Data room / Run 1, stage 1: the nine reviews
Insurance and actuarial
run-1/reviews/insurance-actuarial.md Open the file as written
Insurance & Actuarial Review — Equity Flow™ / Equity Shield™
Reviewer role: credentialed actuary (FSA/MAAA-equivalent), insurance-regulatory and insurance-asset-management analysis.
Materials: disclosures/equity-flow-invention-disclosure.txt (full, 39 sections); investor-facing Equity Shield section of public/equity-flow.html (verified marketing language).
Date: 2026-08-01.
Posture per mandate: adversarial. The product's two insurance claims are assumed to have failed; this report explains why. All arithmetic below was computed in Python (models on file in the session scratchpad); key parameters: standard deal $1,500/mo × 84 months = $126,000 commitment, enrollment value $625,000, first mortgage $350,000 (56% LTV), settlement waterfall per Section 29, junior non-recourse position per Sections 32/34.
Claim tags used: SETTLED LAW / LIKELY INTERPRETATION / STATE-DEPENDENT / AMBIGUOUS / SPECULATIVE.
1. EQUITY SHIELD SOLVENCY
Finding 1.1 — The reserve cannot fund the floor in the correlated tail, and is grossly overpriced outside it. (CRITICAL)
The issue. Equity Shield charges 50–75 bps/yr (spec Section 17 says on total allocation; the website says on deployed capital — the two bases differ by ~58% and the discrepancy is itself a disclosure defect) to fund a reserve that "guarantees" 1.0x of deployed capital portfolio-wide. Housing drawdowns are a common-shock risk: a national correction pauses disbursements and erases appreciation across the entire book simultaneously. A guarantee against a common shock cannot be funded by a small pro-rata fee; it requires capital sized to the tail.
What loss does the floor actually cover? The base structure already returns 1.0x from escrow at exit (Section 29, Priority 3), senior even to commissions. Stress modeling shows the only material loss mechanism for the capital return is foreclosure wipeout (junior interest extinguished; surplus recoveries in practice ≈ 0) plus severe-market short sales. Distressed sales almost never breach the floor because of the 20%+ equity cushion and low entry LTV: on a −45% sand-state replay, the oldest deal's escrow after 7% selling costs and mortgage payoff still nets ~$158K against a $126K capital claim — fully recovered. So floor loss ≈ (foreclosure rate) × (deployed capital) at ~100% severity.
The arithmetic (seasoned $100M book, 794 deals, originated over 24 months, fully deployed at the market peak, 2007–2012 Case-Shiller-style path to −26.5% national / −45% sand-state trough at year 5 post-peak):
| Scenario at trough | Floor shortfall | Reserve @75bps deployed ($5.69M) | Reserve @75bps allocation ($9.0M) |
|---|---|---|---|
| 3% foreclosure + 7% distressed sale | $3.0M | 1.90x | 3.00x |
| 6% foreclosure + 12% distressed sale | $6.0M | 0.95x | 1.50x |
| 12% foreclosure + 20% distressed sale (sand-state severity, concentrated book) | $12.0M | 0.47x | 0.75x |
Note Section 17 explicitly targets "a single market or a concentrated set of markets" for $100M+ allocations — i.e., the concentrated bottom row is the design case, and there the reserve covers less than half the shortfall even after 12 years of maximum fee accrual.
The other side of the same coin: expected floor loss in a normal decade on prime, low-LTV, deep-rate-lock borrowers (cumulative foreclosure ~0.5–1%) is ~$0.5–1M, against $5.7–9.0M of fees collected. The fee is 6–18x expected loss, and the website states "if markets are fine, excess reserve is retained by noteworth." So the product is simultaneously (a) massively overpriced against the loss it actually covers in all normal states, and (b) underfunded by 2–4x in the correlated tail where it is needed. That is the signature of a guarantee priced as marketing, not actuarially. Severity: critical.
