Add KB Arena (knowledge graph + hybrid retrieval benchmark)

[xmpuspus]

Adds KB Arena to the Open-source Project section.

KB Arena is an open-source benchmark that runs nine architecturally distinct retrieval strategies head-to-head on user-supplied corpora. The two GraphRAG-relevant strategies:

• knowledge_graph — extraction-driven Neo4j graph using a universal 5-node-type / 7-rel-type schema (Topic, Component, Process, Config, Constraint + DEPENDS_ON, CONTAINS, CONNECTS_TO, TRIGGERS, CONFIGURES, ALTERNATIVE_TO, EXTENDS). Source provenance is stamped on every entity end-to-end so chunk-level retrieval matches against section ground truth.
• hybrid — RRF-fused vector + graph retrieval with three-stage intent routing (keyword scan → Haiku LLM → regex fallback). Domain-agnostic.

Why it matters for the GraphRAG community: the field still lacks a clean apples-to-apples way to say "graph beats vector on this corpus by X with p<Y." KB Arena ships paired-bootstrap 95% CIs and Wilcoxon paired two-sided p-values on per-question IR metrics (Recall@k, NDCG, MAP, R-Precision, bpref), so GraphRAG vs vector vs hybrid comparisons report effect size with statistical confidence rather than mean-only deltas.

• Active: v0.8.1 released 2026-05-21, 617 tests, MIT, on PyPI as kb-arena
• Zenodo archive (concept DOI): 10.5281/zenodo.20319678
• Ships an aws-compute benchmark corpus (75 questions, 5 difficulty tiers, 35 with chunk-level ground truth) as a pedagogical baseline

Entry appended to the existing Open-source Project list following the badge-prefix format used by the other 14 entries. No other sections modified.