docs(judge): writeup draft v1 — methodology arc + position + v1.2 fix-list

calibrate(jury): v1.1+v1.1.1 — fix weighting bugs; recency-position paraphrase clause

rubric: clarify groundedness reference scope (snippets-only) for v1.1 gold (#20)

docs(plans): judge-layer v1 implementation plan — 12 phases, ~50 tasks

docs(plans): judge-layer v1 design — supersede continuous-scale judges with discrete-anchored 2-judge jury + κ-validated calibration

docs(plan): add Part A OWASP mapping implementation plan

docs(plan): Part A design self-review fixes (LLM02 consistency, anti-padding template, paired-review gate)

docs(plan): add Part A OWASP LLM Top 10 (2025) mapping design

docs: multi-corpus refactor implementation plan

docs: multi-corpus refactor design

style: fix ruff lint — import sorting, line length

docs: add known limitations and future work for self-hosted benchmark

docs: deepen self-hosted analysis in provider comparison

feat: infrastructure sprint — vLLM/Modal, Helm, Terraform (#8)

fix: comparison framing, mock-specific failure analysis, stale test counts

feat: langchain baseline evaluation results (OpenAI + Anthropic)

fix: remove stale V1 docs, update DECISIONS.md for V2

docs: add provider comparison report (OpenAI vs Anthropic Haiku)

feat: real benchmark numbers from OpenAI gpt-4o-mini evaluation

fix: grounded refusal checks no-sources, reference_answer for judge, mock disclaimer

fix: retrieval metrics use ranked sources, LLM judge wired, report complete

feat: Day 7 — evaluation harness, metrics, report, expanded golden dataset

feat: add reproducible retrieval gate check with committed artifact

Add design document for agent-bench V1