Gold Splits and Sum of Best: Timing Your Work Blocks

I plan my days in ninety minute blocks, and for a long time I judged each block with a single question at the end: did I finish? Most days the answer was "mostly," which told me nothing. A block where I lost forty minutes to a flaky test and a block where I was simply slow looked identical in my notebook. Both were just "mostly."

The fix came from watching a friend speedrun Celeste. Her timer did not show one number. It showed seven, one for each chapter, and each one was green or red against her best. She knew, to the second, that her run had been lost in Mirror Temple and nowhere else. I wanted that for my working day, and after about four months of borrowing speedrunners' tools, I have something close to it.

What a split timer actually shows

Most runners use a free program called LiveSplit, or something like it. Before a run, you divide the game into segments: chapters, levels, bosses, whatever makes sense. As you play, you press a key at the end of each segment. The timer records how long that segment took and compares it with your history.

Three numbers matter, and the vocabulary is worth stealing:

• Personal best (PB). Your fastest complete run, and the split times from that run.
• Gold split. Your fastest ever time on a single segment, even if it came from a run that fell apart later. The timer colours it gold.
• Sum of best. All your gold splits added together. It is the run you could do if every segment went as well as it ever has.

The gap between your PB and your sum of best is where the time is hiding. A runner with a 32 minute PB and a 29 minute sum of best knows there are three minutes available, and the splits show exactly which segments hold them.

Splitting a work block

A block of focused work has segments too, even if we never name them. For a typical feature at my job, mine look like this: read and understand the ticket, write the failing test, make it pass, clean up, open the pull request with a description. Five splits.

I do not use LiveSplit for this, although you could. I use a plain text file and a stopwatch on my phone, and I write the time down at each boundary. It takes about four seconds per split. After three or four weeks of the same kind of task, I had enough history to see my own golds.

The surprise was where the time actually went. I had assumed the slow segment was "make it pass," since that is where the real thinking happens. It was not. My worst segment by far was the first one, understanding the ticket. My gold for that split was 6 minutes; my average was 19. The difference came almost entirely from tickets I started without asking the author a single clarifying question. That one finding paid for the whole experiment.

Do not reset the run

Speedrunners reset constantly. If the first two splits are bad, the run cannot beat the PB, so they restart from the title screen. In a game this makes sense, because the only goal is the record.

I tried this at work for about a week and it was a mistake. When the first segment of a block went badly, I would abandon the task and switch to something easier, telling myself I would "restart fresh" later. What I actually did was pile up half-started work, which costs a great deal more than one slow segment. I wrote about that cost in the real cost of switching tasks every ten minutes, and resets are the same problem with a gamer's excuse.

A runner resets to chase a record. A worker who resets is just starting over, with the loss from the first attempt still on the books.

The habit I kept instead is what runners call a practice run. When a segment goes badly, I finish the block anyway and mark that split as red in the file. The run is not a PB, but the data still counts, and I still ship the work.

What golds tell you, and what they do not

Your gold splits show what you are capable of when things line up. They do not show what you should expect every day. A sum of best is almost never achieved in a single run, even by the best runners, because some golds came from lucky days. Treating your sum of best as a daily target is a fast way to feel like a failure every evening.

What golds are good for is diagnosis. If one segment's average sits far above its gold, that segment has a fixable cause. For me it was unclear tickets. For a colleague who tried this, it was the pull request description, where she was spending 25 minutes on average against a gold of 8 because she kept rewriting it for imagined critics. She now writes it from a three-line template and moves on.

If the average is close to the gold, there is not much to gain there, and you can stop worrying about that part of your day. This was a relief. I had been feeling guilty about how long testing took, and the splits showed it was already near my best. Lachlan's piece on why gaming sessions make good models for time blocking gets at a similar idea from the planning side.

Set up your splits this week

Pick one kind of task you do at least three times a week. Write down four or five segments for it, in order, with names short enough to type quickly. Each time you do the task, note the clock at every boundary. After ten runs, look for the segment with the biggest gap between its best time and its average. That is the one worth changing, and you will probably be surprised which it is. If finding uninterrupted time for the runs themselves is the harder problem, Ines covers that in protecting your deep focus hours each day.