Datasets:
id int64 3 15.1k | source_query stringlengths 23 2.46k | task_type stringclasses 32
values | task_type_bucket stringclasses 12
values | resource stringclasses 5
values | data_source stringclasses 2
values | difficulty_tier stringclasses 3
values | difficulty_score float64 0 36 | traj_len_bin stringclasses 3
values | endpoints_called listlengths 0 16 | reference_answer stringlengths 5 17k | output_answer stringlengths 10 17k | reasoning stringlengths 0 4.53k | output_trajectory listlengths 1 7 | golden_trajectories listlengths 3 103 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
5,146 | What was the average of the total deductions for the years 2004, 2005, and 2006? | Reasoning QA | Reasoning_QA | FinCot | finmcp | easy | 3 | medium | [
"income-statement"
] | The average of the total deductions for the years 2004, 2005, and 2006 is 9,813.67. | Assuming “total deductions” refers to the closest available FMP income-statement field for Apple, namely <b>netIncomeDeductions</b>, the average for <b>2004, 2005, and 2006</b> is:
<b>0</b>
This is based on the most relevant available FMP financial field, since Apple’s statements in FMP do not expose a separate line ... | Answer as closely as possible for Apple using available FMP data.
known_information: The annual income statement data for AAPL is available, but the tool output is truncated and does not visibly include 2004-2006 rows. The most relevant field matching “total deductions” in the available schema is netIncomeDeductions, w... | [
{
"turn": 0,
"reasoning": "Identify what “total deductions” refers to and obtain values for 2004, 2005, and 2006.\nknown_information: The user asks for the average of “total deductions” for 2004, 2005, and 2006, but provides no company, ticker, dataset, or statement context.\nmissing_information: The entity... | [
{
"role": "system",
"content": "You are a financial assistant, please answer user's query. If needed, you can call the tools from MCP server."
},
{
"role": "user",
"content": "What was the average of the total deductions for the years 2004, 2005, and 2006?"
},
{
"role": "assistant",
... |
14,516 | I need a 5-year historical performance report comparing the S&P 500, NASDAQ Composite, and Dow Jones Industrial Average. First, confirm the exact index symbols from the index list. Then retrieve historical monthly closes for each index for the past 5 years and calculate monthly total returns, annualized return, annuali... | Index & Benchmark | Sector_Index_ETF | generated_query | self_generate | medium | 11 | medium | [
"indexes-list",
"all-index-quotes",
"index-historical-price-eod-full",
"index-quote"
] | Assumption: I used the standard price index symbols from FMP, not total-return variants, because you asked for the S&P 500, NASDAQ Composite, and DJIA specifically. So the return figures are price-return based on monthly closes, not dividend-reinvested total returns.
**Confirmed FMP index symbols**
- **S&P 500:** `^GS... | Assumption: I used the standard price index symbols from FMP, not total-return variants, because you asked for the S&P 500, NASDAQ Composite, and DJIA specifically. So the return figures are price-return based on monthly closes, not dividend-reinvested total returns.
**Confirmed FMP index symbols**
- **S&P 500:** `^GS... | Provide the final report with calculations based on retrieved data and reasonable assumptions.
known_information: Exact symbols confirmed by FMP data/quotes: ^GSPC S&P 500, ^IXIC NASDAQ Composite, ^DJI Dow Jones Industrial Average. Current quotes retrieved. Historical daily closes over 2021-03-27 to 2026-03-27 availabl... | [
{
"turn": 0,
"reasoning": "Confirm exact symbols for S&P 500, NASDAQ Composite, and Dow Jones Industrial Average, then gather 5-year historical index data and current snapshots.\nknown_information: User wants index symbols confirmed from index list; then historical monthly closes for past 5 years; metrics: ... | [
{
"role": "system",
"content": "You are a financial assistant operating in an interactive environment with access to external tools.\nYour goal is to answer user queries accurately and efficiently.\n"
},
{
"role": "user",
"content": "I need a 5-year historical performance report comparing the S&... |
2,865 | what was the average dividend payment per share of common stock that was repurchased , per quarter? | Reasoning QA | Reasoning_QA | FinCot | finmcp | medium | 6 | short | [
"cashflow-statement",
"income-statement",
"key-metrics"
] | To determine the average dividend payment per share of common stock that was repurchased, per quarter, in 2015, let’s break down the numbers provided:
1. **Total Dividend Payment in 2015**: The company paid a total of $195.5 million in dividends for the year.
2. **Quarterly Dividend Payment**: Generally, dividend ... | Assuming you meant **Apple (AAPL)** and using the **most recent four reported quarters**, the **average dividend payment per common share per quarter** was **about $0.255/share**.
