archive old experiments and publish swe series

experiments/baseline-swebench/README.md

@@ -0,0 +1,6 @@
+# SWE-bench baseline
+
+Three no-communication model populations on the same 20 conflicting SWE tasks.
+See [DESIGN.md](DESIGN.md) for the frozen design and [analysis.md](analysis.md) for reviewed results.
+
+Infrastructure recovery is recorded in [recovery.md](recovery.md). The selected scored results and automatic report are in `logs/baseline-swebench/`.