Capital Intelligence: The Infrastructure for Private Markets
This page is for the buyer who has already decided they need this kind of capability. The question is whether this platform is real.
WHY THE STACK MATTERS
The Compounding Effect of Six Steps in One System
Each capability described on this page is individually valuable. Better data is better than bad data. Correct classification is better than a flat list. Trust path mapping is better than cold outreach. But the reason the platform produces results that no combination of point solutions can replicate is that the six steps compound when they run in sequence inside one system.
Trust path mapping only matters if the classification engine identified the right person first. Finding a warm path to the wrong person at CalPERS wastes the connection and poisons the relationship. Outreach drafting only lands if the trust mapping surfaced a real anchor. A personalized email that references a fabricated commonality is worse than a generic one.
Assembling five vendors into a workflow does not produce the same result. The classification from vendor A does not inform the path mapping in vendor B. The trust anchor from vendor C does not flow into the outreach drafting in vendor D. A platform that runs all six steps in a closed loop does not add the components. It multiplies them.
THE INFRASTRUCTURE
What the Platform Does
Intelligence Layer
Continuous Monitoring Across Global Markets
The platform runs autonomous intelligence agents that monitor, collect, and synthesize institutional data from primary sources around the clock. These are not scheduled batch jobs that run overnight. They are persistent processes that detect changes as they occur and propagate updates through the system in real time.
The agents monitor SEC EDGAR filings (Form ADV, Form D, Form 13F, Form PF), IRS Form 990 nonprofit disclosures, state regulatory databases, beneficial ownership records, government pension board minutes, and institutional websites. When CalPERS publishes board minutes approving a new allocation sleeve for infrastructure, the system captures it. When a family office posts a job listing for a "Director of Direct Investments, Real Estate," the system infers a new mandate opening and flags it.
A database is a snapshot. This is a system that knows what changed since yesterday and what that change means for the client's program.
Autonomous agents monitoring primary sources continuously
Classification Engine
Institutional Taxonomy Below the Organization Level
No commercial database in the market does what follows. A commercial database lists CalPERS as one row. The platform models CalPERS as a set of distinct investment functions: the private equity team, the real estate team, the infrastructure team, the venture capital allocation. Each function has different mandate parameters, different check size ranges, different decision-makers, and different allocation timelines.
Classification operates at the contact level, not the firm level. Two people at the same family office can have entirely different roles: a principal who writes checks and an advisor who recommends managers. The platform routes to the right person in the right function with the right context.
One institution. Four distinct investment personas.
California Public Employees'
CALPERS · $496B AUM
Private Equity Team
$500M–2B checks
Buyout · Growth · Venture
Head of PE Investments
Real Estate Team
$200M–1B checks
Core · Core-Plus · Value-Add
RE Portfolio Director
Infrastructure Team
$300M–1.5B checks
Transport · Energy · Utilities
Infra Investments Director
Venture Allocation
$50M–200M checks
Early · Growth · Late-stage
VC Program Manager
Data Architecture
Every Fact Carries Provenance
Every data point in the platform carries a source, a date, and a confidence score from 0.0 to 1.0. When sources conflict, the system resolves them through a strict hierarchy: regulatory filings rank highest, followed by institutional websites, verified databases, news sources, and self-reported social data at the bottom.
When a professional network profile shows one employer and an institutional website shows another, the system weights by recency and source reliability. A regulatory filing from last quarter overrides a profile that has not been updated in two years. Conflicting signals trigger automated re-verification.
This matters because every decision downstream depends on data quality. If classification is wrong, the wrong person gets contacted. The platform treats evidence quality as a first-class concern.
Source reliability hierarchy — higher-tier overrides lower when sources conflict
When conflicting signals appear, the system fires automated re-verification against primary sources. A regulatory filing from last quarter overrides a profile that has not been updated in two years.
Trust Graph
The Shortest Credible Path from Client to Target
Decision-makers at institutional allocators receive hundreds of unsolicited contacts per quarter. Whether someone takes the call depends on the path: who introduced you, what you share, why the recipient should care.
