FactIQ

List the data sources FactIQ has, plus the shared table structure. Call this once before exploring. Returns the visible schemas (organization + country), the ones with no data loaded yet (`schemas_without_data` — skip them), a per-schema dataset index, and the shared DDL for the `series`, `data_points`, `dimensions` and `compound_series` tables. schemas: optional comma-separated filter (e.g. "bls,bea"). full: heavy per-dataset descriptions instead of the compact index.

get_data_catalog

Full metadata for one dataset: topic, methodology, discontinuities, any base-change notice, plus the live dimensions and example series. Call after `search_datasets` narrows you to a dataset.

describe_dataset

Recent market data the warehouse lags on: quotes, OHLCV, company profiles, FX, and commodities. Set data_type to price_history for stocks, currency pairs, and commodities; the asset format determines which market is queried. Use search_company_filings for financial statements and reported facts.

get_market_data

Get satellite-derived indicators aggregated over a country, one state/province, or a bounding box. Use for fires, emissions-based activity, rainfall, vegetation, temperature, and soil-moisture questions. For a sub-national area, call one state/province at a time. Informal regions such as "north India" or "the Midwest" have no boundary, so cover them with one call per state or a bounding box; using the whole country answers a different question. Grid cells and fire detections are activity snapshots, never fire perimeters. For ndvi_s2 and *_tropomi grids, drop or flag cells with low valid_obs_share. Results include units, source attribution, and caveats; cite the attribution. External-provider calls can take about 30 seconds, and rainfall or grid calls can take longer.

get_geo_data

Fetch one series' data: timeseries, tabular, or COMPOUND:: all work. Returns title, units, frequency, notes and data points, with source metadata. `transform="yoy_pct"` (percent change) or `"yoy_diff"` (difference, for rates) adds a column with the change versus the same period one year earlier, matched by calendar date — use it for inflation or growth from one series instead of a `LAG(value, 12)` query, which goes wrong after any missing month. The cell is null where the prior-year period is absent, and `transform.periods_without_prior` lists those periods. If the result carries a `coverage_note`, the series has periods with no observation (listed in `missing_periods`): say so in the answer, leave them blank rather than interpolating, and label any aggregate that spans them as partial. SEC-backed results include `row_sources` keyed by `result_index`; each entry identifies the supporting filing, accession/form/publication date, whether the observation is reported or derived, and its standardized `source_link`. Cite that row's source URL beside any figure and label derived observations as derived. A document-precision link does not open an exact passage. At most 50 points come back; if the series is longer the result is flagged `"truncated": true` — narrow it with `from_year`/`to_year`, or pull the series through `run_sql` with an aggregation (e.g. monthly/quarterly via `date_trunc`) rather than trying to fetch every point.

get_series

Render a FactIQ-styled interactive chart inline, from data you have already fetched. Before calling, finish all data retrieval, calculations, source reconciliation, and title/value checks. This tool does not publish or host a public page or create a share link. FactIQ-hosted publishing is unavailable. Explain that limitation and offer the inline chart or a local artifact; do not inspect FactIQ's legacy web interface, use browser automation, or probe HTTP endpoints for a publishing workaround. Do not call send_feedback for this intentional capability boundary. For each intended chart, call this tool exactly once, only after its data and title are final; never render a draft and then re-render a correction. Build the final numbers from run_sql / get_series, aggregated to chart resolution. - x_values: the x axis — ISO dates ("2024-01-01") or categories, sorted. These are the positions; keep them as real ISO dates for date axes so the axis spaces correctly. - series: one or more {label, values}; each `values` list lines up 1:1 with x_values (use null for gaps, don't drop entries). - chart_type: "line" (default), "bar", or "area". - units: y-axis units, e.g. "Percent". Axis-label control (both optional — leave unset for normal date axes, which auto-format to the most compact form, e.g. '24 or Jul 24): - x_tick_format: hint the date granularity — "auto" (default), "year", "quarter", "month", or "date". Use e.g. "quarter" to label quarterly points "Q1 '24", or "date" to force full dates. The widget still chooses how many labels fit and spaces them; this only sets the text. - x_labels: fully custom tick text, one per x_value (1:1 with x_values), for labels the widget can't infer — e.g. fiscal quarters ["Q2 FY19", …, "Q2 FY26"] or custom category names. Takes precedence over x_tick_format. x_values still drives positioning; x_labels is only the text. Don't use this just to reformat dates — prefer x_tick_format. Give the chart a finding-style title with the numbers in it (see the chart style guide). Keep it to a handful of series and a few hundred points at most — aggregate in SQL first if needed.

