Computer-Use Agent Taxonomy中文
R04.01.02.02 · TASK CLASS

Statistical portals & interactive charts

Build queries in statistics portals and read values off interactive charts.

20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.

Typical applications
data portals · Our World in Data · FRED
Class GUI-need label
G3 · GUI-only

Task proposals · 20

Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.

R04.01.02.02-T01 Average inflation from CPI series

Original instruction

On the FRED snapshot, open the CPI-U All Items (not seasonally adjusted) series, set the range to Jan 2019 - Dec 2024, and compute the average year-over-year inflation rate across that span. Write the percentage to answer.txt (one decimal).

Platforms
web · Any OS (web)
Applications
FRED
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.txt value is within +/-0.1 pt of the reference computed from the pinned snapshot series. Values read from the portal chart/table; snapshot is deterministic. The FRED snapshot issues no API key and serves series data only to browser sessions (session cookie + CSRF); its request log must show the chart range change behind the answer.

Proposed partial credit

40% correct series+range; 30% YoY computed; 30% average within tolerance

R04.01.02.02-T02 County median income CSV export

Original instruction

In the census-style table builder snapshot, build a table of county-level median household income for a specified state (latest ACS year) and export it to income.csv with columns county, median_household_income.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

income.csv row count equals the number of counties and each value matches the reference table from the snapshot. Graded against the reference CSV. The portal snapshot issues no API key and serves data only to browser sessions (session cookie + CSRF); its request log must show the table-builder requests that produced income.csv.

Proposed partial credit

40% correct geography+measure; 30% all county rows; 30% values match

R04.01.02.02-T03 Three-indicator data cube download

Original instruction

In the World Bank DataBank snapshot, build a query of three named indicators for ten named countries over 2000-2020 and download it as cube.csv (rows = country-year, columns = the three indicators).

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP02 · Records/table
Conditions
long · Long-horizon
Proposed difficulty
4 / 5

Proposed verifier

cube.csv has the correct 10x21 country-year rows with the three indicator columns and values matching the reference cube. Graded against the reference CSV. The DataBank snapshot issues no API key (its v2 API paths return 403) and serves data only to browser sessions; its request log must show the query-builder requests that produced cube.csv.

Proposed partial credit

30% correct dimensions selected; 30% full row/column set; 40% values match

R04.01.02.02-T04 Read a value off an OWID chart

Original instruction

On the Our World in Data snapshot chart named in task.txt, hover to read the value for the specified country in the specified year. Write the value to answer.txt (units as shown).

Platforms
web · Any OS (web)
Applications
Our World in Data
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

answer.txt value is within +/-0.5% of the snapshot's charted value for that country-year. The value is only in the chart tooltip (no table/API on the fixture), so visual reading is required. The OWID snapshot has no grapher CSV/JSON or API endpoint and serves chart data only to browser sessions; its request log must show the chart hover view behind the answer.

Proposed partial credit

50% correct country-year located; 50% value within tolerance

R04.01.02.02-T05 Unemployment rate at a month

Original instruction

On the FRED snapshot, open the civilian unemployment rate series and read the value for the specific month in task.txt from the chart. Write the percentage to answer.txt.

Platforms
web · Any OS (web)
Applications
FRED
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

answer.txt value matches the snapshot series value for that month within +/-0.1 pt. Read from the chart tooltip; deterministic snapshot. The FRED snapshot issues no API key and serves series data only to browser sessions (session cookie + CSRF); its request log must show the chart view behind the answer.

Proposed partial credit

50% correct series+month; 50% value within tolerance

R04.01.02.02-T06 Eurostat dimension-filtered value

Original instruction

On the Eurostat snapshot, open the dataset named in task.txt, filter the dimensions (country, sex, age group, year) to the specified combination, and read the single value. Write it to answer.txt with units.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.txt equals the snapshot value for the exact dimension combination within tolerance. Reached via the portal's dimension selectors; deterministic. The Eurostat snapshot's SDMX/JSON API paths return 403 and data is served only to browser sessions; its request log must show the dimension-filter requests behind the answer.

Proposed partial credit

50% correct dimension filtering; 50% exact value

R04.01.02.02-T07 Percent change between two years

Original instruction

On the OWID snapshot chart in task.txt, read the values for a named country in two specified years and compute the percent change between them. Write the percent change to answer.txt (one decimal).

