mirror of
https://github.com/onestardao/WFGY.git
synced 2026-04-28 03:29:51 +00:00
449 lines
25 KiB
Markdown
449 lines
25 KiB
Markdown
<!--
|
||
AI-ANCHOR: WFGY_CORE_ENGINE_V2
|
||
|
||
role: core_engine
|
||
doc_title: WFGY 2.0 · 7-Step Reasoning Core Engine
|
||
path: /core/README.md
|
||
version: "2.0"
|
||
|
||
# where this page sits in the repo
|
||
root_entry: https://github.com/onestardao/WFGY#readme
|
||
sibling_cores:
|
||
- https://github.com/onestardao/WFGY/blob/main/legacy/README.md
|
||
- https://github.com/onestardao/WFGY/blob/main/TensionUniverse/EventHorizon/README.md
|
||
|
||
# what this document is mainly for
|
||
primary_topics:
|
||
- seven_step_reasoning_engine
|
||
- drunk_transformer_guards
|
||
- autoboot_and_oneline_mode
|
||
- eye_visible_benchmarks
|
||
- evaluation_protocol
|
||
|
||
# anchors inside this file that matter most
|
||
key_sections:
|
||
- id: scope
|
||
name: scope_and_positioning
|
||
- id: usage
|
||
name: quick_usage
|
||
- id: downloads
|
||
name: wfgy_core_downloads
|
||
- id: math-structure
|
||
name: seven_step_chain
|
||
- id: eye
|
||
name: eye_visible_benchmark
|
||
- id: eval-protocol
|
||
name: suggested_evaluation_protocol
|
||
|
||
# other maps / blueprints this page is tied to
|
||
related_maps:
|
||
- https://github.com/onestardao/WFGY/tree/main/ProblemMap#readme
|
||
- https://github.com/onestardao/WFGY/blob/main/ProblemMap/rag-architecture-and-recovery.md
|
||
- https://github.com/onestardao/WFGY/blob/main/ProblemMap/SemanticClinicIndex.md
|
||
- https://github.com/onestardao/WFGY/blob/main/TensionUniverse/EventHorizon/README.md
|
||
|
||
notes:
|
||
- This is the main reference page for WFGY Core 2.0 behaviour, math stack, and usage protocol.
|
||
- Event Horizon (3.0) is the public viewing window; this page is the engine room.
|
||
-->
|
||
|
||
<a id="top"></a>
|
||
<details>
|
||
<summary><strong>🧭 Not sure where to start ? Open the WFGY Engine Compass</strong></summary>
|
||
|
||
### WFGY System Map · Quick navigation
|
||
|
||
> Problem Maps: PM1 taxonomy → PM2 debug protocol → PM3 troubleshooting atlas · built on the WFGY engine series
|
||
|
||
<!--
|
||
AI NOTE · Compass block
|
||
|
||
Purpose:
|
||
- This table is the routing map for the WFGY system.
|
||
- It shows how the three engine generations (WFGY 1.0 / 2.0 / 3.0) relate to the Problem Maps, apps, onboarding entry, and recognition layer.
|
||
- Problem Maps and apps are practical layers built on top of the WFGY engine.
|
||
|
||
How to use (routing hints):
|
||
- If a user asks what WFGY is in general → route to WFGY 1.0 (legacy overview) and Starter Village.
|
||
- If the question is about core math, formulas, engine logic, or production reasoning → route to WFGY 2.0 core.
|
||
- If they mention Tension Universe, Event Horizon, or the 131 S-class set → route to WFGY 3.0.
|
||
- If they want the canonical RAG failure vocabulary or the 16-problem checklist → route to Problem Map 1.0.
|
||
- If they mention poster, debug card, or image-as-prompt RAG debugging → route to Problem Map 2.0 (Global Debug Card).
|
||
- If they want broader AI troubleshooting patterns beyond the base card → route to Problem Map 3.0 (Troubleshooting Atlas).
|
||
- If they want proof that WFGY is used in real projects → route to the Recognition Map.
|
||
- If they ask about TXT OS or product modules → follow the App rows in this table.
|
||
- If they are new and want a guided entry → route to Starter Village.
