What changed after checking the sources
| Item | Source-verified correction | Model consequence |
|---|---|---|
| Kiavi consideration | Approximately $590M net of acquired cash at closing, versus $532.4M June agreement | Use closing cash, and actual $586.5M note proceeds; do not use gross $600M as proceeds. [E2, F1] |
| Kiavi economics | Management disclosed FY25 revenue >$250M, EBITDA >$100M, volume~$7B | Historical floors exist; post-JV revenue/cost accounting remains unknown. [T3] |
| Q3 volume | $5.119B explicitly excludes Kiavi | One September month is separate; do not double count or assume new historical series comparability. [E1] |
| Near-term demand | October preliminary $1.67B versus September rounded $1.78B | Base Q4 legacy anchor is 3×October≈$5.01B, not the old unsupported $5.6B. No claim of seasonally adjusted decline. [D1] |
| Servicing | $16.0B weighted-average HELOC UPB; $17.341B total ending UPB | Forecast stock, new servicing and runoff separately. [F1] |
| Q2 MSR gain | $31.807M creation −$11.109M runoff +$8.395M valuation =$29.093M | Separate noncash creation/runoff from valuation. [F1] |
| EBITDA quality | $119.379M includes~$5.9M realized Reflow disposal gain | Recurring comparison≈$113.5M before other judgments; no recurring sale gain forecast. [F1, T1] |
| History | Q3’25 net take 4.4%; Q4’25 EBITDA 81.330M; latest-vintage Q1’26 EBITDA 82.613M | Corrected source series; original vintages retained. [F1, F2, E3] |
| Next checkpoint | November 12 before open; 8:30 am ET call | Fixed review date, not unspecified mid-November. [D2] |
Business as a machine
There is one reportable segment. Figure-branded origination, partner LOS, Connect, servicing and digital products are operating channels. Figure earns fees on loan dollars, gains when loans are sold, servicing income on outstanding balances, and income from financed assets. Loan timing, retained capital and fair-value accounting sit between volume growth and cash. Kiavi adds a separately acquired loan cohort from September 1; its pre-acquisition interest-led P&L changes as RTL assets move to the Sixth Street structure. [F1, T3]
schematic.md contains three complete Mermaid diagrams: legacy revenue, cost/capital/cash, and Kiavi. The core chain is:
Partner cohorts × activity × loan size
Channel mix × eligible dollars × realized yield
Earnings − capital required − noncash gains
Metric harvest and reporting frame
The long register contains 9,886 source observations and 2,034 concept/dimension IDs, including duplicate source vintages: A 8,015; B 236; C 1,502; D 66; E 67. A means audited-or-XBRL per this workflow; it does not imply interim statements were audited. Assumptions are separate PLUG leaves and are not promoted into disclosed grades. guidance_ledger.csv retains 68 rows including historical preliminary-result ranges, distinguished from prospective guidance. [harvest_summary.json]
Coverage: four available periodic filings with native extension/dimensional XBRL, IPO S1 financial history, eleven SEC response letters,19 8-Ks/30 documents, twelve unique Koyfin events including four earnings calls, and 17 issuer-verified library events for context. 132 transcript files were screened because the FIGR folder also contains unrelated issuers. Numeric dispositions cover 1,199 Koyfin sentences and 1,901 tokens. This is not a claim of eight historical public earnings calls or literal exhaustive normalization of every historical public statement. [source manifests; transcript harvest summary]
Key comparability breaks:2024 separation/common-control perimeter; 2025 recombination; 2026 marketable-income reclassification; broadened CLM definition; adjusted-revenue/EBITDA changes; Prime correction/rename; pending Kiavi volume recast. Blank old marketable-income cells mean included elsewhere, not zero. No weighted-average share count or rate is quarterized by subtracting YTD averages. [F1–F3; definition_changes.csv]