render_chart

Run a read-only SELECT against the FactIQ data warehouse, scoped to one schema — the power tool for joins, pivots and aggregation. Read-only role, 30s statement timeout, with FactIQ's query transforms and series-metadata enrichment. Resolve ids from `series` before filtering `data_points`. `auto_retry=true` opts into a server-side reviser on zero rows. See the `factiq://guide/sql` resource. Returns at most 50 rows. If the query matches more, the result is flagged `"truncated": true` — that is your cue to **aggregate or compute in SQL** (`GROUP BY`, `date_trunc`, a SUM/AVG/rank/ratio) and return that, NOT to try to fetch the raw rows. A chart or a summary needs the aggregated result, and large raw results don't fit in context anyway. Year-over-year and other period comparisons must join on the calendar period (`date_trunc('month', prior.time) = date_trunc('month', cur.time) - interval '1 year'` — the stored day of the month varies by source, so never compare exact dates), never on a row offset such as `LAG(value, 12) OVER (ORDER BY time)`: a series with one missing month compares the wrong periods for the twelve rows after the gap (LEAD: the twelve before it) and the numbers still look plausible. For one series prefer `get_series` with `transform`. When the result carries a `coverage_note`, the rows have periods with no observation (listed in `missing_periods`, per series): name them in the answer, do not interpolate, and label any aggregate that spans them as partial. `page` is the one exception and works only on the `nasa_fires` schema, where a row is a single fire detection that belongs to no series and so cannot be aggregated into one. Each page holds 50 rows; the answer carries `page` and `has_more`, and you read the next block with `page=2`, `page=3` and so on. Give the query an ORDER BY, otherwise Postgres may return the same row on two pages and never return another. Still prefer a count or a total when that answers the question — paging through thousands of detections does not. Runs through FactIQ's read-only database role, which rejects any attempt to modify data (INSERT/UPDATE/DELETE/DROP/…).

run_sql

Find datasets by keyword across every schema — the first discovery step. Keyword ranking (not semantic) over dataset titles and topics. Returns ranked {schema, dataset_code, title, topic}. Follow up with `describe_dataset` for detail, or `run_sql` on the series/dimensions tables for exact series.