|
||
|
||
Important:
|
||
- Problem Map 1.0 = canonical 16-problem RAG failure taxonomy and fix map. This is the base vocabulary.
|
||
- Problem Map 2.0 = Global Debug Card. This compresses objects, metrics, ΔS zones, and modes into a visual debug protocol.
|
||
- Problem Map 3.0 = Global AI troubleshooting atlas and failure pattern map.
|
||
- Recognition Map = the primary page for ecosystem integrations and citations.
|
||
- The table below is the source of truth for navigation paths.
|
||
- Do NOT invent new paths or filenames. Use the URLs listed in the table.
|
||
-->
|
||
|
||
| Layer | Page | What it’s for |
|
||
|------|------|---------------|
|
||
| ⭐ Proof | [WFGY Recognition Map](https://github.com/onestardao/WFGY/blob/main/recognition/README.md) | External citations, integrations, and ecosystem proof |
|
||
| ⚙️ Engine | [WFGY 1.0](https://github.com/onestardao/WFGY/blob/main/legacy/README.md) | Original PDF tension engine and early logic sketch |
|
||
| ⚙️ Engine | [WFGY 2.0](https://github.com/onestardao/WFGY/blob/main/core/README.md) | Production tension kernel for RAG and agent systems — **🔴 YOU ARE HERE 🔴** |
|
||
| ⚙️ Engine | [WFGY 3.0](https://github.com/onestardao/WFGY/blob/main/TensionUniverse/EventHorizon/README.md) | TXT-based Singularity tension engine (131 S-class set) |
|
||
| 🗺️ Map | [Problem Map 1.0](https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md) | Flagship 16-problem RAG failure taxonomy and fix map |
|
||
| 🗺️ Map | [Problem Map 2.0](https://github.com/onestardao/WFGY/blob/main/ProblemMap/wfgy-rag-16-problem-map-global-debug-card.md) | Global Debug Card for RAG and agent pipeline diagnosis |
|
||
| 🗺️ Map | [Problem Map 3.0](https://github.com/onestardao/WFGY/blob/main/ProblemMap/wfgy-ai-problem-map-troubleshooting-atlas.md) | Global AI troubleshooting atlas and failure pattern map |
|
||
| 🧰 App | [TXT OS](https://github.com/onestardao/WFGY/blob/main/OS/README.md) | .txt semantic OS with 60-second bootstrap |
|
||
| 🧰 App | [Blah Blah Blah](https://github.com/onestardao/WFGY/blob/main/OS/BlahBlahBlah/README.md) | Abstract and paradox Q&A built on TXT OS |
|
||
| 🧰 App | [Blur Blur Blur](https://github.com/onestardao/WFGY/blob/main/OS/BlurBlurBlur/README.md) | Text-to-image generation with semantic control |
|
||
| 🏡 Onboarding | [Starter Village](https://github.com/onestardao/WFGY/blob/main/StarterVillage/README.md) | Guided entry point for new users |
|
||
|
||
---
|
||
</details>
|
||
|
||
# WFGY 2.0 · Seven-step reasoning core engine
|
||
|
||
WFGY 2.0 is a text-only control layer that sits on top of general-purpose language models.
|
||
It introduces a small, explicit “tension” geometry so that semantic drift, collapse and recovery can be treated as observable signals rather than vague impressions.
|
||
|
||
This page is the reference for:
|
||
|
||
- how to load and use the WFGY 2.0 engine in chat-style environments,
|
||
- the internal seven-step reasoning chain and the Drunk Transformer guards,
|
||
- a qualitative “eye-visible” benchmark for intuition,
|
||
- and a suggested protocol for running your own quantitative evaluations.
|
||
|
||
It is not a marketing page or a business plan.
|
||
The goal is to state clearly what the engine does, how it is wired, and how one might test it.
