data_library
ChatGPTSearch and describe datasets in the CarbonArc data library. Use this tool for dataset discovery and provenance questions: - "What data do you have about card spend?" - "Where does this data come from?" - "What fields are available in the retail foot traffic dataset?" Routing: - Any ID provided (insight_id, topic_id, dataset_id) → describe mode: returns the full TearSheet for that dataset, including provider, frequency, history, delivery method, panel size, coverage, bias, lag, geographic availability, data dictionary, raw tearsheet_data metadata, and last refresh date. - Only query provided → search mode: semantic search over datasets by name/description. - No arguments → catalog overview: returns the full accessible TearSheet catalog (subject to source access restrictions). Search then describe: Call with a query first to find relevant datasets, then pass the tearsheet_name from any search result as dataset_id to retrieve its full TearSheet details. This tool vs search_insights: - Use data_library for dataset/provenance questions (what data exists, what fields, where does it come from, how is it collected). - Use search_insights for metric discovery (finding specific analytical insights/KPIs to use in framework queries).
framework_to_insight
ChatGPTConvert structured analytical frameworks into insights. This tool takes a pre-defined analytical framework (structured query parameters) and executes it to generate insights. Useful when you have specific query parameters rather than natural language questions. This provides more control than natural language queries and is ideal for programmatic access or when you know exactly what data structure you need.
get_entities_from_event
ChatGPTGet entities linked to a specific event. Retrieves entities associated with a particular parent event (e.g. companies for a corporate filing). Use search_events first to resolve an event name to its event_id and event_type.
get_entities_from_insight
ChatGPTGet entities related to a specific insight. Retrieves entities associated with a particular insight. Useful for exploring what entities are involved in a specific insight or understanding insight-entity relationships. Entities are listed by association, not ranked by any metric. Valid entity_representation values: abundle, actor, aingredient, app, apublisher, artist, bullbearargument, category, cbsa, city, clfiling, clopinion, comcategory, company, country, court, dccategory, ddcategory, director, distributor, dma, docket, dtcategory, educationlevel, entertainer, ethnicity, finished, franchise, gender, generation, genre, judge, label, lawyer, league, legalmatter, maritalstatus, medicalsegment, mine, newsarticle, occupation, pcodecategory, plabel, platform, port, product, quailinvoice, quailmatter, raw, region, retailer, secfiling10k, secfiling10q, service, socialhandle, song, sport, state, team, ticker, title, transcript, transformed, us, vehiclebody, vehiclemake, vehiclemodel, venue, website, ww, yearlyincome, zip
get_events_from_entity
ChatGPTGet events linked to a specific entity. Retrieves events associated with a particular entity through shared insights. Useful for discovering which events are connected to a given company or entity. Use search_entities first to resolve an entity name to its entity_id.
get_events_from_insight
ChatGPTGet events linked to a specific insight. Retrieves events associated with a particular insight node. Useful for discovering which events are connected to a given insight. Use search_insights first to resolve an insight name to its insight_id.
get_filter_options
ChatGPTDiscover available filters for a framework, or fetch selectable values for a specific filter. Metadata mode (filter_key omitted): Returns all available filters for the given entity/insight combination. Each item's `filter_key is the identifier to use in step 2 and in framework_to_insight filters. label is human-readable display text only. No SQL is executed — this is a fast, free discovery call. **Values mode** (filter_key provided): Returns distinct selectable values for the specified filter. Each option's value is what to pass in framework_to_insight filters or dependency_filters. label is human-readable display text only. **Typical workflow:** 1. Call without filter_key to learn which filters are available. 2. For each filter you want to populate, call again with filter_key set to the filter_key returned in step 1. 3. Pass the selected value(s) into framework_to_insight as filters: e.g. {"state": ["California", "Texas"]}`.