Finding 1.2 — Timing mismatch: an early-cycle correction leaves the reserve at ~$0.1–2.25M, but reveals the deeper defect — the Shield insures the wrong loss. (CRITICAL)
If the correction begins at month 12 (book 0–2 years old), the market-pause mechanism freezes disbursements at ~$6.0M of the $100M committed; reserve at month 36 is $0.10M (deployed basis) to $2.25M (allocation basis). But forced sale exits at −18.5% still recover the small deployed amounts from escrow (net ~$146K vs ~$7.5K average deployed) — the floor is not breached. The investor's actual early-cycle loss is three-plus years of 0% return on frozen capital and open-ended duration extension — and a 1.0x top-up guarantee pays $0 against that by construction. The website's own economics panel proves this: bear case without the Shield is "1.02x," i.e., the floor is not breached in their own bear case, so the Shield's bear-case payout is zero; the "capital at risk" it claims to protect is labeled, in their own table, "opportunity cost" — a loss category a 1.0x floor never compensates. The investor pays ~70 bps of IRR (their own numbers: 8.5% → 7.8%) for protection that pays nothing in the seller's own downside scenario. This is a fee for illusory protection. Severity: critical.
Finding 1.3 — No pool size makes this guarantee credible; only capital does. (MAJOR)
The actuarial prerequisite for a fee-funded guarantee is loss independence, so the law of large numbers shrinks relative variance. Housing tail losses are driven by a common factor (national HPI, rates, unemployment); in the tail, inter-deal correlation → 1, and portfolio loss variance does not diminish with N. There is no minimum pool size at which a portfolio-wide 1.0x floor on same-asset-class, same-cycle exposures becomes statistically self-funding. The only actuarially sound structures are: (i) capital sized to a 1-in-200-type tail (≈ 6–12% of deployed, i.e., $6–12M day-one capital on a $100M book — which the fee cannot accumulate before ~year 10+), (ii) genuine external risk transfer (reinsurance/ILS), or (iii) no guarantee. Severity: major.
Finding 1.4 — Conflicted, unvalidated, counterparty-less reserve. (CRITICAL)
- The spec never identifies the guarantee counterparty. The website implies Noteworth ("excess reserve is retained by noteworth"; "recurring revenue stream for noteworth"). There is no trust, no segregation, no custodian, no actuarial opinion, no capital requirement, and no stated liability of anyone if the reserve is short.
- The economics create a textbook adverse incentive: Noteworth profits dollar-for-dollar from under-reserving (excess is its revenue) and bears no stated obligation for shortfall. No independent party validates reserve adequacy. In licensed insurance this is what appointed actuaries, statutory reserve opinions, and RBC exist to prevent.
- Allocator's recovery if the floor is called and the reserve is short: an unsecured contract claim against a startup platform whose fee income has just collapsed in the same scenario. Recovery ≈ 0 exactly when the guarantee binds (wrong-way counterparty risk).
- The fee is "paid by the investor from fund operating expenses" (Section 17): the insured funds its own guarantee and forfeits the reserve's upside. Economically this is the investors pre-paying their own tail losses into a pot owned by someone else.
Severity: critical. As designed, Equity Shield should not be sold in any form.
2. INSURANCE REGULATORY CLASSIFICATION
Finding 2.1 — Equity Shield as marketed is very likely "doing an insurance business" without a license. (CRITICAL)
- A contract under which one party (Noteworth or an affiliate), for a fee, agrees to confer a benefit of pecuniary value upon another party upon the happening of a fortuitous event (portfolio return < 1.0x) in which the beneficiary has a material interest, is the definition of an insurance contract under NY Ins. Law § 1101(a). The statutory definition itself: SETTLED LAW. Its application to Equity Shield — a third-party promise, for an ongoing per-annum fee, to absorb investment loss from housing-market events — LIKELY INTERPRETATION, and the marketing makes it worse: the spec calls it "portfolio insurance" (Section 17) and the website calls it a "structured downside guarantee" the investor "buys." Regulators and courts weigh substance and marketing language; calling it insurance in your own materials largely forfeits the argument that it isn't. LIKELY INTERPRETATION.
- If it is insurance, it is financial guaranty insurance (indemnifying financial loss on an investment) which in New York may be written only by a licensed monoline financial guaranty insurer under Article 69 (§§ 6901–6909); multiline and unlicensed entities are barred. NY monoline requirement: SETTLED LAW. Whether other states' surety/financial-guaranty statutes reach it: STATE-DEPENDENT (California Ins. Code § 22's expansive definition and similar provisions in most states make broad reach likely; a minority of states have narrower guaranty carve-outs).
