What waste pattern are we missing?

[hernancapucci]

We’re launching llm-waste-audit with a minimal but honest MVP focused on two common and expensive waste patterns:

• W1 — Context Bloat
• W2 — Cache Breaker

The broader taxonomy is already documented:

• W1 — Context Bloat
• W2 — Cache Breaker
• W3 — Redundant Reasoning
• W4 — Model Overkill
• W5 — Output Excess
• W6 — Retrieval Waste
• W7 — Session Memory Failure
• W8 — Unsafe Semantic Reuse

We want help expanding it with real production patterns.

If you’ve seen a recurring, avoidable LLM waste pattern, please comment or open a new issue with:

1. A short name or description of the pattern
2. Why it increases cost, latency, or token usage
3. A minimal reproducible example (JSON preferred)
4. Whether it is provider-specific or general
5. The rough savings you would expect if fixed

We strongly prefer concrete examples over abstract opinions.

This repo is meant to be a living technical benchmark, not a marketing artifact.