Accurate viral card: critic purpose, params, usage, funnel

README.md

@@ -7,60 +7,87 @@ pipeline_tag: text-generation
 base_model: unsloth/Qwen3-8B
 base_model_relation: finetune
 tags:
-  - chain-of-thought
+  - reasoning
   - critic
+  - verification
   - evaluation
-  - fableforge
+  - fact-checking
+  - chain-of-thought
+  - agent
+  - qwen3
   - gguf
-  - imatrix
-  - llama.cpp
-  - lm-studio
   - ollama
-  - reasoning
-  - text-generation
+  - no-refusals
+  - fableforge
 ---
 
-# ReasonCritic-7B
+# ReasonCritic-7B — the verifier that catches your agent's bad reasoning
 
-> **A reasoning critic — evaluates and improves chain-of-thought.**
+> **A reasoning critic that scores and repairs logic chains — PASS/FAIL verdicts, confidence scores, fallacy detection, and actionable fixes.**
 
 [![Downloads](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fhuggingface.co%2Fapi%2Fmodels%2Ffableforge-ai%2FReasonCritic-7B&query=downloads&label=downloads&color=blue)](https://huggingface.co/fableforge-ai/ReasonCritic-7B) [![Demos](https://img.shields.io/badge/🤗-Live%20Demos-orange)](https://huggingface.co/spaces/fableforge-ai/fableforge-nexus)
 
-Base model: [unsloth/Qwen3-8B](https://huggingface.co/unsloth/Qwen3-8B)  ·  Part of the **FableForge** ecosystem.
+Most agents fail *silently* — they produce confident, wrong reasoning. **ReasonCritic-7B** is the verification layer: point it at any reasoning chain or agent output and it returns a structured critique you can act on.
 
----
+## 🔍 What it does
+- ✅ **PASS / FAIL verdict** + a **confidence score** on any reasoning chain
+- 🔎 Flags **logical fallacies, unsupported claims, and gaps**
+- 🛠️ Returns **actionable suggestions** to repair the reasoning
+- 🤖 Designed to sit in an **agent loop** as the checker before an answer is trusted
 
-## ⚡ Quickest start — Ollama (pulls straight from this repo → credits the source)
+## ⚡ Quick start — Ollama
 
 ```bash
 ollama run hf.co/fableforge-ai/ReasonCritic-7B:Q4_K_M
 ```
 
-## 🦙 llama.cpp
+**Recommended sampling for deterministic critique:** `temperature 0.3 · top_p 0.9 · top_k 40 · repeat_penalty 1.1 · num_ctx 4096`.
+
+Build the tuned critic locally with those defaults baked in:
+
+```dockerfile
+# Modelfile
+FROM hf.co/fableforge-ai/ReasonCritic-7B:Q4_K_M
+PARAMETER num_ctx 4096
+PARAMETER temperature 0.3
+PARAMETER top_p 0.9
+PARAMETER top_k 40
+PARAMETER repeat_penalty 1.1
+SYSTEM """You are ReasonCritic, a reasoning critic. Evaluate, score, and improve logical
+reasoning chains. Identify fallacies, unsupported claims, and gaps. Produce structured
+verification: PASS/FAIL verdict, confidence score, issue list, and actionable suggestions."""
+```
 
 ```bash
-llama-cli -hf fableforge-ai/ReasonCritic-7B:Q4_K_M -p "Hello!"
+ollama create reasoncritic -f Modelfile && ollama run reasoncritic
 ```
 
-## 💠 LM Studio
-Search **`fableforge-ai/ReasonCritic-7B`** and pick a quant below.
+## 🧾 Example output
 
----
+```
+VERDICT: FAIL  (confidence 0.82)
+ISSUES:
+  - Step 3 asserts X with no supporting evidence (unsupported claim)
+  - Steps 4 and 6 are mutually circular
+SUGGESTIONS:
+  - Cite evidence for X, or mark it explicitly as an assumption
+  - Derive step 6 independently of step 4
+```
 
