Dataset Viewer
Auto-converted to Parquet Duplicate
id
string
purpose
string
notes
string
persona
null
reason_for_call
string
known_info
string
unknown_info
null
task_instructions
string
tool_path
string
num_actions
int64
actions
string
nl_assertions
string
annotations_path
string
0_baseline
Tests whether the agent correctly retrieves and differentiates weather bulletins and alerts for multiple departments to address a user's safety and planning concerns.
The scenario is plausible as a user interaction: the user is checking weather warnings for different departments sequentially, motivated by concern about their local weather conditions.
null
The user wants to check current weather warnings for department 75. They are also checking department 13 to ensure the safety of their broader area.
Your name is Sophie Bernard.
null
You communicate clearly and directly.
bulletin → alerts
2
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
1_baseline
Tests whether the agent correctly chains weather bulletin retrieval, geocoding, and multi-day forecast tools to address a detailed weather inquiry.
The scenario is plausible as a user interaction: the user seeks weather warnings for a department, then follows up by asking for more detailed information about a specific city, using the GPS coordinates to get a long-term forecast, and concludes with a natural expression of satisfaction.
null
The user is concerned about potential weather disruptions in their area and wants to understand the risks involved. They are located in department 75 and are planning activities that depend on stable weather conditions over the coming days.
Your name is Karim Mansouri.
null
You communicate clearly and directly.
bulletin → search_places → forecast
3
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → End_of_Simulation
2_baseline
Tests whether the agent correctly chains geocoding, multi-day forecast, and alert bulletin tools to answer a detailed weather inquiry for a specific location.
The scenario represents a plausible user interaction where a user seeks weather-related information for a specific location, first by identifying its coordinates and INSEE code, then checking a multi-day forecast and a meteorological alert bulletin.
null
Un utilisateur souhaite obtenir des informations météorologiques pour organiser un événement à Paris. Il a besoin de vérifier à la fois les conditions climatiques à long terme et les alertes météo potentielles pour le département concerné.
Your name is Élodie Martin.
null
You communicate clearly and directly.
search_places → forecast → bulletin
3
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}]
[]
Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
3_baseline
Tests whether the agent correctly handles sequential weather alert checks and provides a detailed weather bulletin when requested.
The scenario is plausible as a user interaction: the user is checking weather alerts for multiple departments and then requests a detailed weather bulletin for another department, showing a realistic pattern of inquiry and curiosity.
null
The user is checking weather alerts for departments 75 and 44 to assess risks for their planned outdoor activities. They also want to review the weather bulletin for department 13 to confirm conditions for their travel or work plans.
Your name is Jean-Pierre Lefebvre.
null
You communicate clearly and directly.
alerts → alerts → bulletin
3
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}]
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
4_baseline
Tests whether the agent correctly retrieves and reports weather warning bulletins and alerts for different departments in response to a user's sequential inquiries.
The scenario is plausible as a user interaction, as it represents a user checking weather warnings for multiple departments sequentially, which is a realistic behavior for someone seeking localized weather information.
null
The user wants to check weather warnings for department 75, department 13, and department 06. They are assessing weather-related risks in these specific areas to make informed decisions.
Your name is Amina Traoré.
null
You communicate clearly and directly.
bulletin → alerts → bulletin
3
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_bulletin", "arguments": {"domain": "06"}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
5_baseline
Tests whether the agent correctly chains weather bulletin, weather alerts, and location search tools to answer a weather-related inquiry with location-specific details.
The scenario is plausible as a user interaction: the user is checking weather warnings for two different departments and seeks more specific information by identifying a city and its INSEE code to better understand the alerts.
null
The user is concerned about potential weather risks and wants to understand the current situation in their area. They are seeking detailed and location-specific information to make informed decisions about monitoring or preparing for weather events.
Your name is Lucas Petit.
null
You communicate clearly and directly.
bulletin → alerts → search_places → alerts
4
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_alerts", "arguments": {"domain": "13", "depth": 1}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → End_of_Simulation
6_baseline
Tests whether the agent correctly handles a sequence of weather-related queries, including alerts, location search, long-term forecast, and detailed bulletin retrieval.
The scenario is plausible as a user interaction: the user first inquires about weather alerts in department 75, then seeks specific location data for Paris to get a 15-day forecast, and finally asks for a weather bulletin for department 13. Each step follows a logical progression of information needs.
null
The user is concerned about potential weather disruptions and wants to check the current alerts in their area. They also plan a trip to Paris and need a detailed forecast, and later express interest in the weather conditions of another department for planning purposes.
Your name is Nathalie Rousseau.
null
You communicate clearly and directly.
alerts → search_places → forecast → bulletin
4
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}]
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
7_baseline
Tests whether the agent correctly retrieves and reports weather warning bulletins and current alerts for different departments in a sequential, user-driven manner.
The scenario is plausible as a user interaction because it represents a user checking weather warnings for multiple departments sequentially, likely to assess potential risks in different locations they care about.
null
The user is checking weather warnings for departments 75, 13, 06, and 69 to prepare for potential weather-related disruptions. They need information about current and upcoming weather alerts in these specific regions.
Your name is Youssef El Amrani.
null
You communicate clearly and directly.
bulletin → alerts → bulletin → alerts
4
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_bulletin", "arguments": {"domain": "06"}}, {"name": "meteo_alerts", "arguments": {"domain": "69"}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
8_baseline
Tests whether the agent correctly provides weather alerts for multiple departments in response to sequential user inquiries.