- Doing an insurance business without a license: in NY, § 1102 prohibition and penalties (civil penalties; § 1102(a) violations can be criminal), and unlicensed insurers' contracts are voidable/unenforceable at the option of the insured in many states while the "insurer" can still be held to pay — the worst of both worlds. Existence of penalties: SETTLED LAW; magnitude and enforcement posture: STATE-DEPENDENT. Liability would reach the platform and plausibly its control persons; producers/promoters of unauthorized insurance are separately sanctionable. LIKELY INTERPRETATION.
Finding 2.2 — The fund-level-reserve restructuring escape exists but the current design forfeits it. (MAJOR)
A true fund-internal, first-loss reserve — owned by the fund, held in trust for the LPs, with excess returned to LPs, no third-party promisor, and disclosed as a waterfall feature rather than a "guarantee" — is generally treated as capital structure, not insurance (analogy: subordinated tranches, sponsor first-loss). LIKELY INTERPRETATION. But the shipped design defeats the escape on every element: excess retained by Noteworth (third party takes the premium-like upside), fee expressed in basis points per annum (premium-like), marketed with "guarantee/insurance/protected" language, sold to obtain "favorable regulatory capital treatment" (Section 17) — a purpose that only works if it functions as a real guarantee, which is exactly what makes it insurance. You cannot simultaneously tell insurers "this transfers your risk for capital purposes" and tell insurance regulators "no risk is transferred for licensing purposes." AMBIGUOUS at best; the two claims are mutually destructive.
Finding 2.3 — Equity Flow's own "not an insurance product" positioning (Section 25) is defensible; Equity Shield contaminates it. (MINOR)
The core homeowner contract (contingent equity forward, no guaranteed payments) is unlikely to be insurance: LIKELY INTERPRETATION. But bundling an unlicensed guarantee product into the platform invites the insurance regulator into the entire structure, and NAIC/state HEI scrutiny is already rising (HEI-specific statutes emerging, e.g., Maryland's 2023 treatment of HEIs as lending — STATE-DEPENDENT).
3. TIER 2 CAPITAL TREATMENT (NAIC / RBC)
Finding 3.1 — RBC charge destroys the yield argument. (CRITICAL)
An unrated, novel, open-ended contingent claim on residential appreciation is not a bond (no maturity, no fixed payments) and cannot be a Schedule D asset. It lands on Schedule BA as an "other" invested asset with equity-like treatment: life RBC C-1 factor ≈ 30% pre-tax for unrated/equity-character BA assets. LIKELY INTERPRETATION (exact factor depends on characterization; SVO would drive it). If characterized as a residual/first-loss interest in a structured vehicle, the 2024 NAIC residual-tranche factor of 45% applies. SETTLED that the 45% residual factor was adopted; AMBIGUOUS whether these positions get swept into it.
Capital drag arithmetic (10% cost of capital):
| Asset | Gross yield | RBC | Capital drag | Net |
|---|---|---|---|---|
| NAIC-1 corporate | 5.5% | ~0.4% | 0.0% | 5.5% |
| A-rated structured tranche | 6.2% | ~1.3% | 0.1% | 6.1% |
| Schedule A real estate (2021 factors, ~11%) | 7.5% | 11% | 1.1% | 6.4% |
| Ride-along position (Sch BA, 30%) | 9.5% | 30% | 3.0% | 6.5% |
| Same, residual treatment (45%) | 9.5% | 45% | 4.5% | 5.0% |
Even granting a 9.5% gross expected return (itself generous — see Section 4), the position nets roughly what an A-rated bond nets, with none of the liquidity, rating, or track record — and under residual treatment it nets less than an NAIC-1 corporate. The premise "insurers want this paper" fails on RBC arithmetic alone. Severity: critical.