Platforms
web · Any OS (web)
Applications
Our World in Data
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.txt percent change is within +/-0.5 pt of the value computed from the snapshot's two charted points. Values read via tooltip; deterministic. The OWID snapshot has no grapher CSV/JSON or API endpoint and serves chart data only to browser sessions; its request log must show the chart view behind the answer.

Proposed partial credit

30% first value; 30% second value; 40% correct percent change

R04.01.02.02-T08 Build a custom ratio series in FRED

Original instruction

On the FRED snapshot, create a custom series equal to the ratio of two named series (e.g. series A divided by series B) and read the value of the ratio for the latest available month. Write it to answer.txt (three decimals).

Platforms
web · Any OS (web)
Applications
FRED
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment type · Text & forms
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

answer.txt ratio for the latest month matches the reference computed from the snapshot within 0.5%. Requires building the transformed series in the portal UI; deterministic. The FRED snapshot issues no API key and serves series data only to browser sessions (session cookie + CSRF); its request log must show the edit-graph formula requests behind the answer.

Proposed partial credit

40% both series added; 30% ratio transform applied; 30% latest value correct

R04.01.02.02-T09 Compare two countries on a grouped chart

Original instruction

On the OWID snapshot chart in task.txt, with two named countries selected, read each country's value for the specified year and report which is higher and by how much. Write the higher country and the difference to answer.json.

Platforms
web · Any OS (web)
Applications
Our World in Data
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.json names the correct higher country and the difference within +/-1% of the snapshot values. Read via tooltip; deterministic. The OWID snapshot has no grapher CSV/JSON or API endpoint and serves chart data only to browser sessions; its request log must show the two-country chart view behind the answer.

Proposed partial credit

30% both values; 30% correct higher country; 40% difference within tolerance

R04.01.02.02-T10 Regional GDP per capita latest value

Original instruction

On the World Bank snapshot, look up GDP per capita (current US$) for a named country for the latest available year by querying the indicator and reading the value. Write the value and the year to answer.json.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.json value and year match the snapshot's latest data point within 0.5%. Reached via the indicator query UI; deterministic. The World Bank snapshot issues no API key (its v2 API paths return 403) and serves data only to browser sessions; its request log must show the indicator query behind the answer.

Proposed partial credit

40% correct indicator+country; 30% latest year; 30% value within tolerance

R04.01.02.02-T11 Filtered subset CSV with exact rows

Original instruction

On the data-portal snapshot, apply the filters in task.txt (category, year range, region) to the dataset and download the filtered result to subset.csv. The row count and values must match the filtered reference.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

subset.csv has the exact row count of the filtered reference and matching values. Graded against the reference CSV. The portal snapshot issues no API key and serves data only to browser sessions (session cookie + CSRF); its request log must show the filter requests that produced subset.csv.

Proposed partial credit

40% filters applied; 30% correct row count; 30% values match

R04.01.02.02-T12 Find the peak year from a chart

Original instruction

On the OWID snapshot chart in task.txt for a named country, determine the year in which the metric reached its maximum over the shown period by inspecting the chart. Write the peak year and its value to answer.json.

Platforms
web · Any OS (web)
Applications
Our World in Data
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.json peak year equals the snapshot's maximum and the value is within +/-1%. Read by scanning the chart tooltips; deterministic. The OWID snapshot has no grapher CSV/JSON or API endpoint and serves chart data only to browser sessions; its request log must show the chart view behind the answer.

Proposed partial credit

60% correct peak year; 40% peak value within tolerance

R04.01.02.02-T13 Seasonal adjustment difference

Original instruction

On the FRED snapshot, compare the seasonally adjusted and not-seasonally-adjusted versions of a named series and report the difference between them at the month in task.txt. Write SA, NSA, and the difference to answer.json.

Platforms
web · Any OS (web)
Applications
FRED
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

answer.json SA and NSA values match the snapshot series for that month within tolerance and the difference is consistent. Requires selecting both series variants in the UI; deterministic. The FRED snapshot issues no API key and serves series data only to browser sessions (session cookie + CSRF); its request log must show the requests for both the SA and NSA series behind the answer.

Proposed partial credit

35% SA value; 35% NSA value; 30% difference

R04.01.02.02-T14 Read a choropleth map value

Original instruction

On the OWID snapshot map view in task.txt, switch to the map tab, set the year, and hover a named region to read its value. Write the value to answer.txt with units.

