Expected vs Median Kill Count: Why Average KC Misleads

Cumulative drop probability curve for a 1 in 512 item marking the median at 355 kills and the expected count at 512 kills where only 63.2 percent have the drop
The median sits at 355 KC — half of players are finished there. The 'expected' 512 already reaches 63.2%, well past halfway. Median from ln(0.5) ÷ ln(1 − 1/512); curve plotted from 1 − (1 − 1/512)^k.

“Expected kill count” is the most quoted number in drop-rate discussion and the least useful one for judging your own grind. It is a mean, and the distribution it summarises is heavily skewed — so the mean sits well above the typical outcome.

Two different numbers

Expected KC (the mean) is simply the drop rate denominator: 512 for a 1/512 item. It is the long-run average number of kills per drop across many attempts.

Median KC is the point where half of players have finished:

Median = ln(0.5) ÷ ln(1 − 1/n) ≈ 0.693 × n

For 1/512 that is 355 kills — about 31% fewer than the expected count. Run it in the OSRS dry calculator as the 50% confidence level and you get the same figure.

The 0.693 rule holds everywhere

The ratio is the natural log of 2, and it barely moves across drop rates:

ItemRateMeanMedianMedian as % of mean
Tumeken’s Shadow (max inv.)1/24241667.9%
Scythe of Vitur1/17217211969.1%
Dragon Warhammer1/3,0003,0002,07969.3%
Twisted Bow1/3,4283,4282,37669.3%
Draconic Visage1/5,0005,0003,46569.3%

Above a few hundred, it locks at 69.3%. So a fast mental estimate for “what does a normal grind look like” is seven-tenths of the drop rate.

Why the mean sits so high

Waiting-time distributions have a long right tail. There is a hard floor — you cannot get the drop in fewer than one kill — but no ceiling. A handful of players go 4× or 5× the rate, and those extreme runs drag the average upward without moving the middle.

The median ignores the tail entirely: it is the middle person, whatever the extremes do. That is exactly why it is the better description of a typical grind.

Concretely, at 1/512:

MilestoneKillsShare finished
Median35550%
Expected KC51263.2%
90%1,17890%
99%2,35699%

By the time you reach the “expected” count, nearly two-thirds of players are already done. Arriving there empty-handed puts you in the remaining third — unremarkable, and the ranking in how unlucky am I makes that explicit.

Which one to use

Use the median to set expectations. “About 355 kills is a normal grind” is a truer statement than “512 kills is expected”, and it will not leave you feeling cheated at a completely ordinary kill count.

Use the mean for aggregate planning. If you are estimating the total value of drops across thousands of kills, or costing a long-term grind in bulk, the mean is the correct tool — that is the question it answers.

Use neither as a promise. Both describe distributions. Neither says anything about your account, and neither becomes more true the longer you go without.

The 63.2% constant, restated

Since the median is 0.693n and the expected count is n, the expected count always lands at the same place on the curve: 1 − 1/e ≈ 63.2%. That is why the figure appears identically for every drop rate. It is not an OSRS quirk — it is a property of the geometry, and it is the reason “I hit expected KC and got nothing” describes about a third of every player base on every item simultaneously.

The practical takeaway

If someone tells you an item “takes 3,428 kills”, the honest reply is that half of players have it by 2,376, and a third of players still will not have it at 3,428. Both facts come from the same curve.

Plan around the median, budget for the 90% mark if the item genuinely matters, and treat anything past 2× as the tail you knew existed when you started. The errors that make an ordinary streak feel extraordinary are collected in 5 drop rate mistakes that make you think you’re dry.

References & Sources

  1. [1]
    NIST/SEMATECH e-Handbook - Gallery of Distributions (opens in new tab)

    Supports: Reference properties of discrete probability distributions, including the distinction between mean and median.

    Verified

  2. [2]
    NIST/SEMATECH e-Handbook - Binomial Distribution (opens in new tab)

    Supports: Independent Bernoulli trial model that produces the skewed waiting-time distribution described here.

    Verified

  3. [3]
    Old School RuneScape Wiki - Drop Rate (opens in new tab)

    Supports: Current per-item drop rates used in the worked comparisons.

    Verified