Finding 3.2 — A rated wrapper is a precondition, and it cannot exist by year 7–8 of fund one. (MAJOR)
The spec concedes it: "Requirement: rated structure or actuarial track record (doesn't exist yet)" (Section 22). Precedent exists for lump-sum, forced-settlement HEI securitizations (Point/Redwood and Unlock deals rated by DBRS Morningstar from 2021 on), but those structures have contractual maturities and 10–30-year settlement deadlines. A rating agency looking at ride-along paper would demand: multiple vintages of exit data through at least one adverse rate cycle, legal opinions on every kill-risk (tax recharacterization of monthly payments, bankruptcy characterization, state HEI statutes, enforceability of the ratchet), servicing and backup-servicing infrastructure, a trustee structure, and resolution of the no-maturity problem (rated notes need a legal final; an asset with no outside settlement date forces either a very long legal final or a liquidation mechanism the product philosophically forbids). Realistic timeline: 5–10 years after meaningful exit data begins accruing — i.e., years 12–17, not year 7–8. The originator exit as scheduled depends on infrastructure that cannot exist on that schedule. Severity: major.
Finding 3.3 — Admissibility and fiduciary gates. (MAJOR)
- State investment statutes: positions would have to fit "leeway"/basket provisions (e.g., NY Ins. Law §§ 1404–1405 basket limits of a few percent of admitted assets) rather than any enumerated class. Basket capacity is scarce and competed for by private equity and other alternatives. STATE-DEPENDENT.
- Pension/endowment consultants require typically 3+ years of audited, GIPS-comparable track record, institutional AUM thresholds, and an operating sponsor that survives diligence. A first-time fund of $25–50M with one vintage and no realized exits fails every screen. The spec itself admits pensions "won't touch first-time fund in unproven structure" (Section 22) — correct, and it means Tier 2 at year 7–8 reduces to insurers, who fail on 3.1–3.2. LIKELY INTERPRETATION (industry practice, not law).
4. SECONDARY MARKET THESIS STRESS TEST
Finding 4.1 — The mortality-vs-mobility category error invalidates the core analogy. (CRITICAL)
Life settlements, longevity swaps, and HECM tails price off mortality: actuarially independent of financial markets, stationary, supported by a century of credible tables. The ride-along exit prices off mobility: sale/refinance behavior that is (i) regime-dependent, (ii) adversely correlated with position value — mobility collapses precisely when rates rise and prices fall (the 2022–2026 lock-in phenomenon is the product's own founding observation, Section 2), and (iii) deliberately minimized by the product's own underwriting (the lock-in score selects for low mobility, Section 23) and further suppressed by the contract itself (selling costs the homeowner 25–43% of appreciation — the agreement adds a second lock-in on top of the rate lock). "Actuarially modelable" does not survive this: there are no credible mobility tables conditional on rate regimes, and the one input that is genuinely modelable (mortality) produces exits 25–40 years out, where discounting makes the claim nearly worthless. Severity: critical.
Hold-duration model (enrollment age 52, Gompertz mortality + mobility hazard; conditional on surviving the 7-year term):
| Mobility assumption | Mean exit (yr) | Median | P(hold > 20 yr) | P(hold > 25 yr) |
|---|---|---|---|---|
| Unselected US mover rate (~8%/yr) | 17.1 | 14.5 | 30% | 18% |
| Lock-in-selected book (~3.5%/yr) | 23.7 | 22.5 | 56% | 42% |
| Deep lock-in + contract disincentive (~2%/yr) | 27.5 | 27.5 | 69% | 56% |
The fund underwrites against its own exit trigger: on its selected book, the majority of ride-along positions are still outstanding at year 20+.
Finding 4.2 — Wrong-way duration: the exit extends exactly when the position is worth least. (MAJOR)
Position bought at year 7 expecting a year-12 exit at 3% HPA has PV ≈ $128K (r = 10%). If the market goes flat post-purchase and the exit slips (which is the same event — flat markets suppress mobility): exit year 17 → PV $84K; exit year 22 → PV $56K. Payoff grows $13K while PV falls $72K. This negative convexity is the opposite of the positive-convexity mortality assets in the analogy set, and it is why "duration is the feature" (Section 8) is backwards for a discounted buyer: duration is the risk, and it arrives correlated with low HPA.
