How the careers are simulated

1. What one ball is

A ball is one career: a fresh graduate who takes a first job at a startup, a corporation or a consulting firm, then lives through thirty-five years of pay, promotions, layoffs, shutdowns, exits and the occasional decision to do something else. The plinko drops 1,000 balls per first job and shows the first thirty years. The sankey walks one ball at a time through the same engine with the reader making the choices.

The three cohorts are identical in every respect except the first job. Same pay curves, same promotion clocks, same odds of a layoff, the same chance at each milestone of moving to another track, going to business school or founding a company. That is the point: the plinko isolates the first decision, then lets the rest of a working life happen to it.

2. What happens in a year

Each year, in this order:

1. Leave school. A ball two years into an MBA graduates, lands one rung higher on the track it chose, and carries a pay lift.
2. Milestone choice. After years 3, 5, 10, 15, 20 and 30 the ball may stay, switch to another track, go to business school (through year 5), or found a company (years 5 to 15). In the plinko the choice is sampled from a track-specific propensity that decays with age; in the sankey the reader pins it.
3. The company rolls. Every company the ball still holds equity in raises its next round on schedule (diluting the holding and moving up a stage, if it graduates), shuts down, exits, or runs a tender. The employer rolls first.
4. Consequences. A shutdown or a layoff costs a few months of pay and a persistent scar; a startup employee then either joins another startup or goes corporate. An exit pays out and folds the ball into the acquirer. Otherwise the promotion clock ticks, and in consulting the counseled-out roll follows a missed promotion.
5. Pay. Starting pay for the persona, times the rung multiplier, times a persistent ability draw, times the scar or lift, times a small annual noise, plus any equity turned into cash that year.

The ball's x position in row t of the plinko is that year's realized pay on a log axis. An MBA pins it to the left wall for two years; a liquidity event throws it right. It settles on its thirty-year average.

3. Equity is a lottery ticket, not salary

The single most common mistake in a career comparison is to count a startup grant at its paper value as if it were pay. Here a grant is a ticket that pays only when its company exits, and then only this much:

vested  = 0 before the one-year cliff, then years / 4, capped at 1
diluted = grant % x (1 - dilution) for every round raised since the hire
common  = max(0, exit value - preference stack)
payout  = diluted x vested x common x (1 - strike / preferred price)

Leaving before the exit keeps the ticket only if the options are exercised, which Carta finds most people do not do. Kept tickets keep rolling, so a ball can leap right years after it moved on. Tenders at growth and PE stages let a fifth of a vested holding out early. Bootstrapped and PE-backed employers grant nothing to a new hire, which is the honest reading of the sources.

Where a naive model lies, and what this one does instead

• Carta's round, valuation and exit tables describe companies still on Carta. Shutdown rates come from cohort data (Carta graduation rates, PitchBook's Series C cohort, BLS survival), and the exit medians are shaded down for undisclosed acquihires.
• Per-round dilution of 12 to 22 percent shrinks a seed grant three to five times by the time a company exits.
• A 1x non-participating preference stack (96 percent of rounds) zeroes common at any sale below the money raised. Roughly half of seed-era exits leave employees with nothing.
• Consulting attrition is not failure: counseled-out consultants land one rung up in corporate jobs, which is where most of the value of consulting shows up.
• Without a persistent ability draw every distribution is far too narrow; with it, and no correlation to track, the spread inside a cohort is honest and the difference between cohorts is still the first job.

3b. How the three first jobs differ

One engine does not mean one set of odds. The tracks differ on every axis the sources support:

• Losing the job. A startup employee faces the company's shutdown hazard (16 percent a year at seed, 10 at Series A-B, 5 at growth) on top of a layoff hazard of 5 to 8 percent. A consultant who misses a promotion faces a 10 to 20 percent counseling-out roll each year, rising by rung. A corporate professional carries a 4.5 percent layoff hazard. The plinko's stat block counts how many of each thousand lost a job involuntarily.
• Cash. Consulting starts highest and its ladder is steepest; startups pay 25 percent below the corporate blend at seed and reach parity at growth stage; corporate sits between.
• Equity. Only startup and founder balls hold tickets; the corporate technical blend already counts big-tech stock as pay because it is liquid.
• Retirement money. Employer 401(k) contributions differ by track and vest with tenure, so short startup stints leave matches on the table. They count as realized pay and feed invested wealth (section 4).
• Life and cash strain scores are set per track and rung from weekly-hours data.

4. Wealth: retirement contributions and compounding

Pay is not wealth. Each year the ball's invested wealth grows at a real return, then adds a share of cash pay that rises with income (the Survey of Consumer Finances puts the fourth income quintile at 14 percent and the top at 27), about half of any windfall, and the whole of the employer's retirement contribution. Business school tuition is drawn down. The employer contribution is a percent of base pay set per track (and per stage for startups), and it vests with tenure, so a ball that leaves inside the vesting window keeps only part of it. Every one of those numbers is in the parameter table with its source.

retire  = base x employerRetirement[track] x min(1, tenure / matchVestYears)
pay     = base + equity cash + retire
wealth  = wealth x (1 + realReturn) + base x savingsRate(base) + equity cash x windfallSavingsRate + retire
          savingsRate steps 10% -> 14% -> 18% -> 22% at $80K, $150K and $250K of pay

Invested wealth at year 30 appears in the plinko's stat block and in the adventure pane, and the year-30 medians are benchmarked against the Survey of Consumer Finances in section 6.