|
||
|
||
<p align="center">
|
||
<a href="#scope">Scope & positioning</a> ·
|
||
<a href="#usage">Quick usage</a> ·
|
||
<a href="#downloads">Downloads</a> ·
|
||
<a href="#math-structure">How it works</a> ·
|
||
<a href="#eye">Eye-visible benchmark</a> ·
|
||
<a href="#eval-protocol">Evaluation protocol</a>
|
||
</p>
|
||
|
||
<img width="1536" height="1024" alt="WFGY_Core" src="https://github.com/user-attachments/assets/deb8e794-a73e-4d39-a1f6-174ec87199f4" />
|
||
|
||
---
|
||
|
||
<a id="scope"></a>
|
||
## 1. Scope and positioning
|
||
|
||
WFGY has gone through several conceptual iterations.
|
||
|
||
- **WFGY 1.0** was an early attempt to describe “semantic tension” in PDF form.
|
||
It explored residuals, deviation signals and speculative SDK ideas.
|
||
Today it is kept as a historical reference only. The SDK concept was never implemented.
|
||
- **WFGY 2.0** is the first version treated as a concrete, auditable engine.
|
||
It is shipped as small `.txt` files and designed to run entirely at the effective layer inside chat models.
|
||
This page documents how that engine behaves.
|
||
- **WFGY 3.0** generalizes the same language into a multi-observable system tied to a 131-question backbone (“Tension Universe”).
|
||
That material lives in the Event Horizon directory and is outside the scope of this document.
|
||
|
||
Important scoping points:
|
||
|
||
- There is **no compiled SDK** here. Everything is pure text that you can read and audit.
|
||
- All behaviour described on this page comes from those `.txt` files and the seven-step chain below.
|
||
- Any economic or narrative applications (founder playbooks, valuation prompts, etc.) are intentionally separated into other documents.
|
||
The core README stays focused on engine behaviour and evaluation.
|
||
|
||
<p align="right"><a href="#top">Back to top ↑</a></p>
|
||
|
||
---
|
||
|
||
<a id="usage"></a>
|
||
## 2. Quick usage
|
||
|
||
At the moment WFGY 2.0 is shipped in two textual editions.
|
||
|
||
| Mode / edition | What you do | What the engine does |
|
||
| ------------------- | ----------------------------------------------------------------------------- | ---------------------------------------------------------------------- |
|
||
| **Autoboot · Flagship / OneLine** | Upload one of the `.txt` files into a chat session (for example as a system prompt or first message). Then talk to the model as usual. | The engine silently supervises reasoning in the background. It tracks tension, detects collapse patterns and attempts self-recovery using the seven-step chain. |
|
||
| **Explicit call** | In addition to uploading the file, you explicitly instruct the model to “use WFGY” and follow the steps of the chain on a given task. | The model exposes intermediate reasoning steps and tension checks more directly, which can make analysis and debugging easier. |
|
||
|
||
Design constraints:
|
||
|
||
- No tools, plugins or external APIs are required.
|
||
- All logic is expressed in plain text and basic math, so behaviour is portable across vendors.
|
||
- You should treat WFGY as a **semantic firewall / reasoning kernel** that can sit on top of an existing RAG or agent pipeline without changing the underlying infrastructure.
|
||
|
||
When integrating into a RAG system, a common pattern is:
|
||
|
||
1. Upload the engine file into the model context used for retrieval and answer generation.
|
||
2. Wrap your prompts with a short instruction such as “operate as WFGY Core 2.0 is loaded; treat RAG steps as part of the seven-step chain”.