The scenario is plausible as a user interaction: the user is checking weather alerts for multiple departments sequentially, which represents a realistic pattern of information-seeking behavior.
null
The user wants to check weather alerts for departments 75, 13, and 69 to assess any risks affecting their planned activities. They also need the current weather bulletin for department 75 and information about the location of Paris.
Your name is Claire Dubois.
null
You communicate clearly and directly.
alerts → search_places → bulletin → alerts → alerts
5
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_alerts", "arguments": {"domain": "69"}}]
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
9_baseline
Tests whether the agent correctly retrieves and provides weather warning bulletins, current weather observations, and rain nowcasts for location-specific weather inquiries.
The scenario represents a plausible user interaction where the user is checking weather warnings and current conditions across multiple departments. The repeated use of the weather bulletin tool is justified by the user's desire to compare different regions.
null
The user is planning activities and wants to check the weather risk levels in various departments. They are particularly interested in current and near-term weather conditions to make informed decisions.
Your name is Thomas Girard.
null
You communicate clearly and directly.
bulletin → observation → rain_nowcast → bulletin → alerts → bulletin
6
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_observation", "arguments": {"lat": 48.8566, "lon": 2.3522}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.8566, "lon": 2.3522}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain...
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_observation → meteo_rain_nowcast → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
10_baseline
Tests whether the agent correctly chains weather alerts, geocoding, and weather bulletin tools to address a user's location-specific weather concerns.
The scenario is plausible as a user interaction: the user is checking weather alerts for two different departments and then confirming the weather bulletin for each by requesting GPS coordinates and INSEE codes for specific cities.
null
The user is planning outdoor activities and wants to ensure there are no significant weather risks in their area. They have prior knowledge of department codes and are seeking reassurance for two specific locations.
Your name is Fatou Diallo.
null
You communicate clearly and directly.
alerts → search_places → bulletin → alerts → search_places → bulletin
6
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_bu...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → End_of_Simulation
11_baseline
Tests whether the agent correctly handles sequential weather alert checks across multiple departments and uses location data to infer the relevant department for weather alerts.
The scenario is plausible as a user interaction: the user is systematically checking weather alerts for multiple departments and then seeks information about a specific city (Marseille) to check its associated department. The flow is logical and reflects a realistic information-seeking behavior.
null
The user is planning a trip across several regions in France and wants to ensure there are no significant weather alerts that could affect their travel plans. They start by checking the weather alerts for department 75 and continue to other departments to compare conditions.
Your name is Mathieu Leroy.
null
You communicate clearly and directly.
alerts → alerts → alerts → search_places → alerts → alerts
6
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_alerts", "arguments": {"domain": "33"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_alerts", "arguments": {"domain": "13", "depth": 1}}, {"name": "met...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
12_baseline
Tests whether the agent correctly chains weather warning bulletins, location searches, and rain nowcasts to support multi-department weather monitoring and travel planning.
The scenario is plausible as a user interaction: the user is checking weather warnings for multiple departments and verifying the weather in Lyon before planning a trip, showing a realistic pattern of information-seeking behavior.
null
The user is planning a trip and wants to ensure that there are no significant weather risks in the destination department. They also want to verify the current and near-term weather conditions for a specific city to finalize their plans.
Your name is Isabelle Moreau.
null
You communicate clearly and directly.
bulletin → alerts → bulletin → alerts → search_places → rain_nowcast → bulletin
7
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_bulletin", "arguments": {"domain": "06"}}, {"name": "meteo_alerts", "arguments": {"domain": "69"}}, {"name": "meteo_search_places", "arguments": {"q": "Lyon"}}, {"name": "meteo_rain_nowca...
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
13_baseline
Tests whether the agent correctly chains weather alerts, current weather, rain nowcast, and multi-day forecast tools to answer a user's comprehensive weather inquiry across multiple locations.
The scenario represents a plausible user interaction where the user is systematically checking weather alerts and forecasts for multiple French cities and departments, motivated by a need to assess potential weather-related risks or plan accordingly.
null
Your name is Samir Benali.
null
You communicate clearly and directly.
search_places → bulletin → alerts → alerts → alerts → search_places → observation → rain_nowcast → observation → forecast → alerts → bulletin
12
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "59"}}, {"name": "meteo_alerts", "arguments": {"domain": "13"}}, {"name": "meteo_search_plac...
[]
Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_observation → meteo_rain_nowcast → meteo_observation → meteo_forecast → Run_Conversation → meteo_alerts → Run_Con...
14_baseline
Tests whether the agent correctly handles sequential weather inquiries across multiple departments, including resolving location details and providing both alerts and forecasts.
The scenario represents a plausible user interaction where the user is systematically checking weather alerts and bulletins for multiple departments, and occasionally resolving location details (GPS coordinates and INSEE codes) to get more accurate forecasts. Each step logically follows from the previous one, with the ...
null
The user is planning multiple trips and events across different French departments and wants to ensure they are prepared for any potential weather conditions. They have prior knowledge of weather alert levels and are using the agent to efficiently gather and verify weather information for specific locations.
Your name is Marie-Claire Fontaine.
null
You communicate clearly and directly.
alerts → bulletin → search_places → bulletin → alerts → search_places → rain_nowcast → bulletin → alerts → bulletin → search_places → forecast → alerts
13
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_alerts", "arguments": {"domain": "06"}}, {"name": "meteo_searc...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_b...
15_baseline
Tests whether the agent correctly chains geocoding and weather-related tools (forecast, alerts, rain nowcast, bulletin) to answer a multi-faceted weather inquiry across different locations.
