No. 15 strategy
Getting Better at Ballpark
Ballpark scores two things: where you put each marker, and how honestly you say how sure you are. This guide is mostly about the second, because it's the half you can improve without knowing any more.
It assumes you know the rules. If you don't, the how-to has them, with an example animation. Every figure here was worked out from the game's scoring code and its puzzles. Some are counted straight off real days, some come from a simulated player run through them, and each one says which.
What the score rewards
Each of the five estimates is a card worth up to 100, and what it pays depends on your claim and on how far off your marker was. The within-10% row includes the 5 for spot on.
What each claim pays, by how far off the marker was
±50% ±2x ±3x ±5x within 10% 100 95 87 78 1.5x off 95 90 82 73 2x off 76 90 82 73 3x off 49 63 82 73 5x off 16 29 49 73 10x off 0 0 3 27
Read across a line. When the answer lands, one chip tighter is worth only a few points more. When it doesn't, the tight claims fall away fast while the wide ones still pay in full.
That's the whole strategy, and I built the scoring to make it the best one. Every step wider costs a little when you were going to land anyway, and every step the answer falls outside costs more. The two balance at a range that holds about three times in four. Claim tighter and the misses outweigh the extra; claim wider and you pay for safety you didn't need.
For the builders: where three in four comes from
A claim's price drops by a fixed amount for each tenfold step of width, and a miss costs a bigger fixed amount for each tenfold step past your edge. Widen a little and you pay the first cost for certain, but save the second only on the days the answer was outside.
So the best width is the one where your chance of missing equals the ratio of the two costs, which comes to 27.5%. There are four chips rather than a dial, so the target is only ever approached: my simulated players with a typical miss anywhere from 1.3x to 3x, each picking the best chip for it, landed between 70% and 76% of their claims.
It also means no one claim habit works every day. Hedge every card at ±5x and even a perfect day of it stays at Jam, unless most of the five were spot on, in which case you knew enough to claim tighter. Claim ±50% on everything and the misses pile up. The scoring tiers themselves were priced from a simulation rather than real play, so they may still move.
It also gives you a way to check yourself. In my simulation, a player on the best chip for their own error lands about three claims in four, and all five on only one day in four to six. All five landing most days means you're claiming too wide, and two or fewer means too tight. Your Track record shows how often you use each range.
Tight where you know, wide where you don't
A real day is rarely five cards of equal doubt. More often you know one or two of the things well and are inventing the rest, and that's where the claims earn their keep. So I modeled that player: within 1.3x of the answer half the time on two cards, and within 5x half the time on the other three.
Matching the claim to the doubt beat leaving every chip on the default by about 20 points a day, and it raised the share of days at Cat or better from about one in six to about one in four. Getting it backwards was the worst of the five, worse than hedging everything.
That makes the default worth a second look. A fresh card starts on ±2x, which is the best claim only for someone whose estimates are typically within about 1.5x, a good day on a thing you know. The same model gives the best chip for each kind of error, where "within" means half your estimates land that close:
- Within 1.3x or closer: ±50%
- Within 1.5x to 1.7x: ±2x
- Within about 2x: ±3x
- Worse than that: ±5x
The first move
Start with the thing you're surest of. It becomes the yardstick for the other four, and once one marker is down you're comparing rather than estimating from nothing. On a weight day with a house cat on it, the cat goes first, and it's worth claiming tight, since it's the one card you can afford to.
Then work in ratios from it. You may have no idea what an adult male polar bear weighs, but how many house cats it weighs is a question you can attack. A big dog is several cats, and a polar bear is several big dogs. That's rough, but it's a number with a reason behind it, which is the kind you can size a range around.
Reading the question
The unit sits beside the instruction, and if it isn't the one you think in, the other is one tap away: kilograms or pounds, kilometers or miles, with only the labels changing. And every answer sits somewhere on the board, so an estimate that wants to run off either end is a sign to rethink it.
