Add files using upload-large-folder tool

SOURCES.md

@@ -24,6 +24,13 @@ combined repository public. SWE-Zero records can additionally contain material
 subject to the original code repositories' licenses; repository provenance is
 retained in the session table.
 
+Published upstream terms include MIT for UltraChat, Apache-2.0 for Toucan,
+CC-BY-4.0 for SWE-Zero, and CC-BY-SA-4.0 for Schema-Guided Dialogue.
+[TaskMaster2's original authors](https://github.com/google-research-datasets/Taskmaster/tree/master/TM-2-2020)
+specify CC-BY-4.0. The D2L dataset repository does not supply a dataset card or
+license file in the inspected file listing; do not infer its data license from
+the D2L implementation's code license.
+
 See the code repository's
 [source recipe](https://github.com/Quark-Medical/rpmem/blob/main/experiments/phase1/formal_sources_v1.yaml)
 for source revisions, adapters, and filtering parameters.
@@ -32,10 +39,15 @@ for source revisions, adapters, and filtering parameters.
 
 | Dataset | Upstream source | Prepared version |
 | --- | --- | --- |
-| PERMA | [ustclsc/PERMA](https://github.com/ustclsc/PERMA) | Seven variants, ten users |
+| PERMA | [ustclsc/PERMA](https://huggingface.co/datasets/ustclsc/PERMA) | Seven variants, ten users |
 | PersonaMem-v2 | [bowen-upenn/PersonaMem-v2](https://huggingface.co/datasets/bowen-upenn/PersonaMem-v2) | Text-only 32k histories |
 | PrefEval | [amazon-science/PrefEval](https://github.com/amazon-science/PrefEval) | Topic splits and history-length settings used by RPMem |
 
+PERMA's dataset card specifies Apache-2.0 and PersonaMem-v2's specifies CC-BY-4.0.
+[PrefEval's license](https://github.com/amazon-science/PrefEval/blob/50795054b5ff5f418d2b768a331d71e480f93331/LICENSE)
+is **CC-BY-NC-4.0**, including a noncommercial restriction. These terms apply
+to the corresponding data rather than to the entire mixed-source bundle.
+
 Benchmark metadata records source revisions and processing choices. This
 collection is a prepared input bundle, not a replacement for the upstream
 datasets or their authors' documentation. Consult the linked sources for their