Method used:
- For each quarter: **common dividends paid ÷ weighted average common shares outstanding**
- Then averaged across the quarters... | Compute a concise answer from observed quarterly dividend and repurchase data for the assumed company.
known_information: Assumed company is AAPL due to ambiguity. Cash flow data provides commonDividendsPaid and commonStockRepurchased by quarter. Income statement provides weightedAverageShsOut. For recent four complete... | [
{
"turn": 0,
"reasoning": "Identify a reasonable company assumption and retrieve quarterly dividend-per-share and common-stock-repurchase figures needed to compute an average per quarter.\nknown_information: The user asks for \"average dividend payment per share of common stock that was repurchased, per qua... | [
{
"role": "system",
"content": "You are a financial assistant, please answer user's query. If needed, you can call the tools from MCP server."
},
{
"role": "user",
"content": "what was the average dividend payment per share of common stock that was repurchased , per quarter?"
},
{
"role"... |
8,240 | Pull all U.S. Senate disclosures for Senator Elizabeth Warren's trades in Microsoft (MSFT) from Jan 1, 2020 through Dec 31, 2022. For each trade date, show the analyst EPS estimates for the next fiscal quarter that were in effect on that date and compare them to the estimates 30 days prior — indicate whether consensus ... | Political Trading | Trading_Market | generated_query | self_generate | medium | 7 | medium | [
"senate-trading-by-name",
"senate-latest"
] | No Elizabeth Warren U.S. Senate disclosure records for Microsoft (MSFT) were found in the available FMP Senate-trading dataset for the period 2020-01-01 through 2022-12-31.
Because no matching MSFT trade disclosures were returned, there are no trade dates on which to compute:
- the then-current consensus EPS estimate ... | No Elizabeth Warren U.S. Senate disclosure records for Microsoft (MSFT) were found in the available FMP Senate-trading dataset for the period 2020-01-01 through 2022-12-31.
Because no matching MSFT trade disclosures were returned, there are no trade dates on which to compute:
- the then-current consensus EPS estimate ... | Provide the best answer based on available Senate disclosure search results.
known_information: senate-trading-by-name returned no results for 'Elizabeth Warren' and for 'Warren'. No Elizabeth Warren MSFT Senate trade disclosures were found via available FMP Senate disclosure tools. Without trade dates, EPS estimate co... | [
{
"turn": 0,
"reasoning": "Identify Elizabeth Warren's Senate MSFT trades between 2020-01-01 and 2022-12-31, and obtain analyst EPS estimates for the next fiscal quarter on each trade date and 30 days earlier.\nknown_information: User specified senator Elizabeth Warren, ticker MSFT, and date range 2020-01-0... | [
{
"role": "system",
"content": "You are a financial assistant operating in an interactive environment with access to external tools.\nYour goal is to answer user queries accurately and efficiently.\n"
},
{
"role": "user",
"content": "Pull all U.S. Senate disclosures for Senator Elizabeth Warren'... |
8,007 | "Give me a chronological timeline of regulatory and legal news involving the U.S. SEC and major cryp(...TRUNCATED) | Cryptocurrency | Alt_Assets_Forex | generated_query | self_generate | hard | 12 | medium | ["search-crypto-news","cryptocurrency-historical-price-eod-full","search-by-symbol","search-by-name"(...TRUNCATED) | "Assumption: I interpret “major crypto exchanges or stablecoins” to include Binance, Coinbase, K(...TRUNCATED) | "Assumption: I interpret “major crypto exchanges or stablecoins” to include Binance, Coinbase, K(...TRUNCATED) | "Provide the best possible timeline and correlation summary using only already observed tool outputs(...TRUNCATED) | [{"turn":0,"reasoning":"Gather regulatory/legal news involving the SEC and major crypto exchanges or(...TRUNCATED) | [{"role":"system","content":"You are a financial assistant operating in an interactive environment w(...TRUNCATED) |
3,609 | "what was the percentage increase in the carrying vale of the furniture and equipment from 2004 to 2(...TRUNCATED) | Reasoning QA | Reasoning_QA | FinCot | finmcp | hard | 14 | long | ["search-name","financial-reports-form-10-k-json","financial-reports-dates","search-by-symbol","sear(...TRUNCATED) | "The percentage increase in the carrying value of the furniture and equipment from 2004 to 2005 is a(...TRUNCATED) | "I can’t determine the percentage increase reliably from the information available because the com(...TRUNCATED) | "Provide the best possible answer with current evidence and disclose assumption/limitation.\nknown_i(...TRUNCATED) | [{"turn":0,"reasoning":"Identify the company and retrieve 2004 and 2005 carrying value data for furn(...TRUNCATED) | [{"role":"system","content":"You are a financial assistant, please answer user's query. If needed, y(...TRUNCATED) |