Then read the labels word by word, because the small words move the answer. A few real ones:
Words that move the answer
An adult male African elephantThe heavy end of the animal, not a typical one.
An average adult wolfThe middle, not the biggest wolf you've heard about.
The fastest racehorse ever timedA record, so past any horse you've seen run.
Read the scale's labels too. The scale is marked in powers of ten, but a wide day skips every other one, so neighbors sit ten or a hundred apart: check which before you place anything. Halfway between labels a factor of ten apart is about three times the lower one, and halfway between labels a hundred apart is ten times it.
And be careful with the famous numbers. Some figures everyone has heard turn out to be folklore. When the only thing you know about a number is that you've heard it somewhere, that's a reason to claim wider, not tighter.
When you have no idea
Before you reach for the widest claim, try the Fermi move: break the number into smaller numbers you do know. Say the question were the weight of a full bathtub. You know roughly how big a tub is, and a cubic meter of water weighs a tonne.
A made-up question, worked the Fermi way
1.5 m long × 0.7 m wide × 0.3 m of water = about 0.3 cubic meters × 1,000 kg per cubic meter = about 300 kg
None of the three is exact, and that's fine. In a chain like this the high guesses and the low ones tend to cancel, as long as you aren't leaning the same way on all of them. The depth is the shakiest, so size the claim around it: ±2x on 300 covers anything from 150 to 600.
If there's truly nothing to build from, bracket it. The other four things are clues: you may not know this one, but you can usually say it's bigger than one of them and smaller than another, which puts the marker in the right stretch. Then claim wide. ±5x keeps paying something until the answer is about 15 times your marker, where ±2x stops at about 8.
The order bonus
Half the pairs right pays nothing because five markers thrown down at random get about half of their ten pairs right. Each pair above that is worth 10. It's the part of your answer that survives a bias: estimate everything three times too low but in the right order, and you still bank the lot.
- Tenfold steps from one end of the board to the other
- 3 to 10.7
- Six on a typical day, so the right end is a million times the left.
- Typical gap from one answer to the next one up
- 13x
- The median over every neighboring pair.
- Days with two answers within 2x of each other
- Over 1 in 4
- Where most of the order bonus is won or lost.
Counted day by day, the five answers are usually far apart, and only about one neighboring pair in twelve sat within 2x of each other. So if your estimates are roughly right, the order mostly takes care of itself, and the points are decided on the days with a close pair. Even then a swapped pair costs 10 points, less than a bad claim on a single card.
No clock, one shot
With no timer and no second go, the only reason to rush is wanting to know, and the moment before Submit is the one worth spending. A short check before you press it:
- Read down the five markers. For each pair, is the one further right the one you believe is bigger?
- Check each claim against your doubt. Tight on what you know, wide on what you invented. Widening one is a fair move.
- Type anything you actually know. A dragged marker that's close isn't the same as one that's on the number.
- Keep the default only on purpose. A card still on ±2x is a claim that you're fairly sure of it.
After Submit, read the reveal's words, not just its colors. A miss says something like "2.1x low", measured from the edge of your range, not from your marker: it's how much wider you needed to be. "Spot on, but hedged" says the opposite: you knew more than you claimed. If the reveal keeps saying "low", you have a lean, and the fix is to move your markers, not to widen your claims. The archive keeps the days you missed, and a missed day counts when you play it.
Sources
- Ballpark's scoring code The payout table, the order bonus, the bust points and the three-in-four balance are read from it when this page is built.
- Ballpark's released days Counted directly: how wide each board was, the gaps between neighboring answers, and the days with a close pair.
- A simulated player, run through the same days Each estimate is the true answer off by a random factor, set so that half land within the stated miss, and clamped to the board. A thousand plays of each day, scored by the game's own rules. Real players vary card to card more than this, so read the gaps between habits, not the exact points.
- Fermi problem, Wikipedia Why errors in a chain of rough numbers tend to cancel when there's no consistent bias.
- Calibrated probability assessment, Wikipedia Ranges, and how often the truth lands inside them.