Operating primitives, formulas and evidence
| Output | Primitive identity | Evidence and weak edge |
|---|---|---|
| Partner volume | Platform partner cohorts × assumed activity/productivity × ramp × demand/seasonality | Partners 387→489 Q1→Q2; 3–6 month large-partner ramp. Active-share, churn and mature productivity unresolved; 489 is platform partners, not confirmed active producers. [T1, T2] |
| Technology fee | Eligible loan dollars × effective fee yield | Q2 fee 29.176M; fee-eligible subset undisclosed. Whole-CLM denominator is an E blended proxy. [F1, F3] |
| Ecosystem/program fees | Connect dollars × fee yield + sold/securitized dollars × program yield | Q2 ecosystem 36.768M/program 6.921M. Sliding tiers compress yield. [F1] |
| Origination | Direct dollars × direct fee rate + nonConnect dollars × partner rate | Q2 direct 23.533M on 723.075M; partner 2.813M on inferred nonConnect 762.786M. [F1, E derivation] |
| Sale gains | Loans sold by eligible channel × realized spread; marks separate | Q2 whole-loan realized 54.033M on rounded 2.5B sold. Current-period sale/production ratio is not a cohort identity. [F1] |
| Servicing | Average serviced UPB × annual fee/4 | Q2 fees 11.303M; average 16.0B, annualized effective~28 bp. Stock additions/runoff are explicit. [F1] |
| MSR | Opening rights + created rights − runoff + valuation | Q2 exact three-part bridge above; creation yield E, future marks scenario. [F1] |
| Interest | Average held assets × yield + cash-income residual | Coupon 8.8–9.0% is new-loan proxy; 24.5 day hold midpoint from 3–4 week statement. Residual 9.517M/q is E/PLUG. [F1, T1] |
| Processing / ads | Loan dollars × processing unit cost; direct dollars × ad cost | Q2 processing 19.940M; advertising/other 25.368M. December CFO~750/loan is a separate scope cross-check, not the processing-only rate. Avoid revenue percentage. [F1] |
| People/programs | Functional salary pools × hiring/wage indices + professional/software programs | Absolute functional headcounts undisclosed; indexed pools preserve this gap. [F1] |
| Funding / capital | Held-loan days, advance rates, retained additions/runoff; balances × rates | Contracts 80–100% advances; model 90%. Financing availability not guaranteed. [F1] |
| Kiavi | Separate acquired volume × post-JV revenue/profit yields − integration cost + synergies | Revenue/profit anchored to historical floors, not post-close guidance; 35M synergy target over 24 months, base half. [T3] |
Every adjustable leaf has base/bull/bear values, units, provenance and rationale in assumptions.csv. Fixed anchor/allocation choices are in sources/model_anchors.json; machine-readable formulas/edge types in driver_tree.yaml. The forecast engine is model.py, and outputs are regenerated by build_model.py. FX is not an active primitive because no material non-USD operating exposure was identified. [Model]
History and calibration
| Quarter | CLM $M | Revenue $M | Adj EBITDA $M | Net income $M |
|---|---|---|---|---|
| 2024Q1 | gap | 75.3 | gap | -28.5 |
| 2024Q2 | gap | 80.8 | gap | 15.1 |
| 2024Q3 | 1,450.5 | 101.0 | 49.4 | 27.4 |
| 2024Q4 | 1,171.8 | 83.9 | 15.5 | 5.9 |
| 2025Q1 | 1,365.1 | 84.5 | 28.3 | -0.6 |
| 2025Q2 | 1,838.0 | 106.1 | 52.9 | 30.0 |
| 2025Q3 | 2,469.4 | 156.4 | 86.4 | 89.8 |
| 2025Q4 | 2,704.6 | 159.9 | 81.3 | 15.1 |
| 2026Q1 | 2,902.4 | 167.0 | 82.6 | 45.0 |
| 2026Q2 | 4,258.8 | 225.6 | 119.4 | 87.4 |
Source: history.csv → canonical_history_long.csv accession/table/formula. Definitions differ across the marked presentation break; see that column before calculating growth or margins.
