Every number in the terminal is traceable to a public document or onchain transaction. This page explains where the data comes from, what is missing, and exactly how the Insider Score is calculated.
Open-market purchases and sales, grants, option exercises and tax withholding by officers, directors and 10% owners. Parsed from each filing's XML.
Execution date from the filing; disclosure date is the EDGAR acceptance timestamp.
Periodic Transaction Reports filed by members of the U.S. House. INSIDER reads the Clerk's annual index and extracts rows from each PTR PDF.
Members report within 30 days of notification and 45 days of the trade. Amounts are reported as ranges, never exact values.
Quarter-end long positions of tracked institutional managers. Changes (new, add, trim, exit) are derived by comparing the two most recent 13F-HR filings.
13F does not disclose trade dates. Positions are as of quarter end and filed up to 45 days later.
Transfers, mints and burns of official Robinhood Stock Tokens above a USD threshold. Token list and share multipliers come from Robinhood's registry API; USD uses its mid quote.
Onchain events are public at the block timestamp, so execution and disclosure are the same moment.
INSIDER stores two dates for every transaction: when it was executed and when it was published. Tables show both, plus the lag between them. Nothing in the terminal presents a delayed disclosure as a live trade — “new” always means newly disclosed. The only real-time feed is Robinhood Chain, where the transaction itself is the public record.
The Insider Score (v1.0) is a 0–100 measure of how unusual a disclosed transaction is relative to the factors below. It is a research aid for deciding what to read first. It is not an allegation, and a high score is not evidence of illegal trading. Executives, members of Congress and funds trade for many legitimate reasons.
| Factor | Max | What it measures |
|---|---|---|
| Transaction type | +20 | Discretionary open-market trades carry more signal than grants, tax withholding or option mechanics. |
| Relative size | +25 | Dollar size, and for Form 4 the share of the insider's position that changed. |
| Filing timing | +15 | Lag between execution and public disclosure versus the legal deadline. |
| Historical behavior | +15 | First-ever trades or reversals of an actor's past direction in the same security. |
| Frequency & clustering | +15 | Other insiders, members or wallets moving the same security in the same direction within 14 days. |
| Market events | +10 | Material company events (8-K, earnings, reports) filed within 60 days after the trade. |
| Pre-arranged plan | −10 | Trades under a Rule 10b5-1 plan were scheduled in advance and are discounted. |
Each trade page shows the points awarded per factor and the reason. The total is clamped to 0–100. Scores are recomputed after every ingestion so clustering and event factors reflect newly disclosed activity.
Congressional Exposure lists disclosed trades by a member (or their spouse or dependent) in companies whose industry falls under the jurisdiction of a committee the member sits on. Committee rosters come from the congress-legislators project; company industries come from each issuer's SEC SIC code. The mapping from committee to industry is an INSIDER classification, shown with its basis and a link to the committee's official jurisdiction. Only trades executed since 2025-01-03, under the current roster, are evaluated.
The Exposure Score (0–100) ranks members for research priority. It measures how strong and how unusual the documented overlap is. It is not a probability of wrongdoing, and a committee seat does not mean a member had material non-public information about a company.
| Factor | Max | What it measures |
|---|---|---|
| Jurisdiction overlap vs. Congress baseline | +25 | Share of the member's industry-classified trades inside committee jurisdiction, above the share across all members. |
| Number of overlapping trades | +20 | Log-scaled count of overlapping trades; 20 or more earns the maximum. |
| Reported size | +15 | Log-scaled sum of the lower bounds of the disclosed ranges, from $1K to $5M. |
| Disclosure timing | +15 | Share of overlapping trades filed past a STOCK Act deadline by reported dates; half or more earns the maximum. |
| Concentration | +10 | Herfindahl index across the securities traded, when at least three trades overlap. |
| Independent corroborating records | +15 | Companies with a federal award within 90 days, a lobbying filing within 120 days, or a PAC contribution to the member. Each is a separate public record, not proof of a link. |
The STOCK Act requires a Periodic Transaction Report within 30 days of the filer being notified of a transaction, and in no case later than 45 days after it. INSIDER evaluates both tests on the dates the filer reported. A row past either limit is labeled “past deadline by reported dates”, not as a violation: extensions, late-filing fees and ethics-committee waivers are not visible in the report.
