CANONICAL LABS · POWER LAW OUTLIER LAB

A venture fund simulator.

Describe a fund — its size, the number of investments, the failure rate, the shape of the right tail — and we run 10,000 simulated versions of that fund. The histogram shows the full distribution of plausible outcomes, not just the headline number on a pitch deck.

If you're an LP Plug in a manager's stated strategy (N, ownership, reserves). Is their projected 3x in the top decile of plausible outcomes, or the median? Use this before the IC meeting.

If you're a GP Show your LPs why your N matters, why concentration discipline matters, and why the right tail is your real thesis. Drop a screenshot in your deck.

If you're curious Click any scenario below. Each one isolates a single counter-intuitive thing the math does that most fund pitches gloss over.

Start with a preset:

Fund

Fund size i Fund size. Total dollars committed by LPs. Drives every dollar-denominated number downstream — at the same N and reserves, a bigger fund means bigger checks. Typical seed funds: $30–100M; Series A: $150–400M; growth: $500M+. $50M

Initial investments (N) i Number of first checks. How many companies you initially invest in. More names = lower variance and a thinner right tail (you'll probably miss the outlier). Concentrated seed funds: 15–30. Index-style funds: 50+. This is the single most consequential portfolio-construction choice. 25

Reserves % i Reserves. Share of the fund held back to follow on in later rounds. Higher reserves = smaller initial checks but more ammunition to defend ownership in winners. Industry default is 30–50%. 40%

Ownership multiplier i Ownership multiplier. Scales every company's outcome multiple by this factor. 2.0x = you owned twice as much of every company at exit — what you get from bigger initial checks, lower dilution, or smarter pro-rata defense. Linear effect on fund TVPI. 1.0x

Follow-on strategy i Where reserve dollars go.

Pro-rata: reserves split evenly across the portfolio.

Super pro-rata: reserves loaded into your realized winners (top 30% by multiple).

None: reserves returned to LPs at 1.0x, no follow-on.

The concentration choice has a bigger impact on fund TVPI than most LPs realize.

Initial check ≈ $1.2M

Outcome shape

Loss rate i Loss rate. Share of companies that return $0 (total loss). Real empirical ranges: seed 45–65%, Series A 25–40%, growth 10–20%. Higher loss rate doesn't move expected TVPI linearly — the lost dollars come out of the bottom of the distribution, not the top. 65%

Tail thickness (α) i Tail thickness. The shape parameter of the Pareto distribution. Lower α = fatter tail = the rare 100x outlier matters much more. 1.1–1.3 = seed-style fat tail; 1.5–1.8 = Series A; 2.0+ = growth. The most consequential slider in this lab — try moving it just 0.2 and watch the histogram redraw. 1.50

Tail cap i Tail cap. Hard maximum outcome multiple for any single company. Sets a ceiling on the right tail to keep the math finite. 100–500x is reasonable for seed; 30x for growth. Real outliers like early Uber or Coinbase landed well past 1000x — bump this up to see what that does. 100x

E[per-co multiple] = 3.1x

Fund economics

Mgmt fee i Management fee. Annual fee as % of committed capital, charged for the fund life. Standard 2% × 10 years = 20% drag on gross returns. Lower fees pass through directly to LP net TVPI. 2.0%

Carry i Carried interest. GP's share of profits above 1.0x of committed capital. Standard is 20%. European waterfall is assumed — carry is only paid after LPs receive their full commitment back. 20%

Fund life (yrs) i Fund life. Years over which management fees accrue. Doesn't model exit timing — TVPI is reported as a horizon-free multiple. A 12-year fund pays 24% in fees at 2%; an 8-year fund pays 16%. 10

Simulation

Trials i Monte Carlo trials. Number of simulated funds. 10,000 is enough for stable median/quantile estimates; crank to 100,000 for sharper p99. More trials = more confidence, slower simulation. 10,000

E[TVPI] i Expected TVPI. The mean fund TVPI across all 10,000 simulated runs. Dragged upward by right-tail outliers — the rare 20x outcomes inflate the average even though most funds don't hit them. If a pitch deck quotes one number, it's usually this. It's the most flattering one.

—

p50 TVPI i Median TVPI. Half of the 10,000 simulated funds returned more, half returned less. The "typical" outcome. The gap between p50 and E[TVPI] is the power law working — in a normal distribution they'd be close. Watch them drift apart as you fatten the tail.