All 50 historical accounting identities pass: revenue components, expense components, operating profit, pretax and net income. That demonstrates arithmetic integrity, not forecasting skill. A Q1-only effective revenue yield applied to actual Q2 volume overpredicts Q2 revenue by 8.63% ($19.47M). Even supplying the actual volume does not fix mix/rate/sale-timing differences. Behavioral edges therefore have ranges, not fitted point estimates presented as empirical laws. [calibration.csv]
Prospective Q2 volume guide $3.8–4.1B versus actual $4.259B beat the high end 3.9%; Q3 guide $4.8–5.2B versus preliminary $5.119B is within range. LatestQ3 financial guidance did not specify sufficient revenue/cost primitives for a blind full-P&L test. Q3 volume in this forecast is conditioned on known preliminary data; Q4 is calibrated to October, so neither is advertised as an out-of-sample success. [guidance_backtest.csv; T1, T2, E1]
Q2 source bridge: net income 87.436 −MSR valuation 8.395 +marketable FV2.195 +digital/investment FV1.068 +registration 0.842 +acquisition 4.676 +restructuring 0.002 +SBC26.098 +software amortization 4.352 +nonfunding interest 5.553 −tax benefit 4.448 =adjusted EBITDA 119.379 ($M). Funding interest 14.117 remains deducted. Normalized tax 26% is CFO's longer-term statement; actual cash tax timing remains uncertain. [F1, T1]
Forecast from leaves
All financial values below are $M except per-share figures. The EBIT field means finance-company operating earnings after funding interest, before corporate/nonfunding interest; it is not generic all-interest-added-back EBIT. EPS is a proxy before undisclosed Kiavi purchase-accounting amortization in the base/bull cases; bear includes an illustrative 10M/q amortization deduction. Cash is after modeled interest/tax/capex and balance investments but before new debt funding, with unknown base Kiavi incremental working capital set tozero and prominently flagged. These are not fully reconciled GAAP EPS/CFO forecasts.
| Base period | Legacy CLM | Kiavi volume | Revenue | Adj EBITDA proxy | EPS proxy | Cash proxy |
|---|---|---|---|---|---|---|
| 2026Q3 | 5,119 | 595 | 262.4 | 137.6 | 0.28 | 35.3 |
| 2026Q4 | 5,010 | 1,821 | 302.0 | 151.2 | 0.32 | 82.1 |
| 2027Q1 | 5,217 | 1,857 | 308.6 | 155.2 | 0.32 | 75.7 |
| 2027Q2 | 6,298 | 1,894 | 340.8 | 179.5 | 0.39 | 61.4 |
| 2027Q3 | 6,549 | 1,932 | 348.5 | 184.0 | 0.40 | 94.6 |
| 2027Q4 | 6,764 | 1,971 | 354.7 | 187.5 | 0.41 | 101.3 |
| 2028Q1 | 6,882 | 2,010 | 358.1 | 188.7 | 0.41 | 108.3 |
| 2028Q2 | 8,172 | 2,050 | 392.6 | 214.3 | 0.48 | 94.1 |
Q3’26 legacy volume is preliminary known data, all financials estimated. Q4 volume anchor 3×October is a calibration, not a company guide. Q3 Kiavi month is estimated from a historical annual run-rate. [Model; E1; D1; T3]
| FY2027 | Bear | Base | Bull |
|---|---|---|---|
| Total loan volume $B | 23.2 | 32.5 | 43.5 |
| Revenue $M | 799 | 1,353 | 1,934 |
| Adj EBITDA proxy $M | 237 | 706 | 1,199 |
| Operating earnings after funding $M | 1 | 576 | 1,129 |
| EPS proxy $ | -0.23 | 1.53 | 3.35 |
| Cash before debt funding $M | -26 | 333 | 651 |
| Cash after modeled funding draws $M | 62 | 408 | 745 |
Scenario differences are operating choices: partner adds 25/60/100 quarterly; first-quarter ramp 25/50/75%; mature productivity−2.5/0/+2.5%quarterly; direct demand−2/2.5/6%; tier-fee retention 97/99/100%quarterly; held days 35/24.5/21. All are labeled ranges/assumptions, not management promises. Cash-capital choices intentionally widen downside. Kiavi profit-yield and reported revenue-yield are separate: reported revenue affects a residual cost equally, so unchanged profit under that presentation shock is imposed, not evidence that Kiavi revenue uncertainty is harmless. [assumptions.csv; model.py]
| Base annual | Revenue | Adj EBITDA proxy | EPS proxy | Cash proxy |
|---|---|---|---|---|
| 2027 | 1,353 | 706 | 1.53 | 333 |
| 2028 | 1,549 | 835 | 1.83 | 464 |
| 2029 | 1,719 | 933 | 2.04 | 586 |
| 2030 | 1,876 | 1,017 | 2.19 | 674 |
| 2031 | 2,092 | 1,157 | 2.49 | 716 |
Years 3–5 are mechanical scenario extensions with weaker evidence. terminal_l1.csv gives a 2032 flat-quantity L1 snapshot; no terminal multiple, perpetuity growth or DCF value is asserted. The 2031 notes are assumed refinanced; corporate principal repayment is not silently counted as distributable cash. [Model]
Sensitivities, unit contribution and the crux five
| Top cash leaf | Bear Δ cash $M | Bull Δ cash $M | Share of squared-range diagnostic |
|---|---|---|---|
| retention_rate | -81.6 | 40.8 | 24.9% |
| retained_runoff | -62.3 | 43.4 | 18.5% |
| acquired_wc_cash | -80.0 | 0.0 | 10.6% |
| gain_on_sale_yield | -48.2 | 29.1 | 9.9% |
| connect_step | -44.6 | 22.3 | 7.4% |
The top five account for 71.4% of the chosen independent equal-weight uniform-input squared-range diagnostic. That is a sensitivity-ranking device, not measured forecast variance or a probabilistic confidence interval. Input ranges are subjective and correlated. EBIT/EPS rankings differ; full year 1/year 3 tornadoes and 15-leaf full-chain local derivatives are in sensitivities.md andCSV.