Rows the filer marked as amendments are not evaluated, because the amendment's filing date is not the original report's. Rows marked deleted are excluded. Senate rows from the mirror carry no notification date or amendment flag, so only the 45-day test applies and late labels note that the amendment status is unknown.
Federal awards come from the USAspending.gov API: the largest new contracts and grants signed in a rolling 60-day window, their obligation actions, and aggregate obligations by agency, recipient and industry for that window and the one before it. Three amounts are kept apart. Obligated is money the government has committed. Outlays are cash actually paid. The ceiling (base plus all options) is the most an award could reach. None of them is revenue to the company.
Recipients are linked to a listed company through the SAM-reported parent, a curated subsidiary map (from the MIT-licensed LuxAlgo market-trackers project), or an exact SEC legal-name match. Prefix and loose-name candidates go to a review queue and are never shown as fact. Department of Defense contract data is published with a 90-day delay, so recent windows are understated.
Lobbying reports (LD-2) come from the Senate LDA API, with the CC0 market-trackers-data mirror filling gaps. A company's spend for a quarter is its own in-house report when it filed one, because that figure includes fees paid to outside firms; otherwise it is the sum of outside firms' reported income. Mirror rows are dated by when the mirror retrieved them, which can be later than the Senate posting.
PAC data comes from FEC bulk files for the current cycle. A corporate PAC is linked to a company only through its registered connected organization. Contributions are aggregated per PAC and candidate; memo entries are excluded.
Convergence looks for companies where several independent kinds of public record appear in the same window: member trades (and trades within a member's committee jurisdiction), Form 4 open-market trades, 13F position changes, new federal awards, lobbying reports, contributions from the company's PAC to a member who traded it, and 8-K material events.
Each signal type's base rate is the share of companies showing it in the window. If types were independent, the chance of a company showing a given combination would be the product of those rates. The score is that improbability in bits (−log₂), weighted by source reliability (mirror records 0.9) and discounted by up to 30% when signals spread beyond 30 days; 16 bits, about 1 in 65,000, scores 100. The independence assumption is false in practice: large companies attract every kind of signal at once. Treat the score as a reading order, not a finding.
Every signal is dated by when it became public: filing dates for disclosures, signing date plus 90 days for Defense awards, and the quarterly FEC deadline (quarter end plus 15 days) for PAC contributions. This is what lets Historical Replay show a past date without look-ahead.
Historical Replay recomputes the full convergence ranking once a week for the past year, each time using only signals public on or before that date, and draws prices only up to the frame date. Early frames score lower partly because INSIDER holds fewer records for those weeks; the replay page lists when each source's records begin.
The network graph only draws links that are stated in a source document: an insider reporting on an issuer's Form 4, a member disclosing a trade in a security, a 13F position above 1% of a manager's portfolio, a token transfer between two wallets, or a Stock Token tokenizing a listed company. Every edge links to its evidence. INSIDER does not infer family ties, employment or coordination.
The Investigator retrieves matching records from INSIDER's database and answers only from them. Each claim carries a citation to a numbered source; citations to records that weren't retrieved are stripped before display. If no record matches, it says so instead of answering from memory. Without a language model key configured, it returns a structured summary of the same records.
Beyond trades and company filings, the question is routed to structured searches by keyword: the Disclosure Clock for timing, federal awards (optionally filtered by a named agency), Congressional Exposure for committee overlap, lobbying reports, PAC contributions, Washington Convergence, and the Pathfinder for documented connections between two named entities. These are database queries, not model calls. Every answer lists each dataset searched and how many records it returned, including zero.
Parameters are taken from Robinhood's chain documentation. Stock Token contract addresses are never hard-coded; they're loaded at runtime from the official asset registry. Documented tokens: WETH 0x0Bd7D308f8E1639FAb988df18A8011f41EAcAD73, USDG 0x5fc5360D0400a0Fd4f2af552ADD042D716F1d168. INSIDER deploys no contracts. Premium credits are an ERC-20 transfer to an operator-configured treasury, verified from the transaction receipt.
INSIDER aggregates public disclosures for research and journalism. It is not investment advice, and nothing here asserts that any person broke the law. Data may be incomplete, delayed or parsed incorrectly — always check the original filing linked from every record.