-## 📦 Provided quants (download one file, not the whole repo)
+## 📦 Provided quants
 
 | File | Quant | Size | Notes |
 | ---- | ----- | ---- | ----- |
-| [qwen3-8b.F16.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.F16.gguf) | `F16` | 14.43 GB | Full precision (unquantized). Reference / conversion. |
-| [qwen3-8b.Q8_0.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q8_0.gguf) | `Q8_0` | 8.71 GB | Extremely high quality — usually overkill. |
-| [qwen3-8b.Q6_K.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q6_K.gguf) | `Q6_K` | 6.73 GB | Very high quality, near-perfect. **Recommended** if you have the RAM. |
+| [qwen3-8b.F16.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.F16.gguf) | `F16` | 14.43 GB | Full precision / reference. |
+| [qwen3-8b.Q8_0.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q8_0.gguf) | `Q8_0` | 8.71 GB | Max quality, overkill. |
+| [qwen3-8b.Q6_K.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q6_K.gguf) | `Q6_K` | 6.73 GB | Near-perfect. **Recommended** if RAM allows. |
 | [qwen3-8b.Q5_K_M.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q5_K_M.gguf) | `Q5_K_M` | 5.85 GB | High quality. **Recommended.** |
-| [qwen3-8b.Q4_K_M.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q4_K_M.gguf) | `Q4_K_M` | 5.03 GB | ⭐ **Best size/quality balance — default pick for most users.** |
-| [qwen3-8b.Q4_0.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q4_0.gguf) | `Q4_0` | 4.77 GB | Legacy format — prefer Q4_K_M. |
-| [qwen3-8b.Q3_K_M.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q3_K_M.gguf) | `Q3_K_M` | 4.12 GB | Lower quality; usable when RAM is tight. |
-| [qwen3-8b.Q2_K.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q2_K.gguf) | `Q2_K` | 3.28 GB | Very low quality — surprisingly usable. |
+| [qwen3-8b.Q4_K_M.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q4_K_M.gguf) | `Q4_K_M` | 5.03 GB | ⭐ **Best balance — default pick.** |
+| [qwen3-8b.Q4_0.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q4_0.gguf) | `Q4_0` | 4.77 GB | Legacy — prefer Q4_K_M. |
+| [qwen3-8b.Q3_K_M.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q3_K_M.gguf) | `Q3_K_M` | 4.12 GB | Lower quality, low RAM. |
+| [qwen3-8b.Q2_K.gguf](https://huggingface.co/fableforge-ai/ReasonCritic-7B/blob/main/qwen3-8b.Q2_K.gguf) | `Q2_K` | 3.28 GB | Very low quality. |
 
-**Not sure?** Take **`Q4_K_M`**. Low RAM → **`IQ4_XS`**. Max quality → **`Q6_K`**.
+> ℹ️ Files are currently named `qwen3-8b.*`; the Ollama `:Q4_K_M` tag still resolves correctly.
 
 ## 🧠 Prompt format (ChatML)
 
@@ -68,20 +95,21 @@ Search **`fableforge-ai/ReasonCritic-7B`** and pick a quant below.
 <|im_start|>system
 {system_prompt}<|im_end|>
 <|im_start|>user
-{prompt}<|im_end|>
+{reasoning_chain_to_verify}<|im_end|>
 <|im_start|>assistant
 ```
 
-## 🌌 See it live — the FableForge demos
+## 🌌 Part of the FableForge ecosystem
 
-This family powers a galaxy of free, interactive HF Spaces:
+ReasonCritic is the **verification layer** for the FableForge agent stack. Try the live demos:
 
 - 🧭 [FableForge Nexus](https://huggingface.co/spaces/fableforge-ai/fableforge-nexus)
 - 🍺 [Infinite NPC](https://huggingface.co/spaces/fableforge-ai/infinite-npc)
-- 🎭 [Dual-GM](https://huggingface.co/spaces/King3Djbl/dual-gm-simulator)
 - 💻 [ShellWhisperer](https://huggingface.co/spaces/King3Djbl/shellwhisperer-terminal)
-- 👻 [Ghost Writer](https://huggingface.co/spaces/fableforge-ai/ghost-writer)
+- 🎭 [Dual-GM](https://huggingface.co/spaces/King3Djbl/dual-gm-simulator)
+
+Base model: [unsloth/Qwen3-8B](https://huggingface.co/unsloth/Qwen3-8B).
 
 ---
 
-*⭐ Like & share — it helps people find the source instead of a mirror.*
+*⭐ Like & share — pull from the source, not a mirror.*