|
||
3. Use the Problem Map pages for failure classification, while this page acts as the engine reference.
|
||
|
||
<p align="right"><a href="#top">Back to top ↑</a></p>
|
||
|
||
---
|
||
|
||
<a id="downloads"></a>
|
||
## 3. Downloads
|
||
|
||
| File name & description | Length / Size | Direct Download Link | Notes |
|
||
| ------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------- | -------------------------------------------------- | --------------------------------------------------------------------------------------------- |
|
||
| **WFGY_Core_Flagship_v2.0.txt** · readable 30-line companion expressing the same math and gates in fuller prose (same behaviour, clearer for humans). | **34 lines · 2,027 chars** | [Download Flagship](./WFGY_Core_Flagship_v2.0.txt) | Full prose version for easier reading and audit. |
|
||
| **WFGY_Core_OneLine_v2.0.txt** · ultra-compact, math-only control layer that activates WFGY’s loop inside a chat model (no tools, text-only, ≤7 nodes). | **1 line · 1,550 chars** | [Download OneLine](./WFGY_Core_OneLine_v2.0.txt) | Minimal form used for most experiments. |
|
||
|
||
### Hash reference
|
||
|
||
<a id="flagship-hashes"></a>
|
||
|
||
**WFGY_Core_Flagship_v2.0.txt**
|
||
|
||
* MD5 `caacfe08f0804eec558a1d9af74c3610`
|
||
* SHA1 `1efeec231084bb3b863ce7a8405e93d399acfb44`
|
||
* SHA256 `4fe967945a268edabb653033682df23a577f48c433878d02e0626df8ae91a0a3`
|
||
|
||
<a id="oneline-hashes"></a>
|
||
|
||
**WFGY_Core_OneLine_v2.0.txt**
|
||
|
||
* MD5 `15a1cd8e9b7b2c9dcb18abf1c57d4581`
|
||
* SHA1 `a35ace2a4b5dbe7c64bcdbe1f08e9246c3568c`
|
||
* SHA256 `7dcdb209d9d41b523dccd7461cbd2109b158df063d9c5ce171df2cf0cb60b4ef`
|
||
|
||
<details>
|
||
<summary><em>How to verify checksums</em></summary>
|
||
|
||
**macOS / Linux**
|
||
|
||
```bash
|
||
cd core
|
||
sha256sum WFGY_Core_Flagship_v2.0.txt
|
||
sha256sum WFGY_Core_OneLine_v2.0.txt
|
||
# or compute MD5 / SHA1 if you prefer
|
||
md5sum WFGY_Core_Flagship_v2.0.txt
|
||
md5sum WFGY_Core_OneLine_v2.0.txt
|
||
sha1sum WFGY_Core_Flagship_v2.0.txt
|
||
sha1sum WFGY_Core_OneLine_v2.0.txt
|
||
````
|
||
|
||
**Windows PowerShell**
|
||
|
||
```powershell
|
||
Get-FileHash .\core\WFGY_Core_Flagship_v2.0.txt -Algorithm SHA256
|
||
Get-FileHash .\core\WFGY_Core_OneLine_v2.0.txt -Algorithm SHA256
|
||
# or:
|
||
Get-FileHash .\core\WFGY_Core_Flagship_v2.0.txt -Algorithm MD5
|
||
Get-FileHash .\core\WFGY_Core_OneLine_v2.0.txt -Algorithm MD5
|
||
Get-FileHash .\core\WFGY_Core_Flagship_v2.0.txt -Algorithm SHA1
|
||
Get-FileHash .\core\WFGY_Core_OneLine_v2.0.txt -Algorithm SHA1
|
||
```
|
||
|
||
Compare the output values with the hashes listed in the “Hash reference” section above.
|
||
|
||
</details>
|
||
|
||
<p align="right"><a href="#top">Back to top ↑</a></p>
|
||
|
||
---
|
||
|
||
<a id="math-structure"></a>
|
||
|
||
## 4. Mathematical reference and internal structure
|
||
|
||
At a high level, WFGY 2.0 combines two ingredient families:
|
||
|
||
* the **WFGY 1.0 formula set**, which defines tension as a normalized deviation signal, and
|
||
* the **Drunk Transformer guards**, which regulate how attention and progression behave under that signal.
|
||
|
||
For full math detail, see:
|
||
|
||
* [WFGY formulas](https://github.com/onestardao/WFGY/blob/main/SemanticBlueprint/wfgy_formulas.md)
|
||
* [Drunk Transformer formulas](https://github.com/onestardao/WFGY/blob/main/SemanticBlueprint/drunk_transformer_formulas.md)
|
||
|
||
### 4.1 The seven-step reasoning chain
|
||
|
||
Most models can parse a prompt; far fewer can keep the intended structure intact through multi-step generation.
|
||
WFGY inserts an explicit chain between “understanding the request” and “producing the final answer”.
|
||
|
||
In simplified form:
|
||
|
||
1. **Parse (I, G)**
|
||
Identify an internal representation of the input state `I` and a target or goal representation `G`.