The graph contains 40+ distinct trust anchor types, each with a calibrated weight and a time-decay model. A co-investment carries a base weight of 0.9 and a seven-year half-life. A board overlap carries 0.8 with a five-year half-life. The platform runs betweenness centrality analysis to identify super-connectors who bridge otherwise-disconnected clusters, and community detection to segment the network into professional groups.
In a market that runs on trust and reputation, the path is the product.
Trust graph pathfinding with super-connector identification and community detection
40+ anchor types with calibrated weights, time decay, and confidence intervals. Sub-50ms pathfinding across millions of nodes.
Unconventional Data
The Signals No Database Has
Standard relationship intelligence platforms map what is already visible: professional network connections, email volume, calendar overlap. The platform goes further. It tracks signals that no competitor has the data acquisition infrastructure to replicate.
Private aviation transponder data, art auction co-bidding, museum board overlaps, horse racing syndicates, classic car rally co-participants, wine auction records, ultra-endurance event completions, DAO governance participation, and GitHub collaboration. These data layers are the platform's sustainable competitive advantage.
Nine data layers no competitor has the infrastructure to replicate
ADS-B transponder data maps shared tail numbers and travel patterns. FAA registry cross-referenced with deal timelines.
Christie's and Sotheby's results identify competing bidders for the same lots. Shared collecting sensibility signals network proximity.
Board membership rosters and loan records identify investors who serve together on cultural institution boards.
Co-ownership of racehorses through small syndicates (8-12 members) with shared financial interest and regular social interaction.
Multi-day events like the Mille Miglia and Copperstate 1000 create endurance bonds. Registration data identifies co-participants.
Auction house records reveal collectors who compete for the same vintages. Shared oenological interests create non-professional anchors.
Ironman completions, 100-mile ultramarathons. Two allocators who completed the same event share a bond that scales with difficulty.
DAO governance participation and POAP tokens reveal co-attendance at events and shared governance votes. Public and permanent.
Shared open-source contributions and repository co-maintenance signal professional affinity and technical alignment.
Building parallel data acquisition infrastructure would take years. These layers are the platform's sustainable competitive advantage.
Mandate Matching
Quantitative LP-GP Fit Scoring
GPs waste months pitching to LPs who will never allocate. Wrong check size. Wrong vintage. Wrong asset class. The platform eliminates this with a quantitative fit score that matches GP fund characteristics against LP mandate parameters before any outreach begins.
The scoring combines hard constraints and soft preferences across six dimensions: strategy match, geography alignment, check size fit, pacing and deployment capacity, performance versus the LP's return hurdle, and ESG alignment. The composite produces a fit score from 0 to 100.
Mandate matching runs before trust path mapping. A high trust score to the wrong investment function wastes the connection. The platform identifies mandate fit first, then finds the warm path to the right person.
Mandate matching scores LP-GP fit before trust path mapping begins
An LP that scores 15 out of 100 is filtered before the system spends resources mapping warm paths. An LP that scores 90 gets prioritized. Outreach concentrated on the institutions most likely to allocate.
Quality Controls
What the Platform Filters Out
Relationship intelligence is only useful if the relationships are real, current, and safe to activate. The platform applies two layers of quality control that most systems ignore entirely.
Stale relationship detection flags paths where trust anchors have decayed below usable thresholds. Hostile relationship suppression monitors for active litigation, negative news, and user-flagged overrides. The combination prevents the most common warm introduction failure: asking someone to make an introduction through a relationship that no longer exists or has turned adversarial.
Two layers of quality control most systems ignore entirely
Co-investment 2011 · 12 years old · 15% retention floor · No recent interaction detected
Recommend re-engaging via professional networks before requesting an introduction.
Active litigation detected between intermediary and target. Path suppressed entirely.
Outreach Infrastructure
Trust Anchor Personalization at Scale
The platform does not send mail merges. It drafts outreach that references the specific trust anchors the graph surfaces for each individual target. Every message reads as though a senior professional spent twenty minutes researching the recipient.
Execution runs under the client's brand. The platform manages sending cadence, volume controls, deliverability tuning, and response routing. Responses are classified by intent: active interest, timing preference, referral, mandate mismatch, decline. The client engages when there is a conversation to have.
