Jun 7 05:14 data: normalize TP-RAG model type 04:48 Clarify benchmark evaluation caveats 04:26 Add benchmark research group metadata 04:17 Add pure-text GST benchmarks Add GPSBench, UrbanGeoEval, Compositional GeoQA, and TP-RAG to gsta_benchmarks.csv while preserving the existing 15-column schema. 04:03 docs: clarify TurnBack evaluation dependency 03:13 Fix live benchmark count badge 03:04 Update TripCraft benchmark caveats Jun 6 19:16 data: add GeoAgentBench benchmark entry 18:21 data: clarify missing code links 18:12 data: fill verified benchmark URLs 17:58 docs: clarify URL status fields 17:39 docs(readme): force inline badge rendering Remove the outer centered div around the header and render all badges in one styled inline HTML paragraph, while keeping the benchmark count dynamic. 17:30 docs(readme): make benchmark count dynamic Use a Shields dynamic JSON badge backed by the Hugging Face dataset-server row count and remove row-count-derived hardcoded statistics from the README. 17:25 docs(readme): keep badges on one row Refactor the README badge header into a single inline HTML paragraph with non-breaking spaces so all badges render as one centered row. 17:23 docs(readme): fix citation author Use the preferred author spelling Dawei Zhang in the dataset citation. 17:23 docs(readme): align header badges Make the header badges compact and inline so they render side-by-side on the dataset card. 17:21 docs(readme): polish dataset card Refresh the GSTA dataset card with a richer open-source style layout, badges, quick-start examples, coverage snapshot, schema, contribution guidance, and citation section. 17:06 feat(gsta): publish benchmark atlas dataset 17:05 initial commit 
