Add 800-query evaluation set (expert-annotated trajectories) and dataset card

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+evaluation/testset.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,75 @@
 ---
 license: mit
+language:
+- en
+task_categories:
+- question-answering
+tags:
+- finance
+- tool-calling
+- function-calling
+- agent
+- trajectory
+- llm-evaluation
+size_categories:
+- n<1K
 ---
+
+# FinTrace
+
+**FinTrace** is a benchmark for evaluating LLM tool-calling (function-calling)
+agents on financial queries, built on the Financial Modeling Prep (FMP) MCP
+toolset. It contains expert-annotated multi-turn tool-calling trajectories
+across 30+ financial task categories, evaluated with a nine-metric rubric
+spanning action correctness, execution efficiency, process quality, and
+output quality.
+
+Paper: *FinTrace* (COLM 2026) — link coming soon.
+
+## Repository structure
+
+| Path | Description | Status |
+|---|---|---|
+| `evaluation/testset.json` | 800-query evaluation set with expert-annotated trajectories | ✅ available |
+| `trajectories/` | model-generated trajectories, one folder per model | 🔜 coming soon |
+| `training/` | FinTrace-Training SFT / DPO preference data | 🔜 coming soon |
+
+## Evaluation set (`evaluation/testset.json`)
+
+800 entries (440 labeled `GOOD`, 360 labeled `BAD` by expert annotators), each
+with:
+
+| Field | Description |
+|---|---|
+| `id` | unique query id |
+| `source_query` | the financial question |
+| `task_type` / `task_type_bucket` | task category (32 types / 12 buckets) |
+| `resource` / `data_source` | origin of the query |
+| `difficulty_score` / `difficulty_tier` | difficulty annotation |
+| `trajectory` | full multi-turn message list (reasoning, tool calls, tool responses) |
+| `final_answer` | the trajectory's final answer |
+| `reference_answer` | ground-truth answer |
+| `endpoints_called` | FMP endpoints invoked in the trajectory |
+| `num_turns` / `traj_len_bin` | trajectory length statistics |
+| `label` | expert quality label (`GOOD` / `BAD`) |
+| `bad_reasons` / `reason_summary` | failure taxonomy and rationale for `BAD` trajectories |
+
+## Usage
+
+The trajectories are deeply nested, so we recommend downloading the raw JSON
+directly rather than `load_dataset`:
+
+```python
+import json
+from huggingface_hub import hf_hub_download
+
+path = hf_hub_download(
+    "YupengCao/FinTrace", "evaluation/testset.json", repo_type="dataset"
+)
+data = json.load(open(path))
+print(len(data), data[0]["source_query"])
+```
+
+## Citation
+
+Citation entry coming with the camera-ready release.

evaluation/testset.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af27e4aff42d88ca43d1ad379a1d8d922385aca3ee191f57df2ca74656f9e24a
+size 38637588