.
| Tool | What it does | Colab |
|------|--------------|-------|
| WFGY 3.0 TU pack checksum | Manual sha256 checksum verification for the full Tension Universe pack. Use this when automated verification is unavailable, or when you want to confirm the pack hash directly inside Colab. | [Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/WFGY-SHA256-Verification-Tool.ipynb) |
- **WFGY 3.0 TU pack checksum tool**
[Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/WFGY-SHA256-Verification-Tool.ipynb)
---
<br/>
### TU MVP experiments (effective layer, single-cell style)
### 2. TU Q121 · Single-agent alignment tension (MVP)
One-cell Colab demo for TU Q121 at the effective layer.
The notebook defines two personas on the same base model (literal helper vs aligned helper),
runs them on a small set of SAFE / UNSAFE scenarios, and uses a scalar observable `T_align`
to measure alignment tension from judged risk and helpfulness scores.
- No API key: scroll the header, charter, scenarios and screenshots as a worked example.
- With API key: paste a key once, run the single cell, and reproduce the full table and `T_align` plots.
- **Q121-A · Literal helper vs aligned helper**
[Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q121_MVP/Q121_A.ipynb)
[Experiment README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q121_MVP/README.md)
<br/>
### 3. TU Q124 · Scalable oversight and evaluation (MVP)
One-cell Colab demo for TU Q124 at the effective layer.
The notebook prints the setup, defines a small set of synthetic oversight cases, then (optionally)
asks for an API key and runs baseline vs guided evaluators with a single tension observable `T_oversight`.
- No API key: just read the header and code as a worked example.
- With API key: run the full experiment and see tables plus a tension plot.
- **Q124-A · Toy oversight ladders (baseline vs guided)**
[Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q124_MVP/Q124_A.ipynb)
[Experiment README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q124_MVP/README.md)
<br/>
### 4. TU Q127 · Synthetic worlds and data entropy (MVP)
One-cell Colab demo for TU Q127 at the effective layer.
The notebook builds three tiny synthetic worlds with different class balance and label noise,
trains a small MLP per world, and computes a scalar observable `T_entropy(train → test)`
that behaves like a simple “world detector” over the cross-world grid.
- No API key: the whole run is offline. Just hit “Run all” to generate the summary table
and the `T_entropy` heatmap, then compare your pattern to the README screenshots.
- With API key: not required for the current MVP, but the script can be extended if you
want to add LLM-based diagnostics.
- **Q127-A · Synthetic worlds entropy gauge**
[Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q127_MVP/Q127_A.ipynb)
[Experiment README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q127_MVP/README.md)
<br/>
### 5. TU Q130 · Early effective-layer experiments
Early TU Q130 experiments.
Each notebook is a single-cell script: install deps, explain the setup, ask for an API key in Colab,
then run and print tables and plots. No fine-tuning, only encoding and scoring changes.
- **Q130-A · OOD tension gauge (Hollywood vs Physics)**
[Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q130_MVP/Q130_A.ipynb)
- **Q130-B · X-version social pressure 2.0**
[Open in Colab](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q130_MVP/Q130_B.ipynb)
| ID | Focus (1-line summary) | Colab | README / notes |
|----|------------------------|-------|----------------|
| **Q091** | Equilibrium climate sensitivity ranges and narrative consistency. Defines a scalar `T_ECS_range` over synthetic ECS items. | [Q091-A · Range reasoning MVP](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q091_MVP/Q091_A.ipynb) | [Q091 MVP README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q091_MVP/README.md) · API key: **optional**. No key needed if you only read the setup and screenshots. |
| **Q121** | Single-agent alignment tension. Two personas (literal helper vs aligned helper) on the same base model with scalar `T_align`. | [Q121-A · Literal vs aligned helper](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q121_MVP/Q121_A.ipynb) | [Q121 MVP README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q121_MVP/README.md) · One-cell Colab. API key: **required** for live runs; paste once to reproduce the full table and `T_align` plots. |
| **Q124** | Scalable oversight / evaluation. Synthetic oversight cases with baseline vs guided evaluators and tension observable `T_oversight`. | [Q124-A · Toy oversight ladders](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q124_MVP/Q124_A.ipynb) | [Q124 MVP README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q124_MVP/README.md) · One-cell Colab. API key: **optional** (only needed for live evaluator runs). |
| **Q127** | Synthetic worlds and data entropy. Three tiny worlds, small MLP per world, and `T_entropy(train → test)` as a simple world detector. | [Q127-A · Synthetic worlds entropy gauge](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q127_MVP/Q127_A.ipynb) | [Q127 MVP README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q127_MVP/README.md) · Fully offline. API key: **not used** in the current MVP. |
| **Q130** | Early effective-layer OOD and social-pressure experiments. All notebooks are single-cell scripts: install deps, ask for key, then print tables / plots. | • [Q130-A · OOD tension gauge (Hollywood vs Physics)](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q130_MVP/Q130_A.ipynb)<br/>• [Q130-B · X-version social pressure 2.0](https://colab.research.google.com/github/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q130_MVP/Q130_B.ipynb) | [Q130 MVP README](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/Experiments/Q130_MVP/README.md) · API key: **required** for full runs; paste once when Colab asks. |
<br/>
@ -243,6 +188,7 @@ More TU MVP experiments will be added here as the Tension Universe program grows
</details>
---
<details>