| Crux | Current anchor / base choice | Falsifier and when |
|---|---|---|
| Retained investment rate | Q2 retained securities 354M; base new retention 1.5% of modeled sold flow, undisclosed | Required rate>2.5% or collateral haircut; Q3 balance/cash notesNovember 12. |
| Retained runoff | No disclosed full maturity vintage; base 10%quarterly | Actual repayments fail to cover 5% of opening balance; next two filings. |
| Kiavi incremental cash | Closing~590M; base incremental WC0 is an open gap, adverse 20M/q | Cash demands beyond acquisition consideration or retained/JV capital; purchase allocation andQ3 cash flow. |
| Realized sale yield | Q2 whole-loan 54.033M /rounded 2500M≈2.16% | <1.6% on comparable loan mix or sale lag extends; Q3 note/MD&A. |
| Connect migration | Q2 Connect 2772.935/partner 3535.721≈78.4%; base+1 percentage point/quarter | Adoption stalls, tier pressure exceeds assumed range, or legacy sale economics disappear faster; November 12 print. |
Sources: F1, E2; model assumptions and sensitivity.csv. Historical Q3’25 management illustration was 3%ecosystem +40 bpsecuritization +25 bpannual servicing; this is not a universal current Connect contract. It cannot be directly compared with the latest blended effective fee without eligible-volume/tier data. [transcript register, Q3’25 CEO]
Street-implied values are unavailable because no verifiable consensus series was sourced. No invented disagreement with consensus is claimed.
| FY2027 partial transaction contribution per $1,000 loan dollars | Base |
|---|---|
| Direct | 40.56 |
| Connect | 14.73 |
| Non-Connect | 46.78 |
This is a partial transaction contribution: attributable fees plus assumed realized sale/program economics, less dollar-driven processing and direct advertising. Common people, funding, servicing and credit costs are excluded because channel allocations are undisclosed. Sale participation is an inferred timing ratio, so these are not audited channel margins. MSR creation and marks are excluded from this cash-transaction view. Low-capital Connect can be economically attractive without maximizing reported revenue per loan dollar. [Model]
MSR valuation alone is removed from adjusted EBITDA and reversed in the cash bridge, although tax timing and future MSR runoff can create indirect cash consequences. A genuine deterioration in borrower credit, loan-sale demand or servicing collections is a different joint shock. The capacity curve is illustrative buyer capacity, not contracted funding or a forecast of the exact threshold. [validation.json; thresholds.csv]
Monitoring and implementation
Use the weekly issuer dashboard for volume, but do not equate on-chain transfers/write-scope activity with unique loans, revenue, or cash. November 12 must reconcile legacy versus Kiavi volume, new historical definitions, purchased assets, retained capital and contribution. The existing production nowcast configuration still contains pending-Kiavi and old-guide values; this run does not use those stale assumptions. super_model/FIGR/monitor.py saves independent local snapshots and sends no messages. Full cadence/falsifier/caveat table: monitoring.md. [D1; local pipeline config]
What remains uncertain
The five largest gaps are retained-asset cohort funding, Kiavi post-JV accounting/cash, fee eligibility and tiering, sold-loan vintage timing, and functional headcounts/productivity. The model uses named E/PLUG nodes for each. The ylds_balance leaf is presentation-only, affecting the adjusted-revenue denominator; it has no independent earnings/cash sensitivity. YLDS35 bp and Prime 50 bp pricing statements are captured; gross standalone fee illustrations are not added to consolidated earnings because existing interest/residual lines and intra-group activity require a revenue bridge. A Q2 gross-cash-minus-debt EV is inappropriate without separating servicing collections, YLDS backing, restricted balances and future collateral needs. Cash/loan interest splits, realized funding costs and tax timing are not fully known. Standalone Kiavi historical floors do not fix post-close GAAP revenue. [F1, F3, T3]
RKT and UWMC primary-source peer checks were used only to interrogate cost/holding assumptions. Rocket's loan holding time is on a weeks scale; UWM’s incentives make direct-cost behavior nonstationary. Neither supplies Figure's missing cost elasticity; no peer coefficient is transplanted. FNTK health was up, but most Q&A calls returned timeout errors; direct filings replaced them. The capital-structure tool succeeded and its extracted skyline is preserved under sources/tool_outputs/capital_structure/, with subsequent events checked separately. [tool logs; peer_edge_check]
This run delivers the requested source register, guidance ledger, historical reconciliation, driver tree, eight-quarter and longer scenario forecasts, sensitivities, monitoring and questions. It does not certify every harvest gate as fully resolved: full historical KPI disappearance testing is constrained by four available earnings calls,554 captured narrative passages are not all normalized, and the forward tree cannot be independently validated at every hidden primitive. These limits are part of the deliverable, not filled with invented observations. Detailed questions and every PLUG: open_questions.md.