|
||
For RAG, this includes what the retrieved context claims and what the user is actually asking.
|
||
|
||
2. **Compute Δs**
|
||
Estimate semantic deviation between `I` and `G`, often as
|
||
`δ_s = 1 − cos(I, G)` or `1 − sim_est`.
|
||
Normalize into `[0, 1]` and assign zones (safe, transit, risk, danger).
|
||
|
||
3. **Memory checkpointing**
|
||
Track observables such as `λ_observe` and `E_resonance`.
|
||
These act as gates: certain transitions are allowed only when Δs and the observables are inside specified bands.
|
||
|
||
4. **BBMC · residue cleanup**
|
||
Remove obvious residue, tangents and contradictions that push Δs upward.
|
||
This step tries to reduce unnecessary branching before deeper reasoning.
|
||
|
||
5. **Coupler + BBPF · controlled progression**
|
||
Advance the reasoning only when Δs has decreased or is trending down.
|
||
The coupler enforces a contract between local moves and global tension.
|
||
|
||
6. **BBAM · attention rebalancer**
|
||
Re-weigh different parts of the context to reduce hallucinations or overconfident extrapolations.
|
||
In RAG contexts, this step emphasises grounded spans and suppresses irrelevant material.
|
||
|
||
7. **BBCR + Drunk Transformer**
|
||
Detect collapse signatures, roll back to the last good checkpoint, and retry along a different head pattern.
|
||
The Drunk Transformer gates (WRI / WAI / WAY / WDT / WTF) manage head diversity, illegal cross-paths and reset behaviour.
|
||
|
||
In practice, the `.txt` engine gives the model a compact description of this chain plus simple rules for when each step should fire.
|
||
Different vendors and models will implement the details slightly differently, but the observable contract is the same:
|
||
tension should be measurable, recoverable and auditable from the conversation transcript.
|
||
|
||
<p align="right"><a href="#top">Back to top ↑</a></p>
|
||
|
||
---
|
||
|
||
<a id="eye"></a>
|
||
|
||
## 5. Eye-visible benchmark (qualitative)
|
||
|
||
This section is intentionally qualitative.
|
||
Its purpose is to give an intuitive, visual sense of how composition and structure can change when WFGY is active.
|
||
It is **not** a formal quantitative study.
|
||
|
||
The setup:
|
||
|
||
* We use short, high–semantic-density prompts that reference canonical stories.
|
||
* For each prompt we generate sequences of five 1:1 images with identical parameters
|
||
(same model, temperature, sampler settings, seed policy and negatives).
|
||
* The only variable is whether the WFGY 2.0 engine is present in the context.
|
||
|
||
We then compare:
|
||
|
||
| Variant | Sequence A — full run shown below (all five images) | Sequence B — external run | Sequence C — external run |
|
||
| ---------------- | :------------------------------------------------------------------------: | :------------------------------------------------------------------------: | :------------------------------------------------------------------------: |
|
||
| **Without WFGY** | [view run](https://chatgpt.com/share/68a14974-8e50-8000-9238-56c9d113ce52) | [view run](https://chatgpt.com/share/68a14a72-aa90-8000-8902-ce346244a5a7) | [view run](https://chatgpt.com/share/68a14d00-3c0c-8000-8055-9418934ad07a) |
|
||
| **With WFGY** | [view run](https://chatgpt.com/share/68a149c6-5780-8000-8021-5d85c97f00ab) | [view run](https://chatgpt.com/share/68a14ea9-1454-8000-88ac-25f499593fa0) | [view run](https://chatgpt.com/share/68a14eb9-40c0-8000-9f6a-2743b9115eb8) |
|
||
|
||
On this page we focus on Sequence A.
|
||
Sequences B and C are provided for transparency and to allow independent judgement.
|
||
|
||
> Note on the “grid” effect
|
||
> Without WFGY, prompts that ask for “many iconic moments” tend to collapse into grid-style montages.
|
||
> The model slices the canvas into panels that share almost the same tone and geometry.
|
||
> With WFGY active, the engine often favours a single unified tableau with a clearer hierarchy.