get_insights_from_entity
ChatGPTGet insights related to a specific entity. Retrieves insights associated with a particular entity (e.g., company, state, retailer). Useful for exploring what data is available for a specific entity or understanding entity-insight relationships. Valid entity_representation values: abundle, actor, aingredient, app, apublisher, artist, bullbearargument, category, cbsa, city, clfiling, clopinion, comcategory, company, country, court, dccategory, ddcategory, director, distributor, dma, docket, dtcategory, educationlevel, entertainer, ethnicity, finished, franchise, gender, generation, genre, judge, label, lawyer, league, legalmatter, maritalstatus, medicalsegment, mine, newsarticle, occupation, pcodecategory, plabel, platform, port, product, quailinvoice, quailmatter, raw, region, retailer, secfiling10k, secfiling10q, service, socialhandle, song, sport, state, team, ticker, title, transcript, transformed, us, vehiclebody, vehiclemake, vehiclemodel, venue, website, ww, yearlyincome, zip
get_insights_from_event
ChatGPTGet insights linked to a specific event. Retrieves insights associated with a particular event node. Useful for exploring what data is available for a specific event or understanding event-insight relationships. Use search_events first to resolve an event name to its event_id and event_type.
ontology
ChatGPTExplore Carbon Arc's ontology — the read-only internal knowledge graph linking companies, brands, categories, locations, apps, websites, subjects, topics, insights, and datasets. Ask natural-language questions about how entities relate, what data covers what, and overall graph structure. Returns relationship tables — not metrics, trends, or time-series values. Question shapes: - Anchored — about specific named things (e.g. "What brands does Walmart own?", "What topics does Target have?"). Names are resolved to graph IDs automatically. - Entity-free — global counts, listings, or schema (e.g. "How many companies do we cover?", "What relationship types are in the graph?"). No entity_ids needed. Input: - `question only (default) — routing infers which entities the question refers to. - entity_ids — skip name resolution on retry after checking metadata.entity_resolution_metadata` or calling a discovery tool below. Example questions: - Categories & locations: "What sub-categories does beverages have?", "What sector is Walmart in?", "What cities are in California?" - Ownership & brands: "What brands does Pepsi own?", "Who owns Dunkin?", "What private labels does Costco carry?" - Apps & websites: "What apps does Nike have?", "What websites does Walmart have?" - Subjects & topics: "What KPI topics does Target have?", "What topics do Walmart and Costco share?", "What subjects are covered for Walmart?" - Data coverage: "What tearsheets cover Walmart?", "What insights do you have for Walmart?" - Graph-wide: "What subjects are in the graph?", "How many companies do we cover?", "What relationship types are in the graph?" Not for (use another tool): - Business metrics or trends → text_to_insight or synthesize_answer - Analyzing named-event relationships with other graph nodes (entities, insights) — not supported yet. Use search_events / get_events_from_ / get_insights_from_event / get_entities_from_event. Graph hits about "metric events" or EventSubject topics are insight/topic coverage, not those event↔entity/insight relationship queries. - Dataset provenance or catalog → data_library (including data asset IDs like CA0056, BulkAPI, BlockData) - Platform documentation → search_docs - Finding entities by name → search_entities - Searching insights by keyword → search_insights Wrong-tool examples:* - "What was Walmart's revenue last quarter?" → text_to_insight - "Which companies are linked to Apple Q3 earnings?" → search_events / get_entities_from_event - "Where does foot traffic data come from?" → data_library - "Find companies named Acme" → search_entities
search_docs
ChatGPTSearch Carbon Arc platform documentation and return an answer. Use this tool to answer questions about how the Carbon Arc platform works, what data is available, how to interpret results, or how to use specific features. Returns a natural-language answer grounded in the documentation. Use Cases: - "What is Carbon Arc?" - "How do I interpret card spend data?" - "What entity representations are available?" - "How does the framework system work?"