Finding 4.3 — The year-7–8 originator exit only works in the marketing deck. (CRITICAL)
The claimed ~11% originator IRR (Sections 18–19) assumes settlement at year 7 at full intrinsic value ($189,650 at 5% HPA). But a position that reaches year 7 unsold is precisely one that did not settle; the originator sells an unsettled, open-ended claim. Required price for the marketing IRR vs. what a rational Tier 2 buyer pays:
- Price required at month 84 for 11% originator IRR: $178,270 (1.41x deployed capital); for 8%: $161K (1.28x); for 6%: $150K (1.19x).
- Probability-weighted fair value at year 7 to a 10%-return buyer on the lock-in-selected exit distribution, with the ratchet: $86,866 (0.69x deployed) at 3% HPA; $138,664 (1.10x) at 5% HPA. Even an 8%-return buyer at 5% HPA pays only $182K — and Finding 3.1 shows no insurer accepts 8% gross on a 30%-RBC unrated asset.
- Originator IRR at various novelty discounts to fair value (g = 3%, r = 10%): 0% discount → −10.4%; 30% → −21.5%. At g = 5%: 0% discount → +3.6%; 15% → −1.2%; 30% → −7.0%.
A rational first buyer of an unproven asset class does not pay fair value; life settlements traded at 20–40% discounts to actuarial EV for their first decade. At any plausible discount, the year-7–8 sale is not "crystallizing IRR" — it is realizing a loss. The spec's own Critical Risk Acknowledgment (Section 8) concedes that if the secondary market doesn't develop, the fund must hold to organic exit "requiring a much longer fund life and different LP expectations" — per the duration model above, that means a 20–30-year fund life. The acknowledgment is honest; the fund economics built on top of it are not. Severity: critical.
Finding 4.4 — Seasoning data from fund one is statistically incapable of supporting Tier 2 pricing. (MAJOR)
A $25–50M first fund is ~200–400 positions, one vintage, concentrated markets, one rate regime. At ~4%/yr exit rates that generates ~10–16 exit observations per year. Classical credibility (limited-fluctuation full-credibility standard ≈ 1,082 claims for frequency alone, more for severity) puts full credibility decades away — and the deeper problem is non-stationarity: exit behavior observed in one rate regime does not predict the next (the entire product exists because 2019-2021 mobility assumptions broke in 2022). No amount of first-fund seasoning produces a priceable table; "7+ years of seasoning data" (Section 8) is 7 years of data from a distribution that will have shifted by the time it is used. Severity: major.
5. RATCHET PRICING (+1 pp/yr, cap 50%)
Finding 5.1 — The ratchet does roughly a quarter of the work duration requires; it is real but decisively insufficient. (MAJOR)
Priced at year 7 (V₀ = $625K, capital $126K, g = 3%, r = 10%), PV of the claim by exit year, with and without the ratchet:
| Exit yr | Share | Payoff | PV w/ ratchet | PV flat 25% | Ratchet adds |
|---|---|---|---|---|---|
| 8 | 26% | $169K | $153,955 | $152,439 | +1.0% |
| 12 | 30% | $206K | $127,804 | $119,543 | +6.9% |
| 15 | 33% | $241K | $112,466 | $99,451 | +13.1% |
| 20 | 38% | $317K | $91,954 | $72,982 | +26.0% |
| 25 | 43% | $420K | $75,532 | $53,401 | +41.4% |
| 30 | 48% | $554K | $61,890 | $38,977 | +58.8% |
On the probability-weighted book (3.5% mobility), the ratchet adds 18–38% to expected PV depending on (g, r) — so it is not purely cosmetic. But the test of "correctly compensates for open-ended duration" is whether PV is flat in exit timing, and it fails by a wide margin: the breakeven ratchet that would hold PV constant year-over-year at g = 3%, r = 10% is 5.6 pp/yr at year 8, 4.3 pp/yr at year 12, ~3.5 pp/yr at years 20–30 — versus the 1.0 pp/yr offered. Even with the ratchet, each additional year of hold destroys ~8–10% of position PV. And the 50% hard cap removes all further compensation from year 32 onward — exactly the mortality-only deep tail where duration compensation is most needed. Structurally the ratchet cannot be fixed by turning the dial: at the ~4–5 pp/yr that duration actually requires, the homeowner hits the 50% cap by year 12–13, which merely relocates the problem. The honest conclusion: open-ended duration at a 10% discount rate cannot be compensated on the appreciation-share axis at any consumer-defensible ratchet speed. The ratchet's real function is narrative — it lets the deck claim duration is "compensated" — and Tier 2 buyers who can discount will price through it (Model C: even with the ratchet, EPV = 0.69x deployed at conservative growth). Severity: major (for investor economics; note the retail-facing framing "increases modestly to compensate" is accurate about modestly and wrong about compensate).
