McKinsey Sea Wolf: the ocean cleanup game, explained

The setup

Every microbe has three attributes rated 1–10 and one trait. Every site defines target ranges for the attributes, one desired trait, and one undesired trait. Your job at each site: end up with a treatment of exactly three microbes that fits those targets. You work each site through four one-way steps:

1. Build profile: pick the two characteristics that matter most for the site. This tunes what you'll see next; read the site brief carefully before committing.
2. Categorize microbes — for each candidate microbe: use it here, save it for the next site, or reject it. Locks on confirm. Saved microbes reappear at the next site, which makes this a portfolio decision, not just a local one.
3. Build prospect pool: you get six prospects, then add four more by choosing 1 of 3 across four rounds, ending with ten.
4. Create treatment: choose exactly three from the pool. This is where the score is decided.

The scoring formula

Efficiency starts at 100% and drops in flat 20% penalties:

• −20% for each attribute whose mean across your three microbes is outside the site's target range;
• −20% if none of the three carries the desired trait;
• −20% for every microbe carrying the undesired trait.

Read that middle word again: mean. No individual microbe has to fit the range. The average does. A microbe with an attribute of 2 is perfectly fine next to two 9s if the target range covers 6–7. This is the whole game: picking the three best individuals is not the same as picking the best three.

A worked example

Numbers make this concrete. Suppose a site wants all three attributes (call them Toxicity, Depth and Density) to land in the range 5–7, with a desired trait of "Aerobic" and an undesired trait of "Corrosive." Your pool includes these microbes:

• A: Toxicity 9, Depth 6, Density 4 · trait Aerobic
• B: Toxicity 3, Depth 6, Density 8 · trait Neutral
• C: Toxicity 6, Depth 6, Density 6 · trait Corrosive
• D: Toxicity 5, Depth 7, Density 5 · trait Neutral

The tempting pick is A, B and C. They look like the "strongest" microbes. Watch what happens. A+B+C gives mean Toxicity (9+3+6)/3 = 6 ✓, Depth 6 ✓, Density (4+8+6)/3 = 6 ✓. All three means land in range, and A carries the desired Aerobic trait. So far, 100%. But C is Corrosive: that's the undesired trait, and it costs a flat −20%. Final efficiency: 80%.

Now swap C for D. A+B+D gives mean Toxicity (9+3+5)/3 = 5.67 ✓, Depth (6+6+7)/3 = 6.33 ✓, Density (4+8+5)/3 = 5.67 ✓. Still all in range, A still covers Aerobic, and nobody carries Corrosive. Final efficiency: 100%. The "weaker-looking" trio scores higher because it dodges the trait penalty while the means still balance. That is the entire skill: read the traits first, then let the arithmetic of the means decide the rest.

The four mistakes that cost people the most

• Chasing the best individual microbes. The most common error, and the one the worked example above exists to kill. The score is built on means and trait penalties, not on any single microbe's raw stats.
• Missing an undesired-trait carrier. It's the harshest line item and it stacks: two carriers is −40%, three is −60%. Screen every candidate for the undesired trait before you look at attributes at all.
• Over-collecting the desired trait. You need exactly one carrier of the desired trait, not three. Spending two of your three slots on desired-trait microbes usually wrecks your attribute means for no extra points.
• Burning the clock on site one. Three sites in ~30 minutes means a hard ten-minute ceiling each. Candidates who perfect site one often run out of time on sites two and three, where the same points are available.

Strategy that follows from the formula

• Work backwards from the means. For each attribute, the three values must sum to between 3× the range floor and 3× the range ceiling. Two strong microbes define exactly what the third must contribute. That turns selection from vibes into arithmetic.
• Undesired trait is the harshest line item. It stacks per microbe: three carriers is −60% on its own. Screen for it first and reject early.
• Desired trait needs one carrier, not three. Once one microbe covers it, the other two slots are pure attribute-balancing material.
• Use the save-for-later option strategically. A flexible all-rounder that doesn't quite fit site one may be exactly the mean-corrector site two needs. The categorization step locks, so think one site ahead.
• Pace: three sites, ~30 minutes. Roughly ten minutes a site, and the fourth step deserves the biggest share of each.

Keep going

Before Sea Wolf comes the Red Rock Study walkthrough; after it, on 85-minute invites, the Sustainable Futures Lab. The full picture is in the McKinsey Solve guide. Ready? Start the free Sea Wolf scenario.