Engagement has a shape
If you post on LinkedIn as a founder, your engagement graph looks chaotic: a hundred reactions, then nine, then four thousand, then twelve. It feels random. It mostly isn't - it has a shape, and the shape is knowable.
We assembled a clean panel of 477 YC W24 founder accounts - roughly 11,000 original posts from January 2025 through June 2026 - deduplicated at the person level, with scraper artifacts merged and validated.11Data hygiene mattered more than modeling: about 880 duplicate person nodes were merged, 46 stale school records removed, about 300 numeric IDs re-resolved, and numeric-ID validation added at parse time. Then we asked four questions.
1. What does the distribution of engagement actually look like?
Lognormal, everywhere. Take the log of reactions per post and the chaos turns into a well-behaved bell - with one deviation we'll get to.
Founder engagement is lognormal: most posts cluster low, a fat tail goes viral
Most posts cluster below an account's average, while a long right tail contains rare viral posts. Half of posts fall below the median and a typical post swings two times around trend.
Illustrative shape using study parameters · 477 YC W24 founder accounts · about 11,000 posts · Jan 2025–Jun 2026
Most posts cluster below the account's average; a long right tail carries the outliers. Two practical consequences:
- Half of your posts will land below your median. That is arithmetic, not failure. Founders quit in exactly this valley, mistaking the lower half of a lognormal for decline.
- The viral tail is fatter than lognormal predicts. Outliers are more common than the textbook curve suggests - which is also why mean-based dashboards mislead: a single post can drag the average anywhere. Medians and trimmed means are the honest summaries.22An apparent non-B2B engagement advantage in an early cut came almost entirely from one outlier company — a reminder to use robust summaries in fat-tailed data.
The most surprising regularity: every account shares essentially the same within-account distribution shape. Accounts differ by level - how high the whole curve sits - not by shape. There is no special founder whose posts are consistently viral; there are founders whose entire distribution sits 10x higher.
2. Is engagement predictable?
The single most predictable thing about an account is its level. Your next month looks like your last few months, shifted slightly by trend.
We ran a time-series cross-validation on continuous posters: predict each account's future engagement level from its history.
Momentum is real but weak: trust your level, and only 30% of your trend
Using past level plus 30 percent of trend reduces forecast error by 21 percent. Extrapolating the full raw trend increases error by 70 percent.
YC W24 founder study · 477 accounts · about 11,000 posts · account-level engagement backtest · Jan 2025–Jun 2026
The winning forecast is past level plus about 30% of the fitted trend - it beats a level-only forecast by ~21% in mean squared error. Extrapolating the raw trend is ~70% worse than doing nothing. Momentum is real, but weak: if your engagement doubled last quarter, expect to keep about a third of that slope, not all of it.
Single posts remain genuinely hard to predict - a typical post swings 2x around trend in either direction. The system is forecastable at the account level and noisy at the post level, which is exactly the opposite of how most people read their own analytics.
3. Do accounts drift?
Yes - individually, a lot; collectively, not at all. About half of continuous posters (38 of 77) show statistically significant drift over the study window, some moving 3x per year in either direction. The platform-wide average stays flat. LinkedIn isn't "dying" or "booming" for founders in this cohort; individual accounts are rising and sinking through a stable market.33Drift was detected by fitting within-account trends on continuous posters, with significance established by permutation inference.
4. Is the weekend penalty real?
Everyone believes weekends are dead. The naive comparison agrees: Saturday posts earn about -13% vs weekdays.
The weekend penalty is half as big as it looks — and maybe not real
A naive comparison estimates Saturday engagement at 13 percent below weekdays. Matching the same account and month reduces the estimated penalty to 7 percent and borderline significance.
YC W24 founder study · 477 accounts · about 11,000 posts · matched account-by-month design · about 500 weekend posts
But weekend posters are not average posters, and weekend posts are not average posts. In a matched account-by-month design - the same account, the same month, weekend vs weekday - the penalty shrinks to about -7% and borderline statistical significance. With only ~500 weekend posts in the cohort, the study cannot resolve effects smaller than ~10%, so the honest conclusion is: if a weekend penalty exists, it is small. Day-of-week is a control variable, not a strategy.44The -13% naive estimate conflates which accounts post on weekends, which months they do it in, and the weekend itself. Matching removes the first two.
What this means if you're a founder
- Judge months, not posts. A lognormal generates cold streaks by design. Your operating metric should be a monthly aggregate against your own baseline.
- Your level is your asset. The reliable way to raise every future post is to raise the account's level - cadence, voice, and audience quality - not to chase the tail.
- Discount your own momentum by 70%. Growth persists at about 30% strength.
- Post when you have something to say, including Saturdays.
Next in this series: extending the panel to S24 and F24 cohorts (the fix for underpowered calendar effects), and between-account questions - verticals, activity tiers, company status - using the same matched designs and permutation inference.