What Speedrun Split Timers Teach About Honest Time Estimates

I have a confession that will be familiar to anyone who has tried to plan a week. For about two years, I blocked out 45 minutes every Friday for my monthly client report, and for about two years it took well over an hour. Every Friday I was surprised. Every Friday I blamed something specific: a slow spreadsheet, a missing figure, an email that came in halfway through. It took a speedrunner friend glancing at my calendar to point out the obvious. I was not unlucky. I was planning around my best run.

Most of us estimate time this way, and it quietly wrecks schedules that look sensible on paper. Speedrunners, people who try to finish video games as fast as possible, have a tool for this exact problem, and it works just as well on a Friday report as on Super Mario 64.

How speedrunners split a run

A speedrun is timed from start to finish, but serious runners do not just watch the total. They use a split timer, most commonly a free program called LiveSplit, that breaks the run into segments: a level, a boss, a section of a map. Every time they finish a segment, they hit a key and the timer records that split.

Over hundreds of attempts, the timer builds up a detailed picture. For each segment it keeps a best ever time, which runners call a gold split, along with the times from their personal best run. Add all the gold splits together and you get the sum of best: the theoretical time you would get if every segment went as well as it ever has, all in the same attempt.

Here is the key thing. Almost every runner's personal best is slower than their sum of best, often by several percent. Getting a gold on every split in one run basically never happens. Experienced runners know this, and nobody plans their evening around hitting their sum of best.

Our calendars are full of sums of best

When I estimated my report at 45 minutes, I was quietly adding up the best time I had ever done each part. Pulling the data had once taken me 10 minutes. Writing the summary had once taken 15. Checking and sending had once taken 10. Add some slack and you get 45. But those best times never happened on the same Friday.

This pattern has a name in psychology: the planning fallacy. In a well known 1994 study, Roger Buehler and colleagues asked students how long their senior theses would take. The average estimate was 33.9 days. The average actual time was 55.5 days. Even when asked for a worst case estimate, students said 48.6 days on average, and reality beat their worst case. Fewer than a third finished by the date they first predicted.

We estimate from our gold splits and live through our average ones.

Speedrunners avoid this trap not because they are more disciplined than the rest of us, but because the timer shows them their actual distribution every single time.

What my report really looked like

So I did what my friend suggested. For six Fridays in a row I split the report into four segments and wrote down the time for each one as I went. No changes to how I worked, just recording. The results, in minutes, rounded to the nearest one:

My sum of best was 41 minutes, almost exactly what I had been booking. My average was 71. The report never took 41 minutes in those six weeks, and it went over 90 minutes twice. No wonder every Friday felt like a small failure.

The table also told me where the variance came from. Pulling figures and writing the summary were the two segments with the widest range. The figures varied because one of the three systems was often slow to export. The summary varied because some months had something real to say and some did not, and the empty months were harder.

The advice to "add a buffer" is half right

The usual fix for bad estimates is to add a buffer, often 20 or 25 percent. That would have taken my 45 minutes to about 56, still 15 minutes short of my average. A flat buffer assumes your estimate is roughly right and just needs padding. If your estimate is really a sum of best, the gap is usually much larger than any polite buffer.

What worked better was to plan from the average and then work on the segments with the biggest spread. I now book 75 minutes. I also pull the slow system's figures on Thursday afternoon, which cut that segment's worst case dramatically. Over the next two months my average dropped to around 58 minutes, and I finish early more often than late, which feels completely different from finishing late every week.

If you have tried time blocking and found that your blocks never hold, this is often why. The post on why time blocking fails the first few times covers other reasons. For a broader look at measuring where your hours go, see where your time actually goes.

Running your own splits

You do not need LiveSplit. A note on your phone works. Pick one task you do at least weekly and always underestimate, something like a report, a weekly review, the school run, or batch cooking on Sunday. Break it into three to five segments that have a clear end point. Each time you do it, write down the minute you finish each segment.

After four or five runs you will have a gold, an average and a worst for every part. Book the average. Then look for the segment with the widest gap between gold and worst, because that is where a small change does the most good.

This week, choose the task and write its segments on a sticky note before you start it. Then time just this one run. One run will not give you an average, but it will almost certainly show you that your current estimate is somebody's gold split.