Platform
How Pinchfin Produces Institutional-Grade Output
Deterministic calculations, domain-specific agents, and artificial inference — organized into agentic workflows that produce consistent, committee-grade output and compound across every deal in the pipeline.
Deterministic Python calculations, agentic workflows, and verified government data feeds.
Every calculation in Pinchfin uses standardized CRE formulas written in Python — DSCR, debt yield, LTV, IRR/MOIC, interest reserve shortfall, sensitivity tables. The math is deterministic: same inputs always produce the same output. No AI is involved in any calculation. AI handles research, analysis, and document production. All financial math is deterministic Python.
| Component | Function | Why It Matters |
|---|---|---|
| Six-Constraint Waterfall | LTV As-Is, LTV Stabilized, DSCR, Debt Yield, LTC, and program cap tested simultaneously. Most restrictive governs. Full show-your-work output. | Same inputs always produce the same answer. No analyst judgment in the sizing math. |
| Interest Reserve Shortfall Method | Month-by-month MAX(0, Debt Service - NOI) across the business plan period. Accounts for lease-up velocity, concession burn-off, and renovation timeline. | Reserve precision-sized to the cash flow trajectory. |
| Dual-Scenario Cash Flow | Parallel proformas: Sponsor Proforma (counterparty assumptions) and Lender Underwrite (independently derived). Variance analysis across all material line items. | Committee sees both perspectives. Sizing uses lender figures exclusively. |
Three-agent quality assurance on every critical analytical phase. The Producer executes, the Reviewer audits, and the Resolver decides. Structured escalation hierarchy with defined authority at each level.
| Component | Function | Why It Matters |
|---|---|---|
| Producer | Executes the analytical phase per skill specification. Produces the output, runs the calculations, generates the deliverable. | Single-owner execution. One agent responsible for the work product. |
| Reviewer | Audits the Producer's output against skill requirements, calculation accuracy, source verification, and format compliance. | Independent audit catches errors, source gaps, and format drift before delivery. |
| Resolver | MD-level authority. Receives Reviewer findings and decides: FIX (send back) or ACCEPT WITH EXCEPTION (document and proceed). Decisions are final. | Structured escalation prevents infinite review loops. Defined authority at every level. |
Specialized analytical handling for complex deal attributes. When a deal involves rent control, LIHTC, ground lease, seismic, or other structural complexity, the Nuance Gate Agent routes to purpose-built methodology with dedicated skill chains for each nuance type.
| Component | Function | Why It Matters |
|---|---|---|
| Nuance Detection | Automated classification at intake: rent control, ground lease, tax abatement, Section 8, LIHTC, 721 exchange, environmental, seismic, TIC structures. | Structural complexity identified at intake and routed to specialized analytical methodology. |
| Gate-Keeping vs. Contained | Gate-keeping nuances (could kill the deal) trigger early GO/NO-GO. Contained nuances (known impact) route to specialized methodology in their respective phases. | Deal-killers evaluated early. Known complexity handled with the right tools. |
| Purpose-Built Methodology | Each nuance type has its own analytical pathway: rent ceiling for LIHTC, reversion for leasehold, regulatory stacking for affordable, PML thresholds for seismic. | Generic underwriting misses structural risk. Specialized methodology does not. |
Every rent assumption and value conclusion is derived from independently sourced market comparables. The Comparable Analysis Agent produces lender-underwritten figures with adjustment grids and documented methodology.
| Component | Function | Why It Matters |
|---|---|---|
| Rent Comparable Analysis | 5-7 comparables with $/SF, $/unit, concession adjustment, vintage adjustment, amenity normalization. Market rent conclusion with confidence interval. | Rent assumptions drive all downstream credit metrics — independently derived from market comparables. |
| Sales Comparable Analysis | 5-7 transactions with per-unit, per-SF, and cap rate analysis. Adjustment grids for location, quality, vintage, and market timing. | Value conclusions backed by comparable evidence with documented adjustments. |
| Comparable Database | Every comparable across all deals is indexed and persisted. New deals in the same market start with an existing comp set. 90-day rent / 180-day sales shelf life. | Each deal enriches the comp set for subsequent analysis in the same market. |
Every research phase queries verified institutional sources, government databases, and the platform's own deal and market database before generating analysis. The system reads before it researches, and it researches before it writes.