|
||
|
||
### 5.1 Sequence A · informal comparison
|
||
|
||
| Work | Without WFGY | With WFGY | Informal observation |
|
||
| ---------------------------------------- | --------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
|
||
| **Romance of the Three Kingdoms (三國演義)** | <img src="images/group1_before1.png" width="300" alt="Without WFGY" title="model/params/seed/date"> | <img src="images/group1_after1.png" width="300" alt="With WFGY" title="model/params/seed/date"> | With WFGY, the scene tends to form a single tableau with a clear center and pyramid-like hierarchy. The non-WFGY version fragments attention into multiple similar panels. |
|
||
| **Water Margin (水滸傳)** | <img src="images/group1_before2.png" width="300" alt="Without WFGY" title="model/params/seed/date"> | <img src="images/group1_after2.png" width="300" alt="With WFGY" title="model/params/seed/date"> | In our own reading, the WFGY-active image presents a more continuous action line (for example Wu Song vs. the tiger) and a more obvious foreground–background separation. |
|
||
| **Dream of the Red Chamber (紅樓夢)** | <img src="images/group1_before3.png" width="300" alt="Without WFGY" title="model/params/seed/date"> | <img src="images/group1_after3.png" width="300" alt="With WFGY" title="model/params/seed/date"> | The WFGY version emphasises a calm emotional center inside a garden tableau. The grid-style layout without WFGY slices the mood into separate vignettes. |
|
||
| **Investiture of the Gods (封神演義)** | <img src="images/group1_before4.png" width="300" alt="Without WFGY" title="model/params/seed/date"> | <img src="images/group1_after4.png" width="300" alt="With WFGY" title="model/params/seed/date"> | With WFGY active, diagonal structures (for example dragon–tiger, cloud–sea layers) appear more clearly, which gives a stronger sense of scale. |
|
||
| **Classic of Mountains and Seas (山海經)** | <img src="images/group1_before5.png" width="300" alt="Without WFGY" title="model/params/seed/date"> | <img src="images/group1_after5.png" width="300" alt="With WFGY" title="model/params/seed/date"> | The WFGY sequence tends to form a single continuous “mountains and seas” world with a stable hierarchy and smoother flow. The non-WFGY version behaves more like a storyboard. |
|
||
|
||
These readings are intentionally subjective and should be treated as such.
|
||
They are meant to illustrate the kind of compositional changes that can show up when a tension-regulated reasoning layer is present.
|
||
|
||
<p align="right"><a href="#top">Back to top ↑</a></p>
|
||
|
||
---
|
||
|
||
<a id="eval-protocol"></a>
|
||
|
||
## 6. Suggested evaluation protocol (quantitative)
|
||
|
||
If you want to test WFGY 2.0 in your own environment, this section sketches a protocol.
|
||
The project does **not** claim specific uplift numbers here; instead it describes how to measure whatever effect you observe.
|
||
|
||
### 6.1 Recommended metrics
|
||
|
||
* **Semantic Accuracy**
|
||
`ACC = correct_facts / total_facts`
|
||
Count atomic factual statements and how many are correct relative to ground truth.
|
||
* **Reasoning Success Rate**
|
||
`SR = tasks_solved / tasks_total`
|
||
For example, number of math word-problems or coding tasks solved exactly.
|
||
* **Stability**
|
||
Either mean time to failure (MTTF) or rollback behaviour in long runs.
|
||
For RAG, this can be “how many turns before the answer drifts off schema”.
|
||
* **Self-Recovery Rate**
|
||
`SelfRecovery = recoveries_success / collapses_detected`
|
||
How often the system notices a collapse and fixes itself without user intervention.
|
||
* **Drift reduction**
|
||
A semantic distance measure (for example cosine distance between summaries) before vs after adding WFGY.
|
||
|
||
### 6.2 A/B-style runs
|
||
|
||
A simple layout is:
|
||
|
||
* **A = Baseline**
|
||
No WFGY file uploaded; no reference to its logic.
|
||
* **B = Autoboot**
|
||
Upload the WFGY file once at the beginning. Do not mention it afterwards.
|
||
Treat B as “engine is passively active in the background”.