14,762 | "For ticker AAPL over the last 12 months, list every period where insiders were net buyers (dates an(...TRUNCATED) | Technical Analysis & Backtesting | Trading_Market | generated_query | self_generate | hard | 11.5 | medium | ["insider-trade-statistics","search-insider-trades","relative-strength-index","average-directional-i(...TRUNCATED) | "Assumption: I interpret “period” as the net insider-buy reporting period available from FMP ins(...TRUNCATED) | "Assumption: I interpret “period” as the net insider-buy reporting period available from FMP ins(...TRUNCATED) | "Provide the final synthesized answer listing net insider-buy periods in the last 12 months with RSI(...TRUNCATED) | [{"turn":0,"reasoning":"Gather insider trading stats and technical indicators for AAPL over the last(...TRUNCATED) | [{"role":"system","content":"You are a financial assistant operating in an interactive environment w(...TRUNCATED) |
10,564 | "For GOOGL over the past 8 quarters, find instances where analysts consecutively lowered the next-qu(...TRUNCATED) | Technical Analysis & Backtesting | Trading_Market | generated_query | self_generate | hard | 11.5 | medium | ["financial-reports-dates","financial-estimates","as-reported-income-statements","relative-strength-(...TRUNCATED) | "Assumption: I used the most recent 8 completed GOOGL quarters available from the as-reported income(...TRUNCATED) | "Assumption: I used the most recent 8 completed GOOGL quarters available from the as-reported income(...TRUNCATED) | "Answer with the best possible result using available data while noting the key limitation on revisi(...TRUNCATED) | [{"turn":0,"reasoning":"Gather the last 8 quarters for GOOGL, analyst EPS estimate revision history (...TRUNCATED) | [{"role":"system","content":"You are a financial assistant operating in an interactive environment w(...TRUNCATED) |
7,971 | "Which S&P 500 companies announced earnings between January 1, 2024 and March 31, 2024 that missed E(...TRUNCATED) | Earnings Analysis | Corporate_Events | generated_query | self_generate | easy | 3 | medium | [
"sp-500",
"earnings-calendar"
] | "Using the FMP data I was able to retrieve for the 2024-01-01 to 2024-03-31 earnings calendar, the v(...TRUNCATED) | "Using the FMP data I was able to retrieve for the 2024-01-01 to 2024-03-31 earnings calendar, the v(...TRUNCATED) | "Provide the best possible answer from the observed earnings-calendar pages by listing S&P 500 const(...TRUNCATED) | [{"turn":0,"reasoning":"Identify S&P 500 companies with earnings announcement dates between 2024-01-(...TRUNCATED) | [{"role":"system","content":"You are a financial assistant, please answer user's query. If needed, y(...TRUNCATED) |
8,848 | "Compare weekly price performance of Ethereum and Litecoin for calendar year 2021 with managed money(...TRUNCATED) | Cryptocurrency | Alt_Assets_Forex | generated_query | self_generate | medium | 11 | medium | [
"cryptocurrency-list",
"COT-report-list",
"cryptocurrency-historical-price-eod-light",
"COT-report"
] | "Assumption used: I aligned everything to the COT reporting week by using Tuesday-to-Tuesday crypto (...TRUNCATED) | "Assumption used: I aligned everything to the COT reporting week by using Tuesday-to-Tuesday crypto (...TRUNCATED) | "Provide the requested analytics from gathered data.\nknown_information: ETHUSD and LTCUSD daily 202(...TRUNCATED) | [{"turn":0,"reasoning":"Gather 2021 historical weekly proxy data for Ethereum and Litecoin plus mana(...TRUNCATED) | [{"role":"system","content":"You are a financial assistant operating in an interactive environment w(...TRUNCATED) |
End of preview. Expand in Data Studio
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-curated 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 output and golden 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. Each entry pairs a model output trajectory with a golden reference trajectory (the best of three frontier-model runs, selected by an LLM judge) for the same query:
| 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 |
traj_len_bin |
trajectory length bin |
endpoints_called |
FMP endpoints invoked in the output trajectory |
reference_answer |
ground-truth answer |
output_answer |
final answer of the output trajectory |
reasoning |
reasoning summary of the output trajectory |
output_trajectory |
full multi-turn message list (reasoning, tool calls, tool responses) |
golden_trajectories |
golden reference trajectory for the same query |
Usage
The trajectories are deeply nested, so we recommend downloading the raw JSON
directly rather than load_dataset:
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.
- Downloads last month
- 26