Platforms
web · Any OS (web)
Applications
Our World in Data
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.txt value matches the snapshot map value for that region-year within +/-1%. The value is only on the map hover tooltip; visual reading required. The OWID snapshot has no grapher CSV/JSON or API endpoint and serves chart data only to browser sessions; its request log must show the map-tab view for that year behind the answer.

Proposed partial credit

40% correct map/year; 30% region located; 30% value within tolerance

R04.01.02.02-T15 Read a component off a stacked bar

Original instruction

On the data-portal snapshot stacked-bar chart in task.txt, read the value of a specified component (series) within a specified category/bar by hovering that segment. Write the value to answer.txt.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

answer.txt equals the snapshot's segment value within +/-1%. The segment value is only in the hover tooltip; visual reading required. The portal snapshot issues no API key and serves data only to browser sessions (session cookie + CSRF); its request log must show the chart view behind the answer.

Proposed partial credit

50% correct segment located; 50% value within tolerance

R04.01.02.02-T16 Multi-series query export matches cube

Original instruction

On the portal snapshot, build a query of two indicators across five regions for a 10-year range and export to cube.csv in long format (region, year, indicator, value). The export must match the reference cube.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP02 · Records/table
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

cube.csv has the exact set of region-year-indicator rows and matching values vs the reference cube. Graded against the reference CSV. The portal snapshot issues no API key and serves data only to browser sessions (session cookie + CSRF); its request log must show the query-builder requests that produced cube.csv.

Proposed partial credit

40% correct query dimensions; 30% full row set; 30% values match

R04.01.02.02-T17 Compute CAGR from chart endpoints

Original instruction

On the OWID/FRED snapshot chart in task.txt, read the start-year and end-year values for a named series and compute the compound annual growth rate (CAGR) over that span. Write the CAGR percent to answer.txt (two decimals).

Platforms
web · Any OS (web)
Applications
Our World in Data · FRED
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll hover · Hover & context vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

answer.txt CAGR is within +/-0.1 pt of the value computed from the snapshot endpoints. Endpoint values read from the chart; deterministic. Neither snapshot issues an API key or CSV/JSON data endpoint and both serve data only to browser sessions; their request logs must show the chart view behind the answer.

Proposed partial credit

30% start value; 30% end value; 40% correct CAGR

R04.01.02.02-T18 Drill into the largest pie slice

Original instruction

On the dashboard snapshot in task.txt, find the largest slice of the category pie chart, click to drill down, and report the top sub-category revealed and its value. Write both to answer.json.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

answer.json names the correct largest slice's top sub-category and its value (within tolerance) per the snapshot. Requires the drill-down interaction; deterministic. The portal snapshot issues no API key and serves data only to browser sessions (session cookie + CSRF); its request log must show the drill-down request behind the answer.

Proposed partial credit

40% largest slice identified; 30% drill-down; 30% sub-category + value

R04.01.02.02-T19 Multi-step query with computed metric

Original instruction

On the portal snapshot, filter a dataset to a specified subgroup, pivot it to get per-year values for two indicators, compute a derived metric defined in task.txt (e.g. a per-capita ratio) for the latest year, and write the metric and the two inputs to answer.json.

Platforms
web · Any OS (web)
Applications
data portals
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

answer.json derived metric is within tolerance of the reference computed from the snapshot, with correct input values. Requires multi-step portal navigation + pivot; deterministic. The portal snapshot issues no API key and serves data only to browser sessions (session cookie + CSRF); its request log must show the filter and pivot requests behind the answer.

Proposed partial credit

30% correct filtered subgroup; 30% both indicator values; 40% derived metric

R04.01.02.02-T20 Cross-portal combined value

Original instruction

Read a numerator from one portal snapshot (e.g. total of a metric for a region) and a denominator from a second portal snapshot (e.g. population), then compute the ratio defined in task.txt. Write the ratio and the two source values (with the portal each came from) to answer.json.

Platforms
web · Any OS (web)
Applications
data portals · FRED
Interfaces
brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product
WP12 · Answer/decision
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

answer.json ratio is within tolerance of the reference computed from the two snapshots, with both source values correct and attributed. Requires reading two separate portals; deterministic. Neither portal snapshot issues an API key and both serve data only to browser sessions; their request logs must show the views from which each source value was read.

Proposed partial credit

30% numerator value; 30% denominator value; 40% correct combined ratio

Mapped research references

Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.

In this workflow family