Data Preview: Sample Rows per Table
What does a sample from an alternative data provider show a hedge fund?
Direct Answer: The Gemral Edge data preview lets a research team read up to 20 real rows from each of 24 tables before asking for anything. The rows belong to a small sample universe (AMZN, LMT), carry the ticker that was valid on the day each record became public, and every column is described from the data dictionary. The preview was refreshed on 2026-10-26.
Which government and disclosure records tables can you preview, and how far back do they go?
This group has 8 tables. Each table below shows what one row means, how many rows the sample universe holds and which years those rows cover.
| Table | What it holds | Rows in the sample universe | Years | Columns |
|---|---|---|---|---|
| edge_bid_protests | U.S. federal bid protest dockets, one row per docket number (B-NNNNNN.N), with the protested solicitation linked to the contract award and its listed parent. One row is one row per bid protest docket file number. | 6 | 2019–2024 | 28 columns: file_number, b_number, protester, solicitation_number, agency, sub_agency, case_type, status and 20 more |
| edge_congress_trades | Securities transactions disclosed by members of U.S. Congress (periodic transaction reports), amounts as disclosed ranges. One row is one row per disclosed transaction. | 607 | 2014–2026 | 17 columns: id, politician_id, ticker, asset_description, side, amount_min, amount_max, tx_date and 9 more |
| edge_contract_announcements | Daily U.S. federal defense contract announcements since 2015, one row per announced contract paragraph, resolved to the listed parent as of the announcement date. Includes Foreign Military Sales customer countries and raw weapon-system designators mentioned in the announcement. One row is one row per announced contract paragraph. | 2,985 | 2015–2026 | 23 columns: article_id, para_idx, announced_date, piid, contractor_text, recipient_name, recipient_uei, federal_award_id and 15 more |
| edge_contract_transactions | Transaction-level (modification-level) federal contract actions, each with its own action date and new-money amount, resolved to the entity as of the action date. Enables exact new-money-by-date reconstruction. One row is one row per award transaction. | 19,145 | 2017–2026 | 41 columns: transaction_key, award_id, piid, modification_number, transaction_number, parent_award_piid, action_date, fiscal_year and 33 more |
| edge_contracts | Federal contract awards (one row per award) to public companies and their subsidiaries, resolved to a ticker/entity. Amount fields distinguish lifetime obligation from recent new money. One row is one row per award. | 28 | 2026 | 24 columns: award_id, piid, recipient_name, parent_uei, entity_id, ticker, amount, awarding_agency and 16 more |
| edge_insider_trades | Corporate insider transactions (insider-ownership reports filed with the U.S. securities regulator) with routine-vs-opportunistic classification and reporter identity links. One row is one row per reported transaction line. | 40 | 2025–2026 | 35 columns: accession, issuer_cik, ticker, entity_id, reporter_name, reporter_role, tx_date, filed_at and 27 more |
| edge_lobbying_filings | Federal lobbying disclosure filings (quarterly), resolved to a listed entity where possible, with issue codes and lobbied government bodies. One row is one row per filing. | 588 | 2018–2026 | 23 columns: filing_uuid, filing_year, filing_period, filing_type, client_name, client_id, client_state, registrant_name and 15 more |
| edge_resolved_events | Flagship dataset: every record from the point-in-time source tables resolved to event - person - listed ticker as of the publication date, with the relationship path walked and the catalyst phase. Rows without a listed company are kept so coverage is measurable. One row is one row per source record and directly named entity. | 23,453 | 2014–2026 | 21 columns: resolved_event_id, source_table, source_record_key, entity_slot, event_family, phase, source_published_at, first_seen_at and 13 more |
A sample answers the questions a brochure cannot: whether the columns match your model, whether dates are the dates a record became public, and whether the history is deep enough for a backtest. You can filter the rows by ticker, sort any column and read what each column means without creating an account.
How do you judge alternative data providers from a sample?
Start with the date columns. A usable point-in-time table keeps the date of the underlying activity apart from the date it became public, and a research join should use the publication date so that no future information leaks into a backtest. Next read the grain, the statement of what one row means, because a table that mixes awards with later modifications produces double counts in a naive sum. Then check identifiers: public identifiers such as company filing numbers, contract award numbers and trade codes let your own systems join the rows without a mapping table. Finally compare history depth by dataset, since each public source began publishing in machine-readable form at a different time. The data coverage page shows the same depth for the full database, year by year.
What do alternative data providers for hedge funds offer beyond the sample?
The preview is a window, not the delivery. The full sample pack contains Parquet files for the whole sample universe with a manifest of row counts and checksums, the data dictionary, a methodology note, the license and the Data Due Diligence Questionnaire. Funds that need the entire point-in-time history license it under the Institutional plan, and the REST API, signed webhooks and the MCP server give programmatic access under the API plan. To start, email support@gemral.com with your fund name, the table you previewed and the questions you want the data to answer.
