follow up first commit

data/historical_weights_manual_iot_review.README.md

@@ -0,0 +1,34 @@
+# Manual + IoT historical labels (for review)
+
+**File:** `historical_weights_manual_iot_review.csv`
+
+This file is the production consensus label source (`csv_path` in `configs/weight_cc2_cc3.yaml`).
+
+## Sources
+
+| Column | Source |
+|--------|--------|
+| `manual_avg_g` | `historical_weights_avg.csv` → Berat Aktual (g) |
+| `iot_avg_g` | `data_cycle7.jpeg` → AVG IOT column (days 1–37) |
+| Day 0 Manual 32 g | DOC / Age-0 Manual from sampling sheet |
+
+## Proposed consensus rule (draft)
+
+| Situation | Proposed target | Train weight | Flag |
+|-----------|-----------------|--------------|------|
+| Manual & IoT gap ≤ 5% | mean(Manual, IoT) | 1.0 | AGREE |
+| Gap 5–10% | mean(Manual, IoT) | 0.5 | MODERATE_GAP |
+| Gap > 10% | mean(Manual, IoT) | 0.25 | LARGE_GAP_DOWNWEIGHT |
+| Manual only | Manual | 0.7 | MANUAL_ONLY |
+| IoT only | IoT | 0.7 | IOT_ONLY |
+| Day 0 DOC | Manual 32 | 0.0 | DOC_REFERENCE_ONLY |
+
+## How to review
+
+Fill these columns:
+
+- `reviewer_decision`: `keep` / `exclude` / `prefer_manual` / `prefer_iot` / `custom`
+- `reviewer_target_g`: only if `custom` or override
+- `reviewer_notes`: free text
+
+After review, we can freeze a clean training labels CSV from your decisions.