Model notes & sources
Explorer is a Monte Carlo model of the shape of angel returns, built on Wiltbank 2007 data. Outcome multiples and holding periods are drawn from the distribution in Robert Wiltbank & Warren Boeker, Returns to Angel Investors in Groups (Kauffman Foundation / ACEF, 2007) — 1,137 exits from 539 group-affiliated angels: 52% of single deals returned under 1×, 7% over 10×, 2.6× overall in 3.5 years, with the top 10% of exits producing ~75% of all cash returned. The Low / All / High presets are the study's due-diligence cohorts; the two intermediate presets are blends. Failure rates, >10× shares and per-band holding periods come from the paper; interior band splits are read from its chart. The band editor reseeds from whichever preset you pick — probabilities are normalized to 100% at simulation time, and the within-band skew (draws lean toward the low end of each band) is inherited from the preset. Holding periods use the study's per-band averages (losses ~3 years, the biggest winners ~6 — "lemons ripen faster than plums"), varied ±40% per deal.
The chart shows each run's net cash position — money actually out the door versus cash actually returned — so you see the true J-curve: angel capital is locked and cash-negative for years while exits trickle in. Deals are simulated to their true exit dates, even past the chart's edge, and the outcome statistics use those fully realized figures. The dashed violet line adds back the mark-to-market value of positions still alive (cost basis interpolated toward each deal's eventual outcome) — the value locked up and waiting to land. The green line is an index-fund benchmark: the same contributed dollars deployed on the same cadence, compounding at the chosen market return and always liquid. IRRs are solved by bisection on each run's true dated cash-flow stream (sampled across runs for speed).
Because extracted cash does not compound on its own while the index does, the cash returned selector above the chart decides what distributions do next — three honest answers to the same question. To the market: each distribution goes straight into the investor's own index account (no new checks, no additional carry) and compounds at the market rate; the endpoint is a genuine terminal-wealth comparison against the green line, which holds the same contributed dollars. Offsets next rounds: returned cash funds future capital calls before any new money is called, and earns the market rate while it waits — the deals are identical, but the paths measure the account against out-of-pocket injections, the headline multiple is taken on that much smaller outlay, and the green line shrinks to match: the same out-of-pocket stream sent to the market instead. This view always ends below the to-market view by exactly the growth foregone on dollars recycled into calls — the price of needing less cash. Reinvested in deals: while the program is still deploying, whenever returned cash covers a full check it buys a new deal from the same outcome distribution (on the Network tab carried at the same rate, netted per deal rather than per vintage — slightly conservative for investors; no reserves) — cash out of pocket is unchanged, total invested grows, and the program compounds in itself only while it is investing. Once the deployment window closes, exits sweep into the investor's market account like the first mode, so growth decelerates from venture pace to market pace as positions mature. The orange paths show the cash position, as in every mode: pinned near the outlay floor while deploying — every return is immediately re-locked in a new deal — then surging back once deployment ends; the violet line marks the growing portfolio to market along the way. Terminal figures are the mark-to-market endpoint, essentially fully realized by the chart's edge. The KS-PME stat (Kaplan–Schoar) is mode-independent: it future-values all distributions and all contributions to the horizon at the index rate and takes the ratio — above 1.0 beats the market with timing handled correctly, and IRRs were always timing-fair.
Capital can be split between equity deals and revenue-based finance (the Allocation panel, set per tab). RBF is debt-like: the provider takes a share of revenue until repaid a capped multiple (1.3–2.0×), so returns are bounded — no power-law upside — and arrive as a monthly stream that begins one year after investment (a ramp before the revenue share is meaningful), giving a much shallower J-curve. The risk is business failure, drawn from a broad small-business failure curve (BLS style) graded by diligence cohort exactly like the equity presets — Low / L–M / All / M–H / High, where more diligence means fewer failures (All is the broad baseline, ~49% five-year failure; High is an underwritten/secured book, ~18%; Low is unscreened, ~72%). The revenue share runs continuously from year 1 until the company either fully repays the cap or fails — a failure simply stops the payments and writes off the rest (there is no separate liquidation recovery; your recovery is the stream you collected). The one-year ramp matters: at the default 1.5× cap even a High-diligence book yields only ~13% IRR, and reaching the reported 15–30% RBF fund IRRs needs both underwriting (High) and a cap toward the top of the 1.3–2.0× range (~19% at 1.7×, ~25–28% at 1.9–2.0×) — the cross-check the model is built for, and a sober view of unscreened revenue-based lending. RBF flows through the same recycle modes, carry and per-deal fee as equity; it has no follow-on reserves. See the RBF yield / IRR / blended-IRR readouts and the README's RBF section.
Follow-on reserves model a second check into companies still standing and raising ~2 years in: a company is "followable" with a probability that rises with its eventual outcome, so reserve dollars land mostly — but not only — on winners, and always enter at a 2.5× valuation step-up, capped at 3× the initial check per company. Unused reserve in the pooled tabs is returned to members as cash; the solo angel simply never calls it. This echoes the study's finding that follow-on investing by the same angel correlated with lower returns. Every angel is defined by three numbers: a yearly investing budget, a maximum they'll put into any one round, and a holdback reserved for follow-ons (personal on the solo tab; on the pooled tabs the chapter sets the reserve, and every chapter in the network uses it) — the solo angel's deal count is deployable budget ÷ max per round, with unspent remainders carried forward to the next year. On the Chapter and Network tabs those same members contribute annually and syndicate small SPVs: the group targets a goal round size sliced across its participants-per-round setting (never more than the 249 small-SPV cap, nor the membership), and each participant contributes min(their max per round, goal ÷ participants) — so more participants per round slice thinner, every member lands in more deals, and individual diversification improves with group scale (deal size = participants × slice; deals per year = deployable capital ÷ deal size). These same settings drive every chapter in the network. An optional external fee per deal (default $0, set independently on each tab, e.g. an SPV platform like Sydecar at ~$4,500/deal) is skimmed off each deal's capital before it reaches the company, so only check−fee earns the multiple while the investor still pays the full check — devastating on a tiny solo check, trivial syndicated into a large round; total fees are tracked per member and for the whole chapter or network. The chart follows a single representative member of the first chapter: they join as many rounds as their budget fills, hold a pro-rata share of the pooled reserve and its follow-ons, and pay their own fees and dues — beyond 249 members they hold different subsets of deals than their neighbors, so individual results spread wider than the group aggregate. The totals card and the tables report the whole chapter or network. On the Network tab the manager charges carry — netted against losses at the chapter-vintage level — plus management fees on contributions and annual member dues; all investor figures there are net of all three, while the index benchmark compounds contributions only. Simulations are seeded and deterministic for a given URL; Reseed draws a fresh world. With extreme settings the run count is automatically reduced and checks consolidated to keep the page responsive (the status chip says so). A teaching model, not investment advice.