search_entities
ChatGPTSearch for entities by name or description. Optionally filter by entity_representations (single string or list) to narrow results to specific entity types. See the entity_representations parameter for valid values. Valid entity_representation values: abundle, actor, aingredient, app, apublisher, artist, bullbearargument, category, cbsa, city, clfiling, clopinion, comcategory, company, country, court, dccategory, ddcategory, director, distributor, dma, docket, dtcategory, educationlevel, entertainer, ethnicity, finished, franchise, gender, generation, genre, judge, label, lawyer, league, legalmatter, maritalstatus, medicalsegment, mine, newsarticle, occupation, pcodecategory, plabel, platform, port, product, quailinvoice, quailmatter, raw, region, retailer, secfiling10k, secfiling10q, service, socialhandle, song, sport, state, team, ticker, title, transcript, transformed, us, vehiclebody, vehiclemake, vehiclemodel, venue, website, ww, yearlyincome, zip
search_events
ChatGPTSearch for events by name or description. The entry point for event discovery: resolves an event name to its event_id, event_type, and event_category_id. Feed those into the get_insights_from_event / get_entities_from_event tools or into build_framework to analyze the event. Optionally narrow by event_type.
search_insights
ChatGPTSearch for insights by name or description.
text_to_insight
ChatGPTConvert natural language questions into analytical results from Carbon Arc's own data platform. All answers are computed from data that Carbon Arc has already ingested and hosts internally. This tool does not browse the web, call third-party APIs, or fetch documents at query time. Returns markdown tables for quantitative insight metrics (card spend, foot traffic, rankings, trends) from Carbon Arc's licensed data panels. Returns narrative analysis with source citations from Carbon Arc's internal document archive, a pre-ingested, Carbon Arc-hosted collection of SEC filings (10-K/10-Q), earnings call transcripts, company and market news articles, and bull/bear investment theses. Coverage is limited to what has been ingested into that archive. Query shapes supported (structured analytics): - timeseries — metric values grouped by date (default); e.g., "Show Walmart card spend over 2024" - summary — single aggregate per entity with no date grouping; e.g., "Total card spend by company in Q1 2024" - list — distinct entity or dimension values matching filters; e.g., "What platforms had advertising spend for Walmart in 2024?" - ranking — top-N leaderboard across entities; e.g., "Top 10 companies by card spend last year", "Which brands had the highest revenue in 2024?" - ranking_timeseries — top-N entity discovery followed by a full timeseries for each; e.g., "Show card spend trends for the top 5 retailers" Common Use Cases: - Revenue and spend analysis by company, region, or time period - Growth metrics and trend analysis - Comparative analysis across entities and insights - Management commentary, risk factors, filing narrative, news in Carbon Arc's archive, and bull/bear investment-case questions - Top-N leaderboards and ranking across entities (ranking / ranking_timeseries shapes) - Aggregate summaries without time grouping (summary shape) - Distinct-value lookups for entities or dimensions (list shape) Not Suitable For: - Schema/metadata discovery: "What data is available for Walmart?", "What fields can I filter on?" - Vague or incomplete questions with no metric/insight and no document angle (e.g., "walmart 2025", stock predictions, unknowable private facts) - Questions that don't reference a specific data metric, statistic, quantity, or document topic (filings, earnings calls, news in Carbon Arc's archive, investment theses) - Named events (earnings, Fed decisions, conferences, etc.): this tool does not resolve or query events. Discover with search_events (or get_events_from_entity / get_events_from_insight), then execute via framework_to_insight with events (event_type plus event_id and/or event_category_id). This tool requires a question that references a specific data metric or insight (e.g., card spend, box office revenue, website traffic), or a question about SEC filings, earnings calls, news in Carbon Arc's archive, or bull/bear investment theses. Questions with only an entity name but no metric may return an error with recovery suggestions. If the user's question lacks a clear metric or document angle, ask them what specific data they want before calling this tool.
data_library
Claudeframework_to_insight
Claudeget_entities_from_insight
Claudeget_filter_options
Claudeget_insights_from_entity
Claudeontology
Claudesearch_docs
Claudesearch_entities
Claudesearch_insights
Claudetext_to_insight
Claude