Source appendix and reproduction
- F1: Q2 2026 10-Q, accession 0001628280-26-057061, filed August 14. Notes 4–6,8,12,14; MD&A KPI, revenue/cost and non-GAAP tables. Local raw/text/instance:
harvest_filings/2026-08-14_10-Q_000162828026057061.*. - F2: FY2025 10-K, accession 0002064124-26-000009, filed March 16; Q1 2026 10-Q0002064124-26-000029; Q3 2025 10-Q0002064124-25-000032; final IPO S-1/A0001628280-25-041837. Exact paths/table indexes in
harvest_filings/canonical_history_long.csv. - F3: SEC company response, August 18 2025, comments 11–12: fee-eligible partner subset; sales timing; KPI meaning. Eleven correspondence responses retained.
- E1: Q3 preliminary release, accession 0001493152-26-046036, October 6 event; source raw
sources/q3_2026_prelim_ex99-1.htm. CLM excludes Kiavi; prior series to be recast. - E2: Kiavi closing 8-K0001493152-26-040931, September 1; approximately $590M net consideration and assumed/decommissioned financing details. Files in
harvest_releases/sources/. - E3: Four earnings releases and amendments,19 8-Ks,30 raw documents; URL/date/table row and exact signed units in
harvest_releases/metrics.csv. Q3’25 net take 4.4%; corrected Prime matched offers $22M atSeptember 30’25. - T1:
transcript_downloads/FIGR/Figure_Technology_Solutions_Aug_13_26_002.txt: Michael Tannenbaum and Minchung Kgil, Q2 earnings; individual line/speaker references inharvest_transcripts/metric_register_transcripts.csv. - T2:
Figure_Technology_Solutions_May_12_26_002.txt, Q1 earnings;Figure_Technology_Solutions_Feb_26_26_004.txt, Q4 earnings. Use verified source manifest for duplicate filenames. - T3: June 10 Kiavi acquisition call,
Figure_Technology_Solutions_Jun_10_26_001.txt: CFO historical Kiavi revenue/EBITDA; management post-JV economics and cost synergies. Exact canonical source filename/speaker/line recorded in transcript register. - D1: Issuer dashboard, retrieved November 5; dated window throughNovember 1. RawHTML and parsedJSON in
sources/dashboard_current.*; October $1.67B is rounded preliminary. - D2: Q3 results announcement, October 22: November 12 before open,8:30 am ET call.
- M1:
sources/tool_outputs/market_cap.txt, current-run Massive snapshot $19.98 and August 7 XBRL cover share total 224.377872M. Snapshot is not guaranteed current when this report is opened. - Model:
model.py,assumptions.csv,sources/model_anchors.json,forecast.csv,forecast_annual.json. All forward figures in this report are model outputs; not issuer disclosures or consensus.
The model package below includes the calculation engine, assumptions, forecasts and source register.
All large harvest files retain exact accession/context/dimension or transcript file/speaker/line. Source occurrences are not independent observations; first/last disclosure fields refer to the covered corpus, not first-ever public use. Forecast amounts are generated from the model, not hand-entered in the tables. Prior artifacts remain available for comparison.