v1.1.0

What's New

Benchmark Framework

• Extraction benchmark framework with Anthropic and OpenAI provider support
• Structure benchmark (ADR-0003): compares Monolith vs Flat vs Folder domain structures
• Reduced structure bench from 16 domains/10 repeats to 6 domains/4 repeats (144 calls vs 960), keeping 3 distractor pairs (finance/billing, product/platform, security/compliance)

Quality Scorer

• 10-dimension deterministic quality scorer (no LLM required)

Cache Observability

• Track cache_creation_tokens and cache_read_tokens per run
• Warn when system prompt is below Anthropic's 2048-token minimum for caching ([no cache])

Documentation & ADRs

• Added ADRs 0001–0003
• Unit tests and benchmark results included

v1.0.0

Merge pull request #3 from larsroettig/feat/initial-implementation

docs: clarify the score measures form, not correctness