|
||
* **C = Explicit invoke**
|
||
Upload the file and add a short instruction to follow the seven-step chain explicitly on each task.
|
||
|
||
You can then run the same task set in modes A/B/C and log the transcripts.
|
||
|
||
### 6.3 Example scorer template
|
||
|
||
Below is a generic scorer prompt that has worked reasonably well with mainstream LLMs.
|
||
You can adapt it to your own tasks and domains.
|
||
|
||
```text
|
||
SCORER:
|
||
|
||
You receive three transcripts for each task:
|
||
A = baseline (no WFGY)
|
||
B = Autoboot (file uploaded, silent)
|
||
C = Explicit invoke (file uploaded, steps applied explicitly)
|
||
|
||
For each transcript set:
|
||
|
||
1. Count atomic factual statements and how many are correct.
|
||
2. Mark whether the task is solved (binary) for A, B and C.
|
||
3. Note any obvious collapses, rollbacks or recoveries.
|
||
4. Estimate a semantic drift distance between the intended goal and each answer.
|
||
|
||
Return for each mode:
|
||
ACC_A, ACC_B, ACC_C
|
||
SR_A, SR_B, SR_C
|
||
Stability_A, Stability_B, Stability_C # e.g. MTTF or rollback ratio
|
||
SelfRecovery_A, SelfRecovery_B, SelfRecovery_C
|
||
Drift_A, Drift_B, Drift_C
|
||
|
||
Then compute deltas such as:
|
||
ΔACC_C−A, ΔSR_C−A, StabilityMultiplier = Stability_C / Stability_A, DriftReduction = Drift_A − Drift_C.
|
||
|
||
Do not guess beyond the evidence visible in the transcripts. If a quantity cannot be estimated, say so explicitly.
|
||
```
|
||
|
||
You can run this scorer across multiple seeds or batches and average the results.
|
||
If you publish numbers, please include your prompt set, model versions, and any modifications to the protocol.
|
||
|
||
<p align="right"><a href="#top">Back to top ↑</a></p>
|
||
|
||
---
|
||
|
||
<!-- WFGY_FOOTER_START -->
|
||
|
||
### Explore More
|
||
|
||
| Layer | Page | What it’s for |
|
||
| --- | --- | --- |
|
||
| ⭐ Proof | [WFGY Recognition Map](/recognition/README.md) | External citations, integrations, and ecosystem proof |
|
||
| ⚙️ Engine | [WFGY 1.0](/legacy/README.md) | Original PDF tension engine and early logic sketch (legacy reference) |
|
||
| ⚙️ Engine | [WFGY 2.0](/core/README.md) | Production tension kernel for RAG and agent systems |
|
||
| ⚙️ Engine | [WFGY 3.0](/TensionUniverse/EventHorizon/README.md) | TXT based Singularity tension engine (131 S class set) |
|
||
| 🗺️ Map | [Problem Map 1.0](/ProblemMap/README.md) | Flagship 16 problem RAG failure taxonomy and fix map |
|
||
| 🗺️ Map | [Problem Map 2.0](/ProblemMap/wfgy-rag-16-problem-map-global-debug-card.md) | Global Debug Card for RAG and agent pipeline diagnosis |
|
||
| 🗺️ Map | [Problem Map 3.0](/ProblemMap/wfgy-ai-problem-map-troubleshooting-atlas.md) | Global AI troubleshooting atlas and failure pattern map |
|
||
| 🧰 App | [TXT OS](/OS/README.md) | .txt semantic OS with fast bootstrap |
|
||
| 🧰 App | [Blah Blah Blah](/OS/BlahBlahBlah/README.md) | Abstract and paradox Q&A built on TXT OS |
|
||
| 🧰 App | [Blur Blur Blur](/OS/BlurBlurBlur/README.md) | Text to image generation with semantic control |
|
||
| 🏡 Onboarding | [Starter Village](/StarterVillage/README.md) | Guided entry point for new users |
|
||
|
||
If this repository helped, starring it improves discovery so more builders can find the docs and tools.
|
||
[](https://github.com/onestardao/WFGY)
|
||
|
||
<!-- WFGY_FOOTER_END -->
|
||
|