Update README with detailed model card - Adaption AutoScientist Challenge

README.md

@@ -52,27 +52,20 @@ This model is trained on **real SEC filing data** with gold reasoning programs t
 | **Language** | English (en) |
 | **License** | MIT |
 
-## 🧪 AutoScientist Platform Usage (100% Platform-Driven)
-
-Every step of the training pipeline was executed through the Adaption Labs AutoScientist platform:
-
-1. **Adaptive Data Pipeline** — Ingested financial reasoning data and performed quality validation (Grade **B** maintained at score 8.9).
-
-2. **Prompt Deduplication** — Removed duplicate financial questions to ensure diverse calculation patterns.
-
-3. **Prompt Rephrase** — **Disabled** per platform guidance: rephrasing would corrupt table formatting, column headers, and financial terminology.
-
-4. **Reasoning Traces** — Enhanced completions with structured mathematical chains: extract variables → state formula → execute step-by-step → verify against gold program.
-
-5. **Blueprint Constraints** — Defined financial-specific grading criteria:
-   - MUST extract the relevant variables from the table before computing
-   - MUST state the exact mathematical formula being used
-   - MUST execute arithmetic sequentially and show intermediate results
-   - MUST output the final numeric answer clearly
-
-6. **AutoScientist Training Engine** — Automated LoRA SFT on the 109B MoE architecture.
-
-7. **LLM-as-a-Judge Evaluation** — Compared adapted model vs. base on held-out financial prompts.
+## AutoScientist Platform Integration
+
+The model was adapted using the automated training features of the Adaption Labs AutoScientist platform:
+
+* **Adaptive Data Pipeline:** Validated the financial dataset structure, maintaining a stable Grade B (8.9 score) across multi-step calculation patterns.
+* **Prompt Deduplication:** Removed redundant financial questions, focusing training on unique computation flows.
+* **Prompt Rephrase Constraint:** Prompt rephrasing was disabled to protect the integrity of financial tables, column headers, and tabular structures from corruption.
+* **Reasoning Traces & Blueprint Constraints:** Enforced structured arithmetic thinking. The model is trained to show complete working in a `<think>` trace verified against gold calculation programs before outputting the final answer. The blueprint constraints mandate:
+  1. Extracting the exact cell coordinates and values from the input SEC table.
+  2. Stating the mathematical formula explicitly.
+  3. Executing arithmetic steps sequentially.
+  4. Outputting a clear, verified final numeric value.
+* **Hyperparameter Optimization:** Managed the training run using LoRA SFT targeting the attention and feed-forward layers of meta-llama/Llama-4-Scout-17B-16E-Instruct.
+* **Evaluation:** Compared the adapted model performance against the base model on held-out financial prompts using automated preference scoring.
 
 ## 📊 Dataset