Address README feedback: drop unverified Qwen-Marketing claim, drop 'real/not illustrative' meta-commentary, clarify fixed eval split, cleaner pipeline example (named document var, int output)

README.md

@@ -42,7 +42,7 @@ A high-performance regression model for assessing marketing content quality on a
 This model is built for **large-scale content quality filtering**, not single-document review. Typical applications:
 
 - **Pretraining corpus curation** — score every document in a FineWeb-scale crawl and keep only the top percentile of marketing content, the same way FineWeb-Edu uses an educational-quality classifier to filter Common Crawl.
-- **Dataset construction for domain-specific LLMs** — this classifier is what produces the filtered corpus behind [marketeam/Qwen-Marketing](https://huggingface.co/marketeam/Qwen-Marketing)'s marketing-domain pretraining/fine-tuning data.
+- **Dataset construction for domain-specific LLMs** — build filtered marketing-domain pretraining or fine-tuning corpora.
 - **Marketing content quality scoring** — rank or threshold marketing web pages, blog posts, or landing pages by practitioner-quality writing, as opposed to spam/SEO filler.
 
 It is **not** intended as a fine-grained per-document editorial tool — see [Intended Use](#intended-use) for scope and [Safety & Bias](#safety--bias) for known limitations.
@@ -67,7 +67,7 @@ This model predicts marketing content quality by scoring web pages against a fiv
 | **Output** | Scalar regression score, nominally 0–5 (not clipped by the model itself — see note below) |
 | **Max token length** | 2048 |
 
-The head is not bounded to [0, 5]; scores for exceptionally strong documents can land slightly above 5 (see the real example in [How to Use](#how-to-use)). Downstream consumers that need a bounded integer score should clip and round, as `infer.py` does (`int_score = round(clip(score, 0, 5))`).
+The head is not bounded to [0, 5]; scores for exceptionally strong documents can land slightly above 5 (see [How to Use](#how-to-use)). Downstream consumers that need a bounded integer score should clip and round, as `infer.py` does (`int_score = round(clip(score, 0, 5))`).
 
 ### Model Variants
 
@@ -104,7 +104,7 @@ Only BMse is recommended for production. Others are retained for reproducibility
 | **Optimizer** | Adam, lr=3e-4 (no weight decay) |
 | **Batch size** | 32 |
 | **Train set** | 445,116 docs (natural distribution) |
-| **Eval set** | 50,000 docs (held out, fixed split) |
+| **Eval set** | 50,000 docs, held out from training and reused as the same fixed split across all four head variants (BMse, Head A, Head B, LDS) for direct comparison |
 | **Epochs** | 5 (early stopping patience=3 on Spearman) |
 | **Best checkpoint** | Epoch 2 |
 | **Seed** | 42 |
@@ -138,7 +138,7 @@ Only BMse is recommended for production. Others are retained for reproducibility
 
 **Split:**
 - Training: 445,116 docs
-- Evaluation: 50,000 docs (held out, fixed split)
+- Evaluation: 50,000 docs — held out from training and reused as the same fixed split across BMse, Head A, Head B, and LDS, so their metrics are directly comparable
 
 **Distribution:** Heavily left-skewed (mean 1.52/5), typical of web crawl data. Top ~20% score ≥3; top ~6% score ≥4.
 
@@ -154,17 +154,20 @@ pipe = pipeline(
     model="marketeam/Fineweb-Classifier-Marketing",
     trust_remote_code=True,
 )
-pipe(
+
+document = (
     "Best practices for B2B SaaS pricing: a tactical breakdown of tiered vs. "
     "usage-based models, with real conversion benchmarks from 40+ companies "
     "we've worked with this year."
 )
+result = pipe(document)
+print(round(result["score"]))
 ```
 
-Output (real, captured from this checkpoint — not illustrative):
+Output:
 
-```python
-{'label': 'quality_score', 'score': 5.024752616882324}
+```
+5
 ```
 
 `trust_remote_code=True` is required because this repo ships a custom `PreTrainedModel`/pipeline (frozen encoder + regression head is not a stock `transformers` architecture). Review `modeling_marketing_classifier.py`, `configuration_marketing_classifier.py`, and `pipeline_marketing_classifier.py` in this repo before enabling it, per standard `trust_remote_code` practice.