The scenario represents a plausible user interaction where the user is seeking detailed weather-related information for specific locations, including forecasts, alerts, and rain nowcasts. The repeated use of tools is justified by the user's need to check different locations and weather aspects.
null
The user is planning outdoor activities and needs to check the weather forecast, active alerts, and potential rain for several French cities. They are particularly concerned about the weather conditions in Paris and Lyon, and want to ensure there are no severe weather warnings before making decisions.
Your name is Antoine Dupont.
null
You communicate clearly and directly.
search_places → forecast → alerts → search_places → forecast → bulletin → search_places → alerts → rain_nowcast → bulletin
10
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_search_places", "arguments": {"q": "75001"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.862351, "lon": ...
[]
Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_rain_nowcast → meteo_bulletin → Run_Conversation → En...
16_baseline
Tests whether the agent correctly retrieves and provides weather warning bulletins for different departments, optionally using location data from search tools to enrich the context.
The scenario is plausible as a user interaction because it represents a user repeatedly checking weather bulletins for different departments, seeking detailed information by asking for associated location data and confirming the alerts for multiple regions.
null
The user is checking weather warnings for various French departments to prepare for potential weather-related disruptions. They are inquiring about multiple locations due to travel or regional responsibilities and want detailed, up-to-date information.
Your name is Layla Hamidi.
null
You communicate clearly and directly.
bulletin → search_places → forecast → bulletin → search_places → bulletin → bulletin → alerts → bulletin
9
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "M...
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_...
17_baseline
Tests whether the agent correctly chains geocoding and weather alert/bulletin tools to answer location-specific weather inquiries across multiple departments.
The scenario is plausible as a user interaction: the user is seeking weather-related information for multiple French cities and departments, using a consistent pattern of first identifying a location and then inquiring about weather conditions.
null
The user is planning a trip or managing an event in several French cities and wants to understand the current and upcoming weather conditions. They have prior knowledge of using department codes and are interested in both immediate and extended weather forecasts.
Your name is Pierre Legrand.
null
You communicate clearly and directly.
search_places → rain_nowcast → bulletin → bulletin → search_places → alerts → bulletin
7
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "Toulouse"}}...
[]
Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
18_baseline
Tests whether the agent correctly chains geocoding and multi-day forecast tools to answer a location-specific weather inquiry, and whether it can retrieve and present meteorological risk bulletins for the corresponding regions.
The scenario represents a plausible user interaction where the user systematically inquires about weather forecasts and meteorological risks for multiple cities, each time following a consistent pattern of first retrieving location data and then using it to fetch weather information.
null
The user is planning outdoor activities in several French cities and needs a reliable overview of the upcoming weather conditions and potential meteorological risks. They are familiar with using location identifiers like GPS coordinates and INSEE codes to ensure accurate and specific information.
Your name is Aïssatou Coulibaly.
null
You communicate clearly and directly.
search_places → forecast → bulletin → search_places → forecast → bulletin → search_places → forecast
8
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_forecast", "arguments": {"lat": 43.282, "lon...
[]
Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → End_of_Simulation
19_baseline
Tests whether the agent correctly chains geocoding, rain nowcasting, and weather alert tools to address a multi-step weather inquiry across different regions.
The scenario is plausible as a user interaction: the user first seeks specific weather-related information about Paris, then checks the weather alerts for two different departments, which represents a realistic and coherent information need.
null
The user wants to plan outdoor activities and needs to know the current and upcoming weather conditions for Paris and surrounding departments. They are located in France and are particularly interested in short-term rain forecasts and weather alerts for the evening hours.
Your name is Julien Morel.
null
You communicate clearly and directly.
search_places → rain_nowcast → bulletin → bulletin
4
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}]
[]
Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
20_baseline
Tests whether the agent correctly chains geocoding and weather observation tools to answer a multi-location weather inquiry.
The scenario represents a plausible user interaction where the user is systematically checking weather conditions for multiple French cities by first retrieving location data and then using it to fetch weather information.
null
The user is planning a trip or event across several French cities and wants to check the current and upcoming weather conditions to prepare accordingly. They have a basic understanding of how to retrieve location data and use it for weather-related queries.
Your name is Sophie Bernard.
null
You communicate clearly and directly.
search_places → bulletin → alerts → bulletin → bulletin → alerts → search_places → observation
8
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_bulletin", "arguments": {"domain": "31"}}, {"name": "meteo_alerts"...
[]
Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_observation → Run_Conversation → End_of_Simulation
21_baseline
Tests whether the agent correctly handles sequential weather inquiries across multiple departments and uses geocoding to provide localized nowcasts and bulletins.
The scenario is plausible as a user interaction: the user is systematically checking weather alerts and bulletins for multiple departments, which reflects a realistic information need for someone planning or monitoring weather conditions across different regions.
null
The user is seeking to understand the current and upcoming weather conditions in various departments for planning purposes. They are particularly interested in whether any weather alerts or risks are in effect and are checking multiple locations to ensure comprehensive coverage.
Your name is Karim Mansouri.
null
You communicate clearly and directly.
alerts → alerts → bulletin → alerts → search_places → rain_nowcast → bulletin → bulletin → search_places → bulletin
10
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_alerts", "arguments": {"domain": "33"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_rain_no...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_rain_nowcast → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conve...