Cross-cutting observations
- The two insurance claims are load-bearing and both fail. Equity Shield exists to deliver "favorable regulatory capital treatment" to Tier 2 buyers (Sections 17, 22); Tier 2 buyers exist to give Tier 1 its year-7–8 exit; the year-7–8 exit is what makes the PE-style 7–10-year fund life and the ~11% IRR marketable. Remove the two insurance claims and the true product is a 20–30-year-duration, unrated, negative-convexity asset yielding materially less than marketed, wrapped in a guarantee that is either illusory or unlicensed insurance.
- The spec is more honest than the deck. Section 8's risk acknowledgment and Section 16's pause caveat are creditable. The website's Equity Shield panel ("1.0x guaranteed," "excess retained by noteworth," bear case 1.02x) is where the insurance exposure concentrates.
- What a fixed version looks like (necessary conditions, not sufficient): (i) kill Equity Shield as a fee-for-guarantee product; replace with an LP-owned first-loss reserve or genuine reinsurance/ILS, sized to a modeled 1-in-200 tail (≈$6–12M day-one on $100M), with an independent actuarial opinion and zero Noteworth retention of excess; (ii) re-underwrite the fund as hold-to-organic-exit (evergreen or 25-year vehicle) and re-cut the marketed IRR off the exit-timing distribution above, not off a year-7 settlement fiction; (iii) treat any Tier 2 sale as upside optionality contingent on a rated wrapper in the mid-2030s, never as base case; (iv) either steepen duration compensation via a mechanism that can actually scale (e.g., a small fixed-dollar accrual senior in the waterfall — which reintroduces the debt-characterization problem, and that trade-off must be faced explicitly) or price origination assuming the ratchet under-compensates.
VERDICT: FIX-BEFORE-LAUNCH (with two embedded kills)
- Equity Shield as currently structured: KILL. It is 2–4x underfunded against the correlated tail it purports to cover (reserve coverage 0.47x–0.95x in realistic stress), pays $0 in its own marketed bear case while charging 6–18x expected loss with the excess retained by the platform, has no identified counterparty, no validation, and no capital — and selling it for a fee is very likely doing unlicensed financial-guaranty insurance business (LIKELY INTERPRETATION; NY Article 69 monoline requirement SETTLED LAW). No parameter change rescues it; it must be rebuilt as LP-owned first-loss capital or real risk transfer.
- The year-7–8 Tier 2 exit as a fund-economics assumption: KILL. Rational Tier 2 fair value is 0.69x–1.10x deployed capital versus the 1.41x the marketing IRR requires; RBC drag (~30% Schedule BA, LIKELY INTERPRETATION) nets the buyer bond-like returns for venture-like risk; the rated wrapper and credible seasoning data cannot exist on the stated timeline. Originator IRR at any plausible transaction price is negative to low single digits.
- The core Equity Flow contract (monthly disbursement, ride-along, escrow waterfall) does not, from this domain's vantage, require killing: its escrow-first 1.0x recovery is genuinely robust in sale exits even at −45% (equity cushion holds), and its "not insurance" positioning for the homeowner contract is defensible (LIKELY INTERPRETATION). It can launch only after: Equity Shield is withdrawn/rebuilt, fund documents and all marketing re-based to hold-to-maturity economics with the true duration distribution (mean exit year ~24 on the selected book), and the ratchet either re-priced or honestly disclosed as partial compensation. Those are structural, not cosmetic, changes — hence fix-before-launch rather than acceptable, and kill rather than fix for the two insurance claims the pitch currently stands on.