search_datasets

Search earnings-call intelligence from live calls, decomposed into atomic, quote-anchored claims and Q&A pressure points — never a raw transcript dump. Retrieval for "claims" and "pressure_points" is lexical, not semantic. A nonempty query uses an OR-of-tokens full-text candidate set; strict websearch_to_tsquery matches receive a fixed ranking boost, so strict matches rank ahead of loose partial-term matches. Lower-ranked rows may match only some query terms. If that filtered full-text search returns no rows, a trigram fallback handles typos and sub-word variants. Quoted phrases use websearch syntax, but synonyms are not inferred: try the company's own vocabulary ("capex" and "capital expenditure", segment names, abbreviations). An empty or partial result is not proof that management did not discuss a topic; coverage and vocabulary both matter. Target and filter semantics: - "claims" (default) searches or browses normalized management/analyst claim rows. ticker, quarter_filter, claim_family, section, detail, and limit apply. claim_family matches primary or secondary families; section is "prepared_remarks" or "qa". Base rows include direction/value/unit where checkable, assertion_status, canonical_statement, verbatim_quote, speaker, speaker_role, section, transcript_id, source_block_index, qa_turn_id, and a source_link object. - "pressure_points" searches or browses what analysts pressed for and whether management confirmed, declined, deflected, or partially answered. ticker, quarter_filter, claim_family (the linked family), detail, and limit apply. section is ignored because every pressure point is from Q&A. Base rows include specific_ask, response_quality, refused_number, linked_family, linked_claim_id, verbatim_quote, transcript_id, source_block_index, qa_turn_id, and a source_link object. - "disclosure_profile" is a direct company lookup, not text search. The ticker is the first ticker value, or query when ticker is empty. Profiles are curated separately from transcript coverage, so a covered company may not have one; availability is "available" or "unavailable" accordingly. quarter_filter, claim_family, section, detail, and limit do not narrow it. - "coverage" is corpus inventory, not text search. ticker and limit apply; query, quarter_filter, claim_family, section, and detail do not narrow it. Rows report calls_covered, earliest_period, latest_period, latest_call_date, claims, and pressure_points. An empty query is a capped browse, not a completeness guarantee. Rows are ordered by newest fiscal period/call and then source-block (spoken) order within one call. Every response is limited to at most 50 rows, and a browse that reaches its limit may omit additional rows even if no truncation flag is present. For a latest-call note: (1) call "coverage" with one ticker, (2) read latest_period, (3) call "claims" with that ticker, quarter_filter=latest_period, query="", detail=true, and a deliberate limit, then (4) call "pressure_points" with the same ticker and quarter. Treat both browses as capped evidence, not a complete transcript. Narrow large result sets with ticker, quarter, target, claim_family, and (for claims) section. For media quotes, run_sql can read `transcripts.appearance_claims.verbatim_quote`; other transcript tables remain restricted. Evidence rules: canonical_statement is normalized and must not be quoted as verbatim. Quote only verbatim_quote, with its speaker, role, ticker, and fiscal period. A […] inside verbatim_quote marks omitted transcript sentences between non-adjacent evidence spans: preserve it exactly when quoting, and never present text on either side of it as one continuous statement. When a quote is used, render source_link.source_label as the Markdown link text for source_link.source_url beside the claim. The label names the company, fiscal period, and earnings call transcript; never replace it with an ingestion vendor. Do not make a second call just to find a citation, do not invent a replacement when source_url is null, and do not call a document-level link exact context. source_locator still identifies the stored transcript block or Q&A turn. assertion_status is load-bearing: analyst_hypothesized is the analyst's framing, while mgmt_declined_to_confirm is a refusal, not a management assertion. Spoken call claims are a different source class from filed XBRL actuals (`run_sql` on the `sec` schema) and formally issued targets (the `sec_guidance` dataset); label them separately. Examples: - Latest call: search_earnings_transcripts(query="", search_target="coverage", ticker="MU"), then repeat with search_target="claims", quarter_filter=<latest_period>, detail=true, limit=50, and finally with search_target="pressure_points" for the same ticker and quarter. - Cross-company theme: search_earnings_transcripts(query="capital expenditure capacity", search_target="claims", ticker="MU,NVDA", detail=true); inspect each returned row for topical support, do not assume every term matched, and retry vocabulary as needed. - Disclosure habits: search_earnings_transcripts(query="", search_target="disclosure_profile", ticker="MU"). claim_family codes: driver_attribution, forward_conviction, prior_view_revision, strategic_agreement, pricing_mechanics, demand_magnitude, capacity_milestone, competitive_positioning, cohort_behavior, cost_margin_bridge, capital_allocation, stated_risk_constraint, qa_dynamics, regulatory_policy, pipeline_backlog, commodity_hedging, ma_deal, labor_org. Topic paths (topic_path column) are rooted at: demand, supply_capacity, pricing, cost_margin, capital, competition, customer, product_roadmap, risk, governance.

