nx_qabench_fetch.nx

nx_qabench_fetch.nx -- broaden the researcher's public-benchmark ladder beyond HotpotQA (operator 2026-07-07: "use others like beerqa.github.io, climb the ladder to the top"). Banks, sovereignly (TLS-1.3 + CA store): 1) SQuAD v1.1 validation (HF datasets-server rows API) -- SINGLE context paragraph per Q = isolates the EXTRACTION quality from retrieval (the 1-hop floor rung; should score HIGHEST mechanically). 2) BeerQA dev v1.0 (Stanford, direct JSON) -- UNIFIES SQuAD(1-hop)+HotpotQA(2-hop)+new 3-hop in ONE format tagged by source = the whole difficulty ladder in one dataset. 16MB file; big cap, truncation-tolerant (the harness parses complete records + stops at the cut, like the HotpotQA path). Raw JSON -> knowledge/fetched/<name>.raw for nx_qabench to parse. Run SOLO. expect_exit: 0 license_tier: ORIGINAL