### ⚖️ Challenge 02 — Evaluation

**Accuracy, Integrity, and Reproducibility**

This challenge is about **building trust**, not chasing leaderboard points.
Your goal is to show that your workflow can *detect, repair, and validate* data issues transparently—turning a messy dataset into something reproducible and reliable.

---

#### 🧩 Evaluation Principles

Submissions are reviewed on five professional data-science dimensions:

| Area                        | What Reviewers Look For                                                                                           |
| :-------------------------- | :---------------------------------------------------------------------------------------------------------------- |
| **Audit Completeness**      | All major data-quality problems identified (missing, duplicates, drift, outliers). Clear summary of findings.     |
| **Repair Integrity**        | Fixes are principled and justified—imputation methods documented with rationale and diagnostics.                  |
| **Validation Rigor**        | Evidence that cleaning preserved statistical meaning (distributional checks, leakage tests, re-trained baseline). |
| **Reproducibility**         | Scripts and notebooks run end-to-end without manual edits; outputs match documentation.                           |
| **Clarity & Documentation** | Readable reports, organized folder structure, concise executive summary.                                          |

---

#### 🧮 Scoring Framework

Instead of numeric scores, each submission receives a qualitative review focused on craftsmanship and learning:

| Rating                  | Description                                                                                |
| :---------------------- | :----------------------------------------------------------------------------------------- |
| 🥇 **Gold Integrity**   | Exceptional audit depth and documentation; all fixes reproducible and statistically sound. |
| 🥈 **Silver Integrity** | Complete, functional workflow; minor clarity or reproducibility improvements possible.     |
| 🥉 **Bronze Integrity** | Core requirements met; clear learning progress and honest methodology.                     |
| 🧭 **Needs Revision**   | Missing audits, unclear rationale, or non-reproducible pipeline—resubmission encouraged.   |

---

#### 📊 What Counts as Success

* All injected issues found and either fixed or explained.
* Audit and repair steps clearly logged and repeatable.
* Post-cleaning dataset passes validation and baseline model re-test without suspicious metric jumps.
* Reports and notebooks readable by someone new to the project.

---

#### 💬 Peer and AI-Assisted Review

Participants are encouraged to review one another’s notebooks and logs.
AI tools may suggest repairs or summaries — your job is to verify them and explain your reasoning.
Integrity is earned through transparent collaboration, not automation alone.

---

#### 🔮 Looking Ahead

Challenge 03 (Data → Signal) will introduce automated metrics and forecast validation suites.
The standards you build here — reproducibility, clarity, and honest documentation — will become the benchmark for every future challenge in the series.