22_baseline
Tests whether the agent correctly handles sequential weather inquiries for different administrative areas of a city, including geocoding, rain nowcasting, and current weather observation tools.
The scenario is plausible as a user interaction: the user is seeking detailed weather information for specific administrative areas of Paris, including real-time conditions and short-term rain forecasts, which is a realistic use case for someone planning an activity or project in the region.
null
The user is preparing for a project that requires precise weather information for different parts of Paris. They want to ensure that both current and short-term weather conditions are favorable for their activities in the 4th and 1st arrondissements.
Your name is Élodie Martin.
null
You communicate clearly and directly.
search_places → rain_nowcast → bulletin → search_places → rain_nowcast → observation → rain_nowcast → observation
8
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "75001"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.859, ...
[]
Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_rain_nowcast → meteo_observation → meteo_rain_nowcast → meteo_observation → Run_Conversation → End_of_Simulation
23_baseline
Tests whether the agent correctly uses location search and weather tools to provide alerts and real-time weather observations for user-specified regions.
The scenario represents a plausible user interaction where the user is checking weather alerts and conditions for multiple French departments and cities, seeking both general and location-specific weather information.
null
The user wants to stay informed about current weather conditions and alerts in different French departments where they may be traveling or have family. They provide department numbers and city names to request specific and accurate weather information.
Your name is Jean-Pierre Lefebvre.
null
You communicate clearly and directly.
alerts → bulletin → alerts → search_places → alerts → alerts → search_places → alerts → search_places → observation
10
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_alerts", "arguments": {"domain": "06"}}, {"name": "meteo_search_places", "arguments": {"q": "Nice"}}, {"name": "meteo_alerts", "arguments": {"domain": "06"}}, {"name": "meteo_alerts", "ar...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_search_places → meteo_observation → Run_Conve...
24_baseline
Tests whether the agent correctly chains weather bulletin checks, location searches (GPS and INSEE code), and weather alerts to answer a multi-step weather inquiry.
The scenario is plausible as a user interaction: the user is seeking weather-related information for specific administrative regions (departments and postal codes) and is following up with additional queries to refine their understanding of the weather situation and alerts.
null
The user is concerned about the weather in their area and wants to understand the current and upcoming weather warnings. They are located in department 75 and are interested in both general and more localized forecasts.
Your name is Amina Traoré.
null
You communicate clearly and directly.
bulletin → search_places → forecast → search_places → bulletin → alerts
6
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}, {"name": "meteo_search_places", "arguments": {"q": "75001"}}, {"name": "meteo_bulletin", "arguments": {"domain":...
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
25_baseline
Tests whether the agent correctly chains geocoding, weather bulletin, and weather alert tools to answer a user's inquiry about potential weather risks in a specific location.
The scenario is plausible as a user interaction: the user first seeks location-specific information (GPS and INSEE code of Paris), then asks for the current weather bulletin and active alerts, ending with a logical conclusion based on the information received.
null
The user wants to know if there are any weather alerts for Paris to prepare for an outdoor event. They are familiar with the need for weather preparedness and are seeking detailed and reliable information for planning.
Your name is Lucas Petit.
null
You communicate clearly and directly.
search_places → bulletin → alerts
3
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}]
[]
Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
26_baseline
Tests whether the agent correctly handles sequential weather inquiries for different departments and cities, including geocoding and rain nowcasting.
The scenario is plausible as a user interaction: the user systematically inquires about weather bulletins for multiple departments, then seeks weather information for a specific city (Toulouse) by requesting its GPS coordinates and a rain nowcast. The repeated calls to the same tool for different departments are justif...
null
The user is planning activities across several French departments and wants to stay informed about potential weather risks. They are familiar with department codes and expect the agent to provide timely and accurate weather-related updates.
Your name is Nathalie Rousseau.
null
You communicate clearly and directly.
bulletin → bulletin → bulletin → search_places → rain_nowcast → alerts → bulletin
7
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_bulletin", "arguments": {"domain": "06"}}, {"name": "meteo_search_places", "arguments": {"q": "Toulouse"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 43.604082, "lon": 1.43380...
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
27_baseline
Tests whether the agent correctly handles sequential weather alert checks across multiple departments and uses location search and nowcast tools to provide localized rain forecasts.
The scenario is plausible as a user interaction: the user is checking weather alerts for different departments and then inquires about the weather in Toulouse after obtaining its GPS coordinates. Each step logically follows the previous one and reflects a realistic information-seeking behavior.
null
The user is concerned about potential weather disruptions and wants to check the current weather alerts for their area and nearby departments. They are planning to travel to Toulouse and want to know if rain is expected in the near future.
Your name is Youssef El Amrani.
null
You communicate clearly and directly.
alerts → alerts → search_places → rain_nowcast
4
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_search_places", "arguments": {"q": "Toulouse"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 43.604082, "lon": 1.433805}}]
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → End_of_Simulation
28_baseline
Tests whether the agent correctly handles location-based weather inquiries by chaining geocoding, observation, and alert tools for multiple locations.
The scenario is plausible as a user interaction: the user is seeking weather-related information for two different cities (Paris and Lyon), using their GPS coordinates and INSEE codes to obtain current weather observations, vigilance bulletins, and active alerts. The sequence reflects a realistic and sequential informa...
null
The user wants to know the current and upcoming weather conditions for Paris and Lyon to prepare for weather disruptions. They need information about the weather observation for Paris (coordinates 48.859, 2.347), the bulletin and alerts for department 75 (Paris), and the bulletin and alerts for department 69 (Lyon).
