MARS · Substrate Report 01 · Weekly

What We Can Actually Serve Today

Field-by-field, entity-by-entity — coverage, gaps, and where to invest next.
Snapshot 2026-07-18

Weekly checkpoint on the MARS substrate. tl;dr: the substrate is deep on identity, activity, and SEC filings; thin on descriptive company facts PitchBook leads with (long-form description at 3.2%, founded_year / employee_count still missing as columns); uneven on person depth (32.8% of persons have a bio, only 2.8% have structured work history). Filling the P0 gaps below moves the needle for pilots in both the mid-market PE and family-office segments.

286.1K
Companies
canonical, resolved
1.42M
Persons
strong-ID resolved
100.4K
Investors
canonical, ADV-linked
115.8K
Deals
funding + M&A canonical
11.22M
Form 4 txns
insider trades · fresh 2026-07-17
765.1K
Form D filings
Reg D offerings · fresh 2026-07-17
115.3K
Form 144
planned insider sales
2.47M
Articles
indexed with metadata

01Companies — the identity graph

286,113 canonical companies. Identity fields (name, CIK, city/state) are well-covered thanks to SEC ingest + Atlas backfills. Two gaps that matter: (1) description at 3.2% — over 96% of companies have no long-form description; (2) PitchBook-standard fields founded_year, employee_count, website, hq_lat_lng, business_status are not columns in companies_v2 at all.

What we DO have, coverage:

primary_name
99.9%
country
83.4%
city
69.5%
state
69.1%
location (free-text)
69.5%
cik (SEC anchor)
65.7%
sector
24.8%
primary_domain
22.1%
ticker
2.4%
description
3.2%

What we DON'T have (columns missing entirely):

FieldWhy it mattersFix path
founded_yearEvery PB profile leads with it; screening filter for "companies founded 2020+"Atlas CIK profile has it; SEC Form D has it; add column + backfill
employee_countMost-referenced sizing field; family offices use it as sanity checkLinkedIn Company API (paid); derive from bios corpus
websiteprimary_domain covers most of the need at 22.1%Explicit column low priority
hq_lat_lngGeographic filteringFree geocode from city + state; no external cost
business_statusActive / Acquired / Wound Down / PublicDerivable from existing signals
Country normalization is dirty
"United States" (187,010 rows), "US" (28,305), "USA" (3,911) — three encodings of the same country coexist. Same fragmentation on Canada / United Kingdom / India. Half-day cleanup script; meaningful consumer-visible polish.

02Persons — 1.42M canonical, uneven depth

Persons graph is huge and growing. Coverage is thin where it matters most: only 32.8% of persons have a bio, and structured work history is at 2.8% — far below what a family-office analyst asking "who at these RIA firms went to Wharton" would need.

SignalPersons covered% of graphState
Has a name1.38M97.4% strong
Has a bio464.2K32.8% partial
CIK (SEC-anchored)211.2K14.9% partial
Structured education42.0K3.0% thin
Structured work history39.6K2.8% thin
LinkedIn URL45.4K3.2% thin
CRD (RIA rep)14.2K1.0% thin
The bio → work history gap is the priority
464,235 persons with text bios but only 39,625 with structured work history. Qwen extraction sweep on the existing bio corpus is $0-marginal. Directly unlocks "advisor migration"-style queries family offices want. Cheapest, highest-leverage substrate improvement.

03Investors — 100.4K canonical, ADV-linked

Best-shape entity in the graph for its size. 51.0% carry the CRD strong-ID; 22.4% have real RAUM figures from ADV. Descriptive fields (country, founded_year) still at 0%.

primary_name
100.0%
CRD
51.0%
current_raum
22.4%
website
18.6%
top_fund_type
7.2%
adv_cik
5.3%
country
0.0%
founded_year
0.0%

04Deals — funding is clean, mergers have gaps

Funding deals (79,096) — 100.0% amount, 99.3% date, 100.0% deal_kind. 42.3% linked to a Form D. Investor bridge on 50.8%.

FieldCoverageΔ vs prior
funding — amount / date / kind / tier~100%
funding: SEC Form D linkage42.3%
funding: canonical investor bridge50.8%
funding: valuations (pre + post)0.0% / 0.0%
funding: article provenance bridge65.4%

M&A deals (36,676):

FieldCoverageΔ vs prior
acquirer + target resolved100.0%
integrity_tier assigned100.0%
announcement_date96.0%
deal_value_usd (headline value)23.0%
close_date17.7%
upfront + max (milestone split)0.0% / 0.0%

05SEC substrate — the deepest layer

TableRowsΔ vs priorFresh through
form_4_transactions_v211.22M2026-07-17
form_d_filings_v2765.1K2026-07-17
form_d_officers_v21.42M
form_144_v2115.3K2026-07-18
schedule_13g_v213.8K
schedule_13d_v22.8K

06Verticals — LIVE / SCAFFOLDED split

VerticalRowsΔ vs priorFresh throughState
bios527.9K2026-07-15LIVE
reit (real estate)19.2K2026-07-18LIVE
private_credit94.3K2026-07-18LIVE
flare (thermal assets)16.1K2026-07-17P1 done
flare observations (P2)0pending
prospects (scoring)180.7K2026-07-17scoring live