Continuous Learning
Intelligence That Compounds with Every Program
Every program the platform runs generates intelligence that feeds back into the system. The platform knows which contact information is accurate, which contacts have moved firms, and which investor personas respond to which outreach styles, because it tracked rates across 100+ engagements segmented by persona type, asset class, geography, and message approach.
This intelligence is not available from any database vendor. It was generated inside the platform, from live engagements, with real responses, and it compounds with every program.
Every program compounds the intelligence for the next
THE RESULT
100+ Programs of Front-Line Intelligence That No Vendor Sells
When a pension fund allocator tells the platform "not now, try Q3," that timing preference tags the specific persona at that institution. When a family office principal responds to three different GPs' climate-focused strategies but ignores two traditional buyout pitches, that preference pattern informs future targeting. When an endowment CIO refers the outreach to a colleague on the real estate team, the system creates a new contact, classifies the persona, and records the referral chain as a trust anchor for future use.
Commercial databases do not know which LP replied "interested" to a co-investment opportunity last month. Placement intelligence vendors do not know which family office principal prefers outreach that leads with a shared board connection versus a shared alumni network. That data was generated inside the platform, from live engagements, with real responses, and it compounds with every program.
FREQUENTLY ASKED
Frequently Asked Questions
The platform powers managed programs. Clients engage Praxis Rock to run fundraising or deal origination programs. The intelligence infrastructure, classification engine, trust graph, and outreach systems operate behind the scenes. Clients see qualified conversations, not a dashboard.
Every fact carries provenance: source, date, and a confidence score from 0.0 to 1.0. When sources conflict, the system resolves through a strict hierarchy: regulatory filings rank highest (SEC EDGAR, IRS 990s), followed by institutional websites, verified databases, news sources, and self-reported social data at the bottom. When a professional network profile shows one employer and an institutional website shows another, the system weights by recency and source reliability. Conflicting signals trigger automated re-verification.
The autonomous agents run continuously. Active prospect data refreshes daily. Contacts associated with live programs are monitored in real time for job changes, firm moves, regulatory disclosures, and public signals. The system does not wait for a quarterly database refresh. When a pension fund posts new board minutes, the platform reads them that day.
Everything transfers to the client. Every investor profile, every target company record, every data asset, every piece of intelligence gathered during the program. The client owns it permanently. The platform retains anonymized aggregate intelligence (response patterns, timing preferences, deliverability data) that informs future programs across all clients.
A CRM shows that you share a professional network connection. The trust graph scores 40+ distinct anchor types with calibrated weights and time decay, runs pathfinding across multiple hops, and surfaces the shortest credible introduction chain with a composite trust score and confidence interval. "You share a connection" is a data point. "Your anchor LP co-invested with this allocator three times, most recently eighteen months ago, and that co-investor also sits on a nonprofit board with the target's CIO" is an introduction strategy.
Trust scores include uncertainty estimates calculated through Monte Carlo simulation. The platform does not present a single number as though it were precise. A score of 72 with a 95% confidence interval of [64, 80] tells you the connection is strong but the evidence has some ambiguity. A score of 72 with a confidence interval of [70, 74] tells you the evidence is tight. The platform surfaces both the score and the uncertainty so clients can calibrate their approach accordingly.
Before trust path mapping begins, the platform scores every LP in the universe for fit against the client's fund characteristics. The scoring combines hard constraints (asset class, geography, check size) with soft preferences (ESG alignment, performance hurdles, deployment capacity). An LP that scores 15 out of 100 on fit is filtered before the system spends resources mapping warm paths. An LP that scores 90 gets prioritized. The result is outreach concentrated on the institutions most likely to allocate.
Every trust anchor decays over time using calibrated half-life models. The platform flags paths where the underlying relationships have aged significantly, with explicit warnings and re-engagement recommendations. It does not present a co-investment from 2011 as though it carries the same weight as one from last year. Structural ties retain a minimum floor, but the platform distinguishes between a live relationship and a historical one.
The platform is running. The question is whether it is running for you.
Schedule a Conversation