Your name is Claire Dubois.
null
You communicate clearly and directly.
search_places → observation → bulletin → alerts → search_places → bulletin → alerts
7
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_observation", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Lyon"}}, {"nam...
[]
Run_Conversation → meteo_search_places → meteo_observation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
29_baseline
Tests whether the agent correctly chains location search and weather tools to address a user's sequential weather inquiries for different regions.
The scenario is plausible as a user interaction: the user is seeking weather-related information for multiple French departments and cities, and the agent provides relevant data after retrieving location details and weather forecasts.
null
The user is planning outdoor activities and wants to know the current weather alerts and conditions for several French cities and departments. They are looking for detailed and location-specific weather data to make informed decisions.
Your name is Thomas Girard.
null
You communicate clearly and directly.
alerts → search_places → forecast → bulletin → search_places → forecast → search_places → observation → forecast
9
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "Mar...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_search_places → meteo_observation → meteo_forecast → Run_Conversation → End_of_Simulation
30_baseline
Tests whether the agent correctly handles sequential weather alert checks and appropriately uses geocoding and forecast tools when no alerts are found.
The scenario is plausible as a user interaction: the user systematically checks weather alerts for multiple departments and, finding none, seeks additional weather-related information (GPS coordinates, forecast, and bulletin) to ensure they are fully informed about potential weather conditions.
null
The user wants to check if any active weather alerts are in place for the departments 75, 44, 33, and 13. They are seeking detailed weather information for Marseille and the department 06 to prepare for an upcoming event.
Your name is Fatou Diallo.
null
You communicate clearly and directly.
alerts → alerts → alerts → search_places → forecast → alerts → bulletin
7
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_alerts", "arguments": {"domain": "33"}}, {"name": "meteo_search_places", "arguments": {"q": "Marseille"}}, {"name": "meteo_forecast", "arguments": {"lat": 43.282, "lon": 5.405}}, {"name": "...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
31_baseline
Tests whether the agent correctly chains weather bulletin retrieval, location search, and weather alert tools to answer a multi-step weather inquiry.
The scenario is plausible as a user interaction: the user is seeking weather warning information for specific departments and then follows up with a request for location-specific data (GPS and INSEE code) to refine their inquiry. The progression of the conversation reflects a realistic pattern of information-seeking be...
null
Your name is Mathieu Leroy.
null
You communicate clearly and directly.
bulletin → bulletin → search_places → alerts
4
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_search_places", "arguments": {"q": "Lyon"}}, {"name": "meteo_alerts", "arguments": {"domain": "69", "depth": 1}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → End_of_Simulation
32_baseline
Tests whether the agent correctly handles sequential weather alerts and bulletins for multiple departments to address a user's ongoing situational weather inquiry.
The scenario is plausible as a user interaction: the user is checking weather alerts and bulletins for multiple departments, likely to assess conditions for travel or events, and the agent provides consistent and relevant responses based on the data retrieved.
null
The user is planning multiple trips and outdoor events in different French departments and wants to ensure there are no significant weather risks involved. They are familiar with the concept of weather alerts and use department codes to specify their queries.
Your name is Isabelle Moreau.
null
You communicate clearly and directly.
alerts → search_places → bulletin → bulletin → bulletin → alerts → search_places → alerts → alerts
9
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name": "meteo_bulletin", "arguments": {"domain": "31"}}, {"name": "meteo_alerts"...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simula...
33_baseline
Tests whether the agent correctly chains place search, rain nowcast, and weather alerts tools to address a detailed weather inquiry for a specific location.
The scenario is plausible as a user interaction: the user seeks weather-related information for a specific location, first confirming its coordinates, then checking for rain forecasts and active alerts, which is a coherent sequence of actions for someone planning in a rainy or alert-prone environment.
null
The user wants to know if it will rain in the coming hours and if there are any active weather alerts in their area, as they are planning to be outdoors. They are located in the Paris region and need accurate, up-to-date information to make a decision.
Your name is Samir Benali.
null
You communicate clearly and directly.
search_places → rain_nowcast → alerts
3
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}]
[]
Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
34_baseline
Tests whether the agent correctly chains geocoding and weather alert/bulletin tools to answer a location-specific weather inquiry.
The scenario represents a plausible user interaction where the user is seeking weather-related information for different locations (Paris and Toulouse) and departments (75 and 13), with a logical flow of queries and responses.
null
The user is planning outdoor activities and wants to check the weather conditions for multiple French cities and departments. They are particularly interested in any active weather alerts or bulletins to ensure safety and adjust their plans accordingly.
Your name is Marie-Claire Fontaine.
null
You communicate clearly and directly.
search_places → rain_nowcast → bulletin → alerts → bulletin → search_places → bulletin
7
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}, {"name...
[]
Run_Conversation → meteo_search_places → meteo_rain_nowcast → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → End_of_Simulation
35_baseline
Tests whether the agent correctly chains a weather bulletin lookup, geocoding, and current weather observation tools to address a user's inquiry about weather conditions and associated risks in a specific department.
The scenario is plausible as a user interaction: the user seeks weather warnings for a department, then asks for more details by locating a specific city and checking current weather conditions, which aligns with a realistic information-seeking pattern.
null
The user is concerned about potential weather-related disruptions in their area and wants to understand the current and upcoming weather conditions. They are located in department 75 and have a specific interest in the city of Paris for further details.
Your name is Antoine Dupont.
null
You communicate clearly and directly.
