The MARS Person Graph

A profile of persons_v2 — for the Insider Focus donor-research conversation

MARS data platform · snapshot 2026-08-24

Insider Focus's donor-research product is built today from SEC insider data — individuals who carry a CIK. That's roughly 220K people. The MARS person graph is the same substrate scaled out: SEC insiders plus RIA principals, fund officers, property owners, political donors, and the relationships that connect them — 2.78 million people, ~12.6× the CIK-only universe, and 99.7% of them attached to at least one piece of real evidence.

2.78M

total resolved persons

220K

carry a CIK (the current IF universe)

12.6×

expansion over CIK-only

0.3%

orphans (no evidence at all)

How the graph is identified

Every person is a canonical entity with a stable ID. A tenth carry a machine-strong identifier (the thing that makes a match unambiguous); the rest are anchored by employer, address, filing, or relationship and are continuously being resolved upward.

Identifier	Persons	% of graph
CIK (SEC)	220,442	7.9%
LinkedIn URL	45,997	1.7%
CRD (FINRA/SEC advisor)	14,183	0.5%
Other external IDs	1,167	0.0%
Any strong identifier	278,834	10.0%

The other 90% aren't unknowns — they're people resolved from filings, bios, deeds, and news whose strong ID simply hasn't arrived yet. About half the graph (49.8%) is flagged entity_resolution_pending: that's not a defect count, it's the active worklist our resolver and weekly dedup sweep grind down continuously.

How connected the graph is

The value isn't the row count — it's that people are attached to evidence and to each other. Coverage across the major bridges:

Linkage	Persons	% of graph
In the relationship graph (co-directors, kin, colleagues)	2,001,117	72.0%
Work history (employer + title + dates)	1,621,283	58.3%
Person↔person edges	997,184	35.9%
Has a narrative bio	755,801	27.2%
SEC Form D officer	737,785	26.5%
Property / deeds ownership	255,057	9.2%
SEC Form 4 insider transactions	208,612	7.5%
Education	206,062	7.4%
M&A / funding deal officer	59,257	2.1%
Political donations (FEC)	31,153	1.1%
Executive / director compensation	28,767	1.0%
Orphans (in none of the above)	8,580	0.3%

The wealth-signal layer IF doesn't have today. A CIK tells you someone is an insider. It doesn't tell you what they own, who they know, or how they give. MARS attaches 255K people to property ownership records, 31K to political donations, and 29K to disclosed executive/director compensation — the exact signals donor research is built on, resolved to the same person entity.

Where the people come from

The graph is deliberately multi-source — no single feed dominates, which is why it reaches so far beyond SEC insiders.

Source	What it is	Persons	Share
bios	RIA principals & advisor bios (SEC ADV + web)	1,236,943	44.5%
mars	SEC Form D / 144 / 13D-G officers + news	800,758	28.8%
deeds	Property owners & public records	555,799	20.0%
insider	SEC Form 4 insiders	181,602	6.5%
sedar / signals / re	Canadian filings, news signals, real estate	4,123	0.1%

What MARS adds to the CIK universe itself

This isn't only "more rows." The 220K CIK insiders IF already has get materially richer inside MARS — same people, more attached signal:

Of the 220,442 CIK persons…	Also in MARS	Coverage
…have work history	218,781	99.2%
…are in the relationship graph	216,304	98.1%
…have a narrative bio	24,607	11.2%
…link to property ownership	1,244	0.6%

Property coverage on the CIK slice is low today because our property↔SEC person linking is early — it's an active workstream (see below), and the 255K property owners already in the graph are the pool it links against.

How we keep it clean, and keep improving it

A person graph this size is only useful if it's deduplicated and continuously linked. Both run as standing processes, not one-off cleanups:

A write gate prevents duplicates at creation. Every new person is matched on strong ID first, then on anchors (employer, resolved company, full address, relationship); a name alone can't create a row, and a name that collides with an existing person defers to review instead of duplicating.
A weekly dedup sweep collapses what slips through. Just this cycle it merged ~46K exact-name cross-source duplicates and ~13K same-name-same-address duplicates, all reversible and logged.
Linking is never "done." New strong IDs (a CIK, a CRD) arriving later trigger re-resolution; a weekly deterministic sweep re-links foreign keys as new evidence lands. Property↔SEC and donor↔person linking are the current frontiers.

On duplicate scale, stated honestly. 143K names are shared by two or more rows, which puts a theoretical upper bound of ~288K on possible duplicates. That number is an upper bound, not a defect count — most people who share a name are genuinely different people (there are many real "James Smith"s). The actionable duplicate set is the far smaller slice where two rows also share an anchor (address, employer, or ID), and that's exactly what the weekly sweep targets.

The bottom line for Insider Focus

Today's CIK-only universe (~220K) is fully present in MARS and 98–99% of it is already enriched with work history and relationships.
MARS adds ~2.56M additional people beyond CIK — RIA principals, fund officers, property owners, and donors — the individuals donor research actually targets.
It ships with the wealth-signal bridges built in: property, compensation, donations, and a relationship graph spanning 2M people.
It's kept clean by an always-on gate + weekly dedup, and it gets richer every week as linking runs.

Figures are live counts from mars.persons_v2 and its bridge tables, 2026-08-24. Percentages are of the 2,779,225-person total unless noted. Prepared by the MARS data platform for Raul Peralta.