| Component | Function | Why It Matters |
|---|---|---|
| Prior Analysis Review | Before external research, the system checks the deal and market database for existing market reports, comp sets, and expense benchmarks. Prior analysis loaded as baseline and enriched with current data. | Each deal builds on prior work. 60-day freshness enforced without redundant queries. |
| Source Hierarchy | Tier 1: CBRE, JLL, CoStar, RealPage, Yardi Matrix, Trepp, Census/BLS. Tier 2: Apartments.com, Zillow (cross-reference only). Tier 3: excluded from citation. | If it wouldn't appear in a CREFC white paper, it doesn't appear in the output. |
| 25 Verified Data Feeds | SOFR, Treasury curve, CPI, demographics, employment, flood zone, environmental, HUD rents, AMI limits, construction costs, utility rates, Walk Score — from Census, BLS, FRED, FEMA, EPA, HUD. | Government-verified data integrated at every analytical juncture. |
The foundational systems that make consistent, repeatable, committee-grade output possible across every deal, every analyst, every market.
| Component | Function | Why It Matters |
|---|---|---|
| Deal & Market Database | Persistent storage for market reports, comp records, expense benchmarks, and deal data records. Indexed, searchable, with freshness rules (60/90/180/365 days by data type). | Cumulative data — each deal contributes to a growing analytical base. |
| Credit Box Configuration | Constraint thresholds, sizing parameters, and credit policy codified in a single authoritative configuration. Three independent credit boxes: transitional, construction, preferred equity. | Every deal sized against the same constraints regardless of analyst or origination channel. |
| Execution Audit Trail | Every phase logs: skill loaded, timestamp, execution status, output file, duration. Full audit trail for every deal. | Every output traces to its source phase. Complete execution transparency. |
330+ analytical skills organized as independent, composable modules. Each skill has a defined input contract, output schema, and execution specification. Skills compose into skill chains executed by domain-specific agents. Agents chain together into agentic workflows — each governed by its own credit box, phase gates, and documentation standards.
| Component | Function | Why It Matters |
|---|---|---|
| Skill-Based Architecture | Each analytical capability is an independent skill with defined inputs, methodology, outputs, and validation criteria. Skills compose into skill chains executed by domain-specific agents. | New product types assembled from proven, validated components. |
| Agentic Workflow Execution | Maps skills to phases, enforces execution order, validates inter-phase data contracts, and manages blocking gates. 4 workflow modules, 330+ skills, 25 data feeds. | Consistency enforced by architecture across every execution. |
| Multi-Product Credit Box | Independent credit box per product type: transitional, construction, preferred equity. Each with its own constraint thresholds, sizing methodology, and documentation requirements. | Product expansion without compromising analytical rigor. |
End-to-end agentic workflows from deal intake through committee-grade deliverables. Multiple domain-specific agents chain together — each executing its own skills within its own analytical scope — to produce committee-grade output autonomously.
| Workflow | Scope | Pipeline |
|---|---|---|
| Transitional Debt | Bridge loans, value-add, lease-up, and balance sheet origination. Full credit analysis from intake through committee presentation, closing coordination, and portfolio surveillance. | 17 phases · 4 blocking gates · 143 skills · ~60 min |
| Construction Lending | Ground-up and substantial rehabilitation. Budget validation, GC diligence, draw administration, cost-to-complete monitoring, contingency burn analysis, and stabilization tracking. | 27 phases · Modified credit box (LTC ≤80%, DSCR ≥1.20x) |
| Preferred Equity | Two-pass constraint waterfall for combined capital structures. Senior debt and preferred equity sized independently with combined leverage tested. PIK modeling and waterfall integrity. | Two-pass waterfall · Combined LTV ≤80% |
| Sponsor Acquisitions | Full acquisition lifecycle from sourcing through hold-vs-exit. Five strategies: core-plus, value-add, opportunistic, lease-up, heavy repositioning. | 27 stages · 5 blocking gates · 13 red flag triggers |
| Quick Review | Rapid screening pass for every inbound deal: six-constraint sizing, preliminary credit metrics, market snapshot, sponsor profile, and 1-page screening memo. | 10 phases · ~10 minutes · Pipeline triage |