bulletin → search_places → observation
3
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_observation", "arguments": {"lat": 48.859, "lon": 2.347}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_observation → Run_Conversation → End_of_Simulation
36_baseline
Tests whether the agent correctly chains weather alert checks, rain nowcast, and weather bulletin retrieval to answer a multi-step weather inquiry.
The scenario is plausible as a user interaction: the user is checking weather alerts for multiple departments and, finding none, asks for a rain nowcast and a complete weather bulletin, which represents a realistic progression of a weather-related inquiry.
null
The user is checking weather alerts for several departments to ensure there are no risks for upcoming outdoor activities. They are particularly interested in Lyon and other regions to get a comprehensive view of current and potential weather conditions.
Your name is Layla Hamidi.
null
You communicate clearly and directly.
alerts → alerts → rain_nowcast → observation → alerts → alerts → bulletin
7
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 45.764, "lon": 4.8357}}, {"name": "meteo_observation", "arguments": {"lat": 45.758097, "lon": 4.8407}}, {"name": "meteo_alerts", "arguments": {"domain": ...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_rain_nowcast → meteo_observation → Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
37_baseline
Tests whether the agent correctly retrieves weather alerts and then provides a detailed weather bulletin for a different location.
The scenario is plausible as a user interaction: the user first inquires about weather alerts in one department, then follows up with a request for a detailed bulletin in another department, and finally concludes the conversation with a thank you.
null
The user is checking weather alerts for department 75 to confirm there are no disruptions to their plans. After finding no alerts, they decide to investigate the weather in department 13 where they have an upcoming event.
Your name is Pierre Legrand.
null
You communicate clearly and directly.
alerts → bulletin
2
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}]
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
38_baseline
Tests whether the agent correctly handles weather alerts, location-specific weather data retrieval, and provides consistent information across multiple tools.
The scenario is plausible as a user interaction: the user seeks weather alerts, then detailed weather information for a specific location and sub-area, which the agent provides through a logical sequence of tool calls.
null
The user is planning outdoor activities in department 75 and wants to know if there are any current weather alerts. They are also interested in checking the weather for a specific area of Paris and need detailed forecasts and observations to make informed decisions.
Your name is Aïssatou Coulibaly.
null
You communicate clearly and directly.
alerts → search_places → forecast → search_places → observation → bulletin
6
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}, {"name": "meteo_search_places", "arguments": {"q": "75001"}}, {"name": "meteo_observation", "arguments": {"lat": 4...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_search_places → meteo_observation → meteo_bulletin → Run_Conversation → End_of_Simulation
39_baseline
Tests whether the agent correctly handles sequential weather alerts and bulletins for different departments and cities, including geocoding and nowcasting tools.
The scenario is plausible as a user interaction: the user is checking weather alerts for multiple departments and a city, and the agent provides consistent and relevant information based on the tools used. The repeated calls to 'meteo_alerts' and 'meteo_bulletin' are justified by the user's curiosity and follow-up requ...
null
The user wants to check the current weather alerts for their department and others to prepare for potential weather-related disruptions. They are located in an area where they are familiar with department codes and are interested in both immediate weather conditions and longer-term forecasts.
Your name is Julien Morel.
null
You communicate clearly and directly.
alerts → alerts → search_places → observation → rain_nowcast → observation → bulletin → search_places → bulletin → alerts
10
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_search_places", "arguments": {"q": "Toulouse"}}, {"name": "meteo_observation", "arguments": {"lat": 43.604082, "lon": 1.433805}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 43.6047...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_observation → meteo_rain_nowcast → meteo_observation → meteo_bulletin → Run_Conversation → meteo_search_places → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
40_baseline
Tests whether the agent correctly handles a multi-level location refinement for weather information, including alerts, forecasts, observations, and nowcasts.
The scenario is plausible as a user interaction: the user progressively refines their location-based weather inquiry from a department to a city and then to a specific postal code, with the agent responding appropriately to each level of specificity.
null
The user is planning outdoor activities and wants to know if there are any weather alerts in department 75. They are located in France and have a specific interest in Paris and its neighborhoods.
Your name is Sophie Bernard.
null
You communicate clearly and directly.
alerts → search_places → forecast → search_places → observation → rain_nowcast
6
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_forecast", "arguments": {"lat": 48.859, "lon": 2.347, "lang": "fr"}}, {"name": "meteo_search_places", "arguments": {"q": "75001"}}, {"name": "meteo_observation", "arguments": {"lat": 4...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_search_places → meteo_observation → meteo_rain_nowcast → Run_Conversation → End_of_Simulation
41_baseline
Tests whether the agent correctly handles sequential weather inquiries across multiple departments and cities, including alerts, forecasts, nowcasts, and bulletins.
The scenario is plausible as a user interaction: the user is checking weather alerts and forecasts for multiple French departments and cities, which reflects a realistic information-seeking behavior for weather-related planning.
null
The user is trying to stay informed about current and upcoming weather conditions in various French regions to plan activities and travel. They have no prior knowledge of weather alerts and are using the agent to check for any potential issues in different locations.
Your name is Karim Mansouri.
null
You communicate clearly and directly.
alerts → alerts → search_places → alerts → search_places → forecast → bulletin → alerts → search_places → rain_nowcast → bulletin
11
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}, {"name": "meteo_search_places", "arguments": {"q": "Toulouse"}}, {"name": "meteo_alerts", "arguments": {"domain": "31", "depth": 1}}, {"name": "meteo_search_places", "arguments": {"q": "31000"}}, {"name": ...
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → meteo_search_places → meteo_forecast → Run_Conversation → meteo_bulletin → Run_Conversation → meteo_alerts → Run_Conversation → meteo_search_places → meteo_rain_nowcast → Run_Co...
42_baseline