search_earnings_transcripts

Deterministically search precomputed intelligence from podcasts, TV interviews, and conference talks outside earnings calls, and from official documents: Federal Reserve speeches and congressional testimony, FOMC statements, implementation notes, minutes, press-conference transcripts and Summary of Economic Projections tables. The speakers are company executives, investors, fund managers, analysts, economists, journalists, central-bank officials, and other guests, and their claims cover listed and unlisted companies (OpenAI, DeepSeek, MiniMax), institutions (the Federal Reserve, the European Central Bank, regulators), and whole markets and industries, not only the speaker's own employer. No serving-time model interprets or expands the query. Retrieval for structured findings is lexical, not semantic: strict full-text matching is tried first, loose any-term full-text matching only when strict matching returns no candidates, then trigram matching only when both full-text stages are empty. Natural-language questions work because person names and common question scaffolding are normalized. Prefer concise topical terms; retry synonyms or the speaker's own vocabulary before concluding there is no coverage. Canonical search_target values: - "search" (default): blended structured claims and broad passage cards. - "claims": normalized decision-relevant claims only. - "passages": broad substantive paraphrase cards only. - "pressure_points": stored refusal and declined-to-confirm claim rows. - "appearances": video-level catalog rows. - "coverage": company-level structured-corpus inventory. Compatibility aliases remain accepted for older clients: "all" maps to "search", "videos" to "appearances", and "companies" to "coverage". New callers should use the six canonical values; "search" is the primary blended target, not a legacy alias. Result contracts: - search/claims/passages/pressure_points return result_kind, canonical_paraphrase, speaker, primary ticker, topic labels, video/channel/publication metadata, lexical relevance, and source_url (a timestamped YouTube deep link, the episode page for a podcast, the document page for an official document, or the PDF page of a press-conference transcript). detail=true adds normalized claim and attribution fields plus institution, document_kind and event_date (null for interviews), but never raw transcript text or evidence spans. - appearances returns video metadata, primary ticker, attribution metadata, matching claim count, URL, and relevance score. detail has no effect. - coverage returns company-level appearance, publication-date-span, channel, claim, and low-confidence-attribution counts. It is structured coverage, not a per-person acquired-video/hour inventory; query, sort, and detail have no effect. Filter and ordering semantics: - company is comma-separated tickers and company names, mixed freely. A stored ticker matches the appearance's company or a section subject with that ticker; any other value is read as the name of a company, institution, or organisation a podcast section is about, matched after dropping punctuation, a leading "The", and corporate suffixes. Every ticker attached to a matched name joins the ticker filter, so "Alphabet" also reaches GOOGL rows. The result records company_tickers, company_names, company_matched, and company_unmatched (a value that matched nothing, with up to five possible matches); an unmatched value returns no rows, not an error. company_filter is the old name of company and behaves the same; passing both is an error. The appearances and coverage targets also match every value, upper-cased, against the entity references of claims (products, rivals, partners such as "CUDA"); a value found only there is listed under company_references. Alternate spellings can return different appearances or coverage rows because reference names are matched exactly ("open ai" and "OpenAI" select the same subject but not the same references); the claims rows are the same. - person is a case-insensitive substring over the finding speaker and/or appearance speaker metadata. appearance_type applies to every target: podcast, tv_interview, conference and other for interviews; speech, testimony, minutes, statement, press_conference and hearing for official documents. institution (substring, "Federal Reserve") and country ("US") restrict every target to official documents. - claim_family restricts structured claims. With blended search it suppresses passage cards; pairing it with search_target="passages" is an error. For appearances/coverage it requires videos with matching structured claims. - date_from/date_to are inclusive YYYY-MM-DD bounds on the video's stored publication/upload date, which can differ from the recording or event date. - show is a case-insensitive substring of the show or channel name ("Odd Lots", "Les experts") and applies to every target. - relevance orders lexical score first, then publication date; newest orders publication date first, then lexical score. Coverage has fixed inventory ordering by latest covered publication date. - limit is one page of 1-50 rows and offset skips that many ranked rows before the page, so limit=15 with offset=15 returns rows 16-30 of the same ranking. Every result carries has_more, plus next_offset when more rows follow; call again with that offset to read the next page. With an empty query, search and claims browse recent high-signal claims without generic passage cards; passages browses recent passage cards; pressure_points browses recent refusal rows; appearances browses the video catalog; coverage returns the company inventory. Each result set is capped at 1-50 rows. Evidence safety: canonical_paraphrase is a normalized paraphrase, not a transcript quotation. Cite source_url (timestamped for YouTube, page-anchored for PDF transcripts) with the returned speaker, publication date, and video or document metadata. FOMC minutes describe the meeting of event_date and are published three weeks later; the Committee, not one person, is their speaker. Do not put canonical_paraphrase in quotation marks unless you independently verify the wording at the source link. The tool never returns raw transcript text. Detail-field safety: detail adds normalized claim fields only. The extraction pipeline's internal audit flags are not published, because they describe how a row was built rather than anything checkable about the claim; judge a claim from value, assertion_status, and the timestamped source link instead. company_attribution_confidence and company_attribution_source concern only the video-to-company/ticker assignment; inferred attribution is not independent confirmation.

search_media_appearances