5. Reference results

One thousand careers per cohort under a fixed seed, first thirty years. These are the numbers the browser must reproduce (section 8).

p10, median and p90 are of the thirty-year average of realized annual pay in 2026 dollars. Wealth is invested savings at year 30. A payday is $100K or more of equity cash in any year.

6. Benchmarks the model is held to

verify.py runs 4,000 careers per check and exits nonzero if any lands outside the range the sources support. All 14 pass in the shipped build.

7. Every parameter and where it came from

236 values. 43 are read directly off a source, 151 are estimates with the reasoning stated, and the rest are derived from other rows. The research notes behind them are in the repository under projects/career-paths/research/. All money is 2026 real dollars.

8. Two implementations, one model

The engine is written in Python and ported line for line to TypeScript so the browser can re-simulate when the reader changes the persona or the stage. Two implementations are two chances to be wrong, so both are driven from the same seeded generator (mulberry32, four lines in each language) and a parity test requires 500 careers to agree exactly on every event, track and rung, and to within one part in a billion on every dollar. A second gate re-runs the reference cohorts above through the shipped TypeScript and fails if any statistic drifts by more than half a percent. Both run before every publish.

9. Limitations

• Balls are assigned, people self-select. Somebody who takes a seed-stage job at 22 is not a random graduate. The ability draw widens every cohort but is uncorrelated with the first job; the real world may not be.
• “Corporate” is a blend. For the technical persona it mixes big-tech pay with the far larger population of engineers elsewhere; for the non-technical persona it is a Fortune 500 analyst track. A reader with a specific offer in hand should compare against that offer, not the blend.
• Pre-tax, real dollars, no geography. Nothing here is after tax, so invested wealth is overstated by roughly the tax bill on savings; a $200K year in San Francisco is not a $200K year in Columbus.
• Savings depend on pay alone. The savings rate steps up with income but does not fall in a layoff year or rise for a two-earner household; both would move the wealth figures.
• Survivor bias in the sources. Most private-market data comes from Carta, which sees companies that are alive, organized and venture-backed. The failure mass is spliced in from cohort data and that splice is an estimate.
• Consulting's exit premium is folklore. No study measures what a counseled-out consultant earns against a peer who never consulted; the one-rung bump is a judgment call.
• Milestone choices are the least sourced part. The propensity to switch, go to school or found is estimated from turnover and enrollment rates, and the “stay the course” toggle exists so a reader can remove that layer entirely.

10. Sources

• Levels.fyi End of Year Pay Report 2025
• NACE Class of 2025 salary projections
• MBB and Big 4 salary compilations (roadtooffer, preplounge, strategycase)
• Deming, NBER w31373 (college premium over the life cycle)
• Consulting up-or-out timelines (hackingthecaseinterview, casecoach)
• BLS JOLTS 2025 layoffs and discharges, scaled to tenured professionals
• BLS Displaced Worker Survey, Jan 2026
• Carta seed and Series A data via SaaStr / Chronograph
• VC M&A size distribution (92% under $50M by count)
• Carta, first-ten-hires equity (9,000+ grants)
• Carta State of Startup Compensation H2 2025
• Index Ventures OptionPlan benchmarks
• Carta State of Private Markets 2025 in review (tender offers)
• Carta option exercise rates, Q4 2024
• Carta median dilution per round (2,005 US software startups)
• PitchBook Series C cohort 2010-2015 (via Collective Liquidity)
• BLS Business Employment Dynamics, establishment survival
• Davis, Haltiwanger, Handley, Jarmin, Lerner, Miranda (AER 2014; NBER w26371)
• S&P Global, PE holding periods 2025 (median 6.0 years)
• GMAC tuition survey + Stanford/HBS class of 2025 employment reports
• Kruze Consulting startup CEO salary report 2026
• Carta seed and Series A data via SaaStr / Chronograph
• Correlation Ventures, distribution of VC investment outcomes
• Vanguard, How America Saves 2025
• BLS Employer Costs for Employee Compensation
• Firm 401(k) plan filings (McKinsey PSRP, Deloitte, others via MyPlanIQ) and firm salary guides
• BLS Employee Benefits Survey 2025 (retirement access by establishment size)
• Dynan, Skinner and Zeldes, Do the Rich Save More? (SCF saving rates by income)
• Damodaran, historical returns on stocks, bonds and bills 1928-2025
• Consulting and tech weekly-hours compilations

Stack: Python 3.12 for the reference engine and calibration, TypeScript for the browser port, canvas for the plinko, hand-laid SVG for the sankey, Amazon Bedrock (Claude Sonnet 4.6) for the “ask about this” box on the brief, grounded in the brief's own text.