Tests whether the agent correctly provides a rain nowcast, retrieves location identifiers, and checks for active weather alerts in a location-specific weather inquiry.
The scenario is plausible as a user interaction: the user first asks for a rain nowcast in Lyon, then requests GPS and INSEE codes to confirm the location, and finally asks for weather alerts in the department using the INSEE code. The sequence reflects a logical progression of a user verifying location-specific weathe...
null
The user is planning outdoor activities in Lyon and wants to know the likelihood of rain in the near future. They are also concerned about verifying the exact location using GPS and INSEE codes, and checking for any active weather alerts in the area.
Your name is Élodie Martin.
null
You communicate clearly and directly.
rain_nowcast → search_places → alerts
3
[{"name": "meteo_rain_nowcast", "arguments": {"lat": 45.764, "lon": 4.8357}}, {"name": "meteo_search_places", "arguments": {"q": "Lyon"}}, {"name": "meteo_alerts", "arguments": {"domain": "69", "depth": 1}}]
[]
Run_Conversation → meteo_rain_nowcast → Run_Conversation → meteo_search_places → meteo_alerts → Run_Conversation → End_of_Simulation
43_baseline
Tests whether the agent correctly retrieves and communicates weather alert status for different departments in response to a user's sequential inquiry.
The scenario is plausible as a user interaction: the user asks about weather alerts in two different departments and ends the conversation upon receiving consistent information.
null
The user is seeking information about current weather alerts in their local department to assess potential risks. They first inquire about department 75 and then about department 44.
Your name is Jean-Pierre Lefebvre.
null
You communicate clearly and directly.
alerts → alerts
2
[{"name": "meteo_alerts", "arguments": {"domain": "75"}}, {"name": "meteo_alerts", "arguments": {"domain": "44"}}]
[]
Run_Conversation → meteo_alerts → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
44_baseline
Tests whether the agent correctly retrieves and reports weather warning bulletins for different departments in response to a user's sequential inquiry.
The scenario is plausible as a user interaction: the user asks for weather bulletins for two different departments, which represents a realistic information need for someone monitoring weather risks in multiple regions.
null
The user is seeking weather warning information for department 75 to assess potential risks. They then request the same type of information for department 13.
Your name is Amina Traoré.
null
You communicate clearly and directly.
bulletin → bulletin
2
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_bulletin", "arguments": {"domain": "13"}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_bulletin → Run_Conversation → End_of_Simulation
45_baseline
Tests whether the agent correctly handles a sequential weather inquiry involving rain nowcasting and weather alerts for a specific region.
The scenario is plausible as a user interaction: the user first asks for a rain nowcast in Lyon, then inquires about weather alerts in the Paris region (department 75), and finally ends the conversation after receiving reassuring information.
null
The user wants to know the rain forecast for the next few hours in Lyon to plan their activities. They also want to check if there are any active weather alerts in the Paris region (department 75) to assess potential risks.
Your name is Lucas Petit.
null
You communicate clearly and directly.
rain_nowcast → alerts
2
[{"name": "meteo_rain_nowcast", "arguments": {"lat": 45.764, "lon": 4.8357}}, {"name": "meteo_alerts", "arguments": {"domain": "75"}}]
[]
Run_Conversation → meteo_rain_nowcast → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
46_baseline
Tests whether the agent correctly chains location search, current weather observation, and weather bulletin tools to address a user's weather inquiry for a specific department.
The scenario represents a plausible user interaction where a user seeks weather-related information for a specific location, using a logical sequence of tools to obtain relevant data.
null
The user wants to know the current weather conditions and any ongoing or upcoming weather alerts for Paris. They need information about the current weather and the bulletin for the department 75.
Your name is Nathalie Rousseau.
null
You communicate clearly and directly.
search_places → observation → bulletin
3
[{"name": "meteo_search_places", "arguments": {"q": "Paris"}}, {"name": "meteo_observation", "arguments": {"lat": 48.859, "lon": 2.347}}, {"name": "meteo_bulletin", "arguments": {"domain": "75"}}]
[]
Run_Conversation → meteo_search_places → meteo_observation → meteo_bulletin → Run_Conversation → End_of_Simulation
47_baseline
Tests whether the agent correctly chains weather warning and rain nowcast tools to address a user's sequential inquiry about potential risks and immediate weather conditions in a specific area.
The scenario is plausible as a user interaction: the user first inquires about a weather warning bulletin for a department, then follows up with a specific rain nowcast query for a location within that department, and finally concludes the conversation after receiving the information.
null
Your name is Youssef El Amrani.
null
You communicate clearly and directly.
bulletin → rain_nowcast
2
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_rain_nowcast", "arguments": {"lat": 48.8566, "lon": 2.3522}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_rain_nowcast → Run_Conversation → End_of_Simulation
48_baseline
Tests whether the agent correctly switches between weather observation and alert tools to address a user's sequential weather-related inquiry.
The scenario is plausible as a user interaction: the user first asks for current weather in Marseille, then follows up with a request about weather alerts in department 06, and finally concludes the conversation after receiving the information.
null
Your name is Claire Dubois.
null
You communicate clearly and directly.
observation → alerts
2
[{"name": "meteo_observation", "arguments": {"lat": 43.2965, "lon": 5.3698}}, {"name": "meteo_alerts", "arguments": {"domain": "06"}}]
[]
Run_Conversation → meteo_observation → Run_Conversation → meteo_alerts → Run_Conversation → End_of_Simulation
49_baseline
Tests whether the agent correctly handles a weather warning inquiry followed by a current weather observation request using location coordinates.
The scenario is plausible as a user interaction: the user first inquires about a weather warning bulletin for a specific department, then asks for current weather conditions for a specific location (Paris), and finally ends the conversation after receiving the information.
null
The user is seeking information about weather warnings in a specific department to prepare for potential conditions. They then ask for the current weather at a specific set of coordinates to understand the immediate situation in that location.
Your name is Thomas Girard.
null
You communicate clearly and directly.
bulletin → observation
2
[{"name": "meteo_bulletin", "arguments": {"domain": "75"}}, {"name": "meteo_observation", "arguments": {"lat": 48.8566, "lon": 2.3522}}]
[]
Run_Conversation → meteo_bulletin → Run_Conversation → meteo_observation → Run_Conversation → End_of_Simulation

