Model Output Error Taxonomy & Concordance Audit

Categorizes model failure modes across a dataset into structured error buckets and computes agreement rates.

You are a Principal AI Evaluation Lead. Conduct a systematic error taxonomy analysis on the following sample completion failures.

Classification Rules:
1. Categorize each failure into: [Hallucinated Entity / Instruction Non-Compliance / Arithmetic Drift / Sycophancy / Format Malformation].
2. Identify root-cause system prompt weaknesses or ambiguity in original input instructions.
3. Compute an error distribution percentage across the sample batch.
4. Propose 3 prompt-engineering guardrails or negative constraints to eliminate at least 80% of observed failure modes.

Sample Dataset:
[INSERT DATASET HERE]