Which AI-infrastructure and physical-world records tables can you preview, and how far back do they go?
This group has 11 tables. Each table below shows what one row means, how many rows the sample universe holds and which years those rows cover.
| Table | What it holds | Rows in the sample universe | Years | Columns |
|---|---|---|---|---|
| edge_capex_quarterly | Quarterly capital expenditure (cash paid for property, plant and equipment) of the listed AI-infrastructure companies, single-quarter figures derived from the cumulative amounts in their quarterly and annual reports. A restated figure is a new row with a later filing date. One row is one row per ticker, quarter end and filing date. | 72 | 2009–2026 | 7 columns: ticker, period_end, capex_usd, filed_at, form, concept, accession |
| edge_grid_zone_load_daily | Daily electricity demand per grid sub-region, or per whole balancing area where the balancing area has no sub-regions (local calendar day), from hourly federal demand statistics since 2019. One row is one row per grid area (sub-region or balancing area) and local day. | 285,278 | 2019–2026 | 11 columns: grid_zone, grid_zone_parent, granularity, day, mwh_total, hours_present, is_complete, source_published_at and 3 more |
| edge_land_dc_buildings | Permitted data-center buildings from U.S. county building records: every original field (permit number, issue date, floor area, build status, parcel number, address), one row per version. One row is one row per building record version. | 239 | 2026 | 25 columns: version_id, record_key, record_hash, objectid, am_unit_id, am_mcpi, permit_num, am_id and 17 more |
| edge_land_dc_links | Vendor-derived links from the county land and building records to physical sites (distance) and to listed tickers (recorded owner entity matched by documented subsidiary or brand rules). One row is one row per record, link kind and linked site or rule. | 835 | 2026 | 13 columns: link_id, layer, record_key, link_kind, site_id, match_method, match_dist_m, ticker and 5 more |
| edge_land_dc_parcels | Data-center land parcels from U.S. county land-use records: every original field (recorded owner entity, project name, permit numbers, floor area, zoning, acreage) and the parcel polygon, one row per version. One row is one row per parcel record version. | 139 | 2026 | 44 columns: version_id, record_key, record_hash, objectid_1, data_center_status, pa_gis_acre, pa_legal_acre, existing_acres and 36 more |
| edge_land_dc_pipeline_areas | Planned data-center project areas from U.S. county planning records: every original field (application number, status, zoning case, proposed floor area, acreage) and the area polygon, one row per version. One row is one row per project-area record version. | 85 | 2026 | 20 columns: version_id, record_key, record_hash, objectid, application, first_status, first_subdivision, first_zoning_case_num and 12 more |
| edge_physical_site_events | Dated events at physical sites: data-center air permits (new / modified), power-plant pipeline status (planned, under construction, operating, retired) and reported annual NOx emissions. One row is one row per site, event kind, registration number and date. | 58 | 2011–2026 | 16 columns: event_id, site_id, entity_id, event_kind, registration_no, version_no, issued_on, value_num and 8 more |
| edge_physical_sites | Physical sites relevant to AI infrastructure: data centers (air-permit and environmental registries), power plants (federal generator inventory) and off-grid mining sites, with location, grid zone, nearest high-voltage substation and the listed company linked to the site. Bundles carry primary sites only (duplicate registry records of the same facility are folded into the primary site) and only the columns not marked exported=false. One row is one row per site. | 87 | 2026 | 29 columns: site_id, source, source_site_key, site_name, site_kind, operator_name_raw, state, county and 21 more |
| edge_site_satellite_monthly | Satellite metrics per site and period: built-up index, bright-roof share (500 m and 1,000 m discs) and land-surface temperature delta vs. a vegetated ring around the site. Quarterly through 2024, monthly from 2025. Compare the same period year over year: the built-up index has a seasonal swing larger than the construction signal and also rises when land is cleared; in arid regions use the bright-roof share, which separates new roofs from dry bare ground. One row is one row per site, period and metric. | 9,298 | 2017–2026 | 10 columns: site_id, month, metric, period_kind, value, scenes_used, source_published_at, first_seen_at and 2 more |
| edge_trade_hs_country_monthly | Monthly U.S. import value by 6-digit HS code and country of origin for the AI-infrastructure equipment watchlist (generators, transformers, switchgear, cooling, servers, processors, memory, networking, optical fibre). cty_code '-' is the all-countries total; every other row is a single country with a positive value (country groups are excluded). Sector context, not a company's own purchases. One row is one row per HS6 code, country (or the all-countries total) and month. | 73,165 | 2018–2026 | 7 columns: hs6, cty_code, cty_name, month, gen_val_usd, source_published_at, first_seen_at |
| edge_trade_hs_district_monthly | Monthly U.S. import value by HS code and customs district for a watchlist of AI-infrastructure equipment codes (transformers, generators, switchgear, cooling, servers and parts). National totals; importer names are not available. One row is one row per HS code, customs district and month. | 3,864 | 2025–2026 | 8 columns: hs_code, district_code, month, value_usd, source_published_at, first_seen_at, is_backfill, record_hash |
Which reference and identity tables tables can you preview, and how far back do they go?