—

P(>3x) i Probability of returning 3x+. The fraction of simulated funds that returned 3.0x net to LPs or more. The "good fund" threshold. For seed-style strategies this lands around 10–20%; for growth it's typically <5%.

—

P(>5x) i Probability of returning 5x+. The "outlier fund" threshold. Single-digit-percent territory even for great strategies — these are the funds that build a firm's reputation. A strategy that can't produce a meaningfully positive P(>5x) is not selling venture risk.

—

P(loss) i Probability of losing capital. Fraction of simulated funds that returned less than 1.0x to LPs (failed to return committed capital). Always non-trivial in venture — 25–35% is typical for a seed fund. Don't pretend this number is zero.

—

Sim time i Simulation time. How long the Monte Carlo run took, in milliseconds. Mostly a debug indicator — if it climbs past a second, lower the trials slider.

—

Fund TVPI — distribution across 10,000 simulated funds

Each bar = the % of simulated funds that landed in that TVPI bucket. Quantile markers (p10–p90) overlaid in white.

<1x (loss) 1–3x 3–10x 10x+

Where the return came from

Averaged across all simulated funds. The fewer winners drive the return, the more the math is power-law shaped.

—

Top 1 company i Top 1 contribution. Across all simulations, on average, what % of the fund's total dollar return came from the single largest winner. The signature power-law number. In seed strategies it's often 40–60%; in growth it's closer to 20–30%.

share of total fund return

—

Top 3 i Top 3 contribution. Same idea as Top 1, but for the three largest winners combined. If this exceeds 80%, the fund is fundamentally power-law dominated — most of your TVPI comes from three calls.

share of total fund return

—

Top 10 i Top 10 contribution. Share of fund return from the top 10 winners combined. For most seed strategies this is north of 90% — meaning the bottom 15+ investments contribute almost nothing.

share of total fund return

—

                                            TVPI without your top winner
                                            
                                                i
                                                
                                                    TVPI minus the top hit. What the fund's net TVPI would have been if the single largest winner had returned $0 instead.
                                                    The most underappreciated number in venture — most "good" funds become 1.0x funds once you remove their best deal.
                                                    This is the LP question a GP can't dodge.
                                                
delete the biggest hit — what's left?

Explore

Six scenarios. Each loads a preset and a short argument. Disagree freely.

Methodology — what's actually being simulated +

The model

Each company independently draws an outcome multiple from a mixture distribution: with probability loss rate it returns 0; otherwise it draws from a Pareto distribution with shape α, lower bound 1, truncated at tail cap. Pareto is the standard parametric form for venture outcomes — it produces the long tail empirical fund data shows.

Sampling

Survivor sampling: u ~ Uniform(0, U_max) where U_max = 1 − (1 / tailCap)^α, then x = 1 / (1 − u)^(1/α). Truncating at tail cap bounds the largest possible outcome. With α near 1 the tail is fat and a single 100x outcome dominates; with α near 3 the tail collapses and the distribution looks almost normal.

Fund economics

Initial check = (1 − reserves) × fundSize / N. Follow-on capital is split across the portfolio according to your selected strategy (pro-rata = equal split; super pro-rata = top 30% of winners by realized multiple; none = reserves returned to LPs at 1.0x). Gross return is summed across all positions and scaled by the ownership multiplier. We subtract management fees over the fund life and apply carry on profits above 1.0x of committed capital. The reported TVPI is net to LP.

Calibration

Seed preset (α=1.5, loss rate 65%, cap 100x) is consistent with public empirical work — Correlation Ventures, Kauffman, AngelList. Series A and Growth presets shift loss rate down and tail thickness up, matching the conventional view that later stages have lower variance.

Limitations (read these)

Company outcomes are assumed independent. They aren't — vintage years, sector concentration, and macro shocks correlate failures. Real fund TVPI variance is wider than this model shows.
Outcomes are net of dilution. The Pareto sample represents what your initial check became at exit. Cap-table mechanics are folded into the distribution, not modeled separately.
Time is collapsed. The simulation reports TVPI, not DPI, and treats the fund life as a single horizon. Time-value-of-money matters for IRR but not for the visceral point this lab is making.
This is a teaching tool. Don't use it to underwrite a manager. Use it to argue with their pitch deck.