delhIA — Evaluation Scenarios

This dataset contains 50 evaluation scenarios for delhIA, a Météo-France weather assistant powered by 7 live API tools. Each scenario is a structured task definition specifying a user persona, instructions, and the expected sequence of tool calls the assistant should perform.

The format extends TAU-bench with richer evaluation criteria and an explicit tool-path annotation derived from a Markov-weighted tool-transition graph.

Dataset at a Glance

Scenarios 50
Domain Météo-France weather assistant
Tools covered 6 / 7 (search_places, forecast, observation, rain_nowcast, alerts, bulletin)
Language French
Format JSON

Tools

The delhIA assistant exposes 7 Météo-France tools:

Tool Description
meteo_search_places Resolve a city name to GPS coordinates
meteo_forecast Multi-day weather forecast for a location
meteo_observation Real-time observation at the nearest station
meteo_rain_nowcast Short-term rain nowcast (Metropolitan France only)
meteo_alerts Vigilance alerts by department or nationwide
meteo_bulletin Official weather bulletin by department
meteo_warning_dictionary Lookup phenomenon definitions

Data Structure

{
  "domain": "meteo",
  "policy": "...",        // assistant system prompt
  "guidelines": "...",    // user simulator guidelines
  "tasks": [
    {
      "id": "0_baseline",
      "description": {
        "purpose": "...",
        "notes": "..."
      },
      "user_scenario": {
        "persona": null,
        "instructions": {
          "task_instructions": "...",
          "domain": "meteo",
          "reason_for_call": "...",
          "known_info": "Your name is Sophie Bernard.",
          "unknown_info": null
        }
      },
      "evaluation_criteria": {
        "actions": [
          {
            "action_id": "0_0",
            "name": "meteo_bulletin",
            "arguments": {"domain": "75"},
            "requestor": "assistant"
          }
        ],
        "communicate_info": [],
        "nl_assertions": []
      },
      "annotations": {
        "story_id": 0,
        "path": ["Run_Conversation", "meteo_bulletin", "..."]
      }
    }
  ]
}

Usage

import json

with open("data/tasks.json") as f:
    data = json.load(f)

policy    = data["policy"]       # assistant system prompt
guidelines = data["guidelines"]  # user simulator guidelines

for task in data["tasks"]:
    task_id        = task["id"]
    instructions   = task["user_scenario"]["instructions"]
    expected_tools = [a["name"] for a in task["evaluation_criteria"]["actions"]]
    print(task_id, expected_tools)

Scenario Generation

Scenarios were produced by a five-stage pipeline:

  1. Tool audit — static + dynamic profiling of each API tool (probing, failure modes, I/O labels)
  2. Tool graph — finite-state automaton over tool transitions, pruned by type compatibility
  3. Markov estimation — edge weights from empirical trajectory sampling
  4. Trajectory sampling — paths drawn from the weighted graph, filtered by an LLM judge
  5. Narration — each trajectory hydrated into a full user scenario (persona, instructions, real API calls, success conditions)

Companion Dataset

Actual conversation runs on these scenarios (3 models, 150 runs) are at Jourdain/delhIA-runs.

License

Creative Commons Attribution 4.0 International (CC BY 4.0)

Downloads last month
16