This group has 5 tables. Each table below shows what one row means, how many rows the sample universe holds and which years those rows cover.
| Table | What it holds | Rows in the sample universe | Years | Columns |
|---|---|---|---|---|
| edge_entities | Entity master: companies and organisations with ticker, CIK, industry classification and listing lifecycle (listing/delisting and successor). One row is one row per entity. | 25 | 2026 | 19 columns: id, ticker, cik, name, parent_id, is_public, org_type, source_layer and 11 more |
| edge_entity_parent_links | Point-in-time parent/subsidiary links between entities, each with a validity window. One row is one row per link and validity window. | 23 | reference table | 8 columns: id, child_entity_id, parent_entity_id, valid_from, valid_to, source, source_ref, created_at |
| edge_entity_relationships | Typed relationships between entities, persons and other nodes (e.g. employment, lobbying client, subsidiary), each with a validity window and confidence. One row is one row per relationship and validity window. | 30 | 2025–2026 | 13 columns: id, src_type, src_id, rel_type, dst_type, dst_id, valid_from, valid_to and 5 more |
| edge_entity_relationships_all | Union view of materialised relationships and relationships derived from parent links, with a flag for unknown start dates. One row is one row per relationship. | 53 | 2025–2026 | 13 columns: rel_id, src_type, src_id, rel_type, dst_type, dst_id, valid_from, valid_to and 5 more |
| edge_persons | Person master: insiders, members of Congress and lobbyists resolved to one identity across sources. One row is one row per person. | 221 | reference table | 7 columns: id, full_name, name_norm, cik, legislator_id, lobbyist_id, created_at |
How are the preview rows chosen and checked?
The preview uses the same export path as the sample pack. For each table it keeps up to 20 of the newest rows and rotates across the sample tickers so that the ticker filter always has something to show. Table and column names are the names a fund receives, and text fields that could name an issuing agency are removed from the public preview even though the sample pack keeps them. A record type that fails the accuracy gate is left out of the export, so a table missing from this page is a table that has not yet passed.
What does the preview not tell you?
20 rows cannot show how a table behaves across a decade, so the row counts and the year range in each table row above describe the sample universe, not the full database. Public records are described by category rather than by name, and the Data Due Diligence Questionnaire at https://www.gemral.com/edge-ddq.md lists known gaps, refresh cadence and the rules that keep every record point in time. The data contains no material non-public information and is research data, not investment advice.
What related alternative intelligence streams are available?
Cross-reference correlated public records across our alternative intelligence directories, statutory trackers, and quantitative research libraries:
- Data Coverage & Completeness
- Verification Methodology
- Track Record
- Public Datasets & Machine Feeds
- API for Funds & Fintech
- Subscription Plans & Pricing
Frequently asked questions
What can a fund read in the Gemral Edge data preview?
The preview shows up to 20 real rows from each Gemral Edge table, with every column described from the data dictionary. Rows belong to a small sample universe of tickers and carry the ticker that was valid on the day each record became public. You can filter by ticker, sort any column and read the grain and point-in-time rule of each table before requesting the sample pack.
How do I judge alternative data providers from a sample?
Check four things in the rows: that the date a record became public is separate from the date of the underlying activity, that the grain says what one row means, that public identifiers let you join the rows to your own systems, and that history depth matches your backtest window. The data coverage page shows the same depth for the full database, year by year.
Is the preview the full dataset?
No. The preview is a window of up to 20 rows per table. The full sample pack contains Parquet files for the whole sample universe with a manifest of row counts and checksums, the data dictionary, a methodology note, the license and the Data Due Diligence Questionnaire. Funds that need the whole point-in-time history license it under the Institutional plan.
How do I get the full sample pack?
Email support@gemral.com with your fund name, the table you previewed and the questions you want the data to answer. The request button on the preview page opens a message with those fields prefilled.
Risk Disclaimer
Trading and investing in digital assets, financial instruments, and predictive events involve substantial risk of loss and are not suitable for every investor. The predictive intelligence, probability distributions, historical precedents, and scenario modeling presented on this page are compiled for informational and research purposes only and do not constitute financial, investment, legal, or tax advice. Past performance and statistical precedents do not guarantee future outcomes. Always conduct independent due diligence before committing capital.