A week became a day
Somebody asked how long the checkout would take and I said about a week. Two days later there was a date in the plan. December 4, in a row with my name next to it, and the word “about” wasn’t anywhere in the document.
The honest version of what I thought had a hinge in it. Five days if the payment provider’s sandbox matched their current documentation, closer to nine if it was a version behind, which it had been on the last two projects I’d used it on. And if there turned out to be a 3-D Secure requirement nobody had said out loud yet, longer than that and less pleasant.
I did say the hinge. I remember saying it, because I remember the pause afterward where everyone waited politely to see whether a number was coming.
So the guess wasn’t the problem. Five days is about what the good version would have taken, and I’ve gone back and checked. The problem is what happened to the sentence on its way to becoming a document. I can’t help but think the reason nobody writes about this half of it is that there’s no villain anywhere in it.
A game of telephone with a deadline
I traced the route afterward, mostly out of morbid curiosity. It’s four hops long and it took two days.
On the call I said about a week, depending on their sandbox. In the notes it became “checkout ~1wk (Madison),” which is a fair record and still has a tilde in it, to its credit. In the plan the tilde came off, because a bar on a chart starts somewhere and ends somewhere and there’s nowhere on it to draw a maybe. In the client update it became the week of November 26. And in the client’s own internal deck it was December 4, which is simply the date the bar happened to stop on.
Four people handled it. Every one of them did their job correctly and produced the document they were being paid to produce. Nobody deleted the condition, exactly. There was just less and less room to keep carrying it, and then there wasn’t any.
Nobody can put a maybe in a cell
I suspect there’s nothing more interesting going on than paperwork. Every artifact downstream of my mouth has a fixed shape, and not one of those shapes has a field for a condition.
A bar on a chart has a start and an end. A cell holds one value. A SOW has a date in it because a date is the thing a client can hold you to, which is most of why the document exists at all. So the qualifier never actually gets argued with. It gets dropped for want of anywhere to put it, the way a footnote vanishes when you paste into a text box.
And the qualifier is the valuable part. The number is a summary of what I know. The condition is the knowledge, which is that the whole job turns on one third party’s docs being current. That’s a fact a person could plan around if they had it. Same shape as a brief, which is the minutes of an argument with all the disagreement edited out to make the document look decisive. They cut the part you wanted.
Nothing takes negative time
If the condition could break either way, losing it would cost nothing much on average. You’d be over just about as often as you were under, so six one way half dozen the other.
But the range isn’t symmetric around the number and it can’t be. There’s a floor on the good end, which is the version where everything goes right. There’s no ceiling at all on the other one. Nothing takes negative time. So when a range gets flattened into a value it flattens toward the optimistic end, because the optimistic end is the only one close enough to reach, and the bad case behind the nine has no particular edge to it that I could name a date past.
Which makes every hop a small quiet subsidy pointing the same way, and four hops of that is a week that ends on a single date.
The literature calls the personal half of this the planning fallacy, and the finding everyone quotes is that we underestimate even for work we’ve done before, even knowing we underestimate. I believe all of it. My only complaint is that it locates the whole problem inside one person’s head, and about half of what I’ve watched go wrong happened in a shared spreadsheet after the head was finished with it.
The other half is the work that arrives instead of the work you picked, which never makes it into an estimate because it never makes it into anything. Five days of checkout work does not fit inside five days of calendar and never has, on any project I’ve been on.
Put it where they’ll copy it from
So what I’ve changed isn’t the guess. It’s where the guess lives.
Answer on the call if you’re asked on the call. But the number that reaches the plan is not the one I said, it’s the one somebody wrote down, so now I write it down myself, in the thread the plan gets built out of, before the end of the day. Whoever copies it copies whatever is sitting there.
Two things about how it’s written, both of which I got wrong for a good while.
Put the condition in the same sentence as the number. If it’s in the following paragraph it doesn’t travel, because the following paragraph is not what gets pasted into the cell. One sentence with a hinge in it is much harder to shave than a number with a caveat underneath it.
And name the question, with a date for answering it. “I can tell you which by Wednesday” turns a number somebody has to feel uneasy about into a thing that resolves itself on a known day, which is a far easier item for a project manager to hold. In my experience they’ll take that trade every single time, because a question with a date is schedulable and a bad feeling isn’t.
Checkout is five days if the payment provider's sandbox
matches their current docs, and nine if it's a version
behind, which it was on the last two projects.
I can't tell which from here. Give me half a day against
their sandbox and I'll know by Wednesday. Until then the
plan wants the nine, and if the nine doesn't fit, that's
worth finding out now instead of on the 4th.It isn’t elegant and it’s a few sentences longer than “about a week.” It also survives being copied by somebody in a hurry, which the spoken version demonstrably does not.
This is the same argument I made about commit messages last month, which I didn’t notice until I was most of the way through writing this. The moment you know the most about something is the moment right before you stop thinking about it, and the standard practice in that moment is to say something short and go get a coffee.
Just double it, then
The traditional fix is to work the honest number out in private and then double it. I did that for about two years and I know plenty of people who still do. I’m not above it either.
It doesn’t help with any of the above and I suspect it makes things worse. A doubled number is a single value with the reasoning stripped out of it, which is exactly the format that travels well, so all you’ve really done is hand the chain a cleaner thing to copy. It arrives in the plan carrying precisely as much explanation as the honest one would have, which is none.
Then there’s the second-order part, which is that everybody knows. Padding is an open secret in every agency I’ve worked in, so the number gets quietly discounted on the way past, and now two people are negotiating over a fiction neither of them will name. Meanwhile the real risk, a third party’s documentation, is still invisible to every person who might have done something about it. The project manager would have happily moved the checkout later in the schedule if anyone had told them which week the danger was sitting in. Nobody told them, because I’d folded it into a number.
December 4
The sandbox was a version behind. Of course it was.
It took nine days, which is the number I’d have given if anyone had asked me for the bad one, and the date moved to the following week, and the meeting about it was perfectly civil. Everybody was reasonable. Nobody so much as raised their voice.
But it was a meeting about me being late. That’s the actual cost of the whole thing, and it isn’t measured in days. A week earlier it could have been a conversation about a payment provider whose sandbox lags their own docs, which is a genuinely useful fact about the world for a company that sells things online. Instead it was an item about a developer who said a week.
I don’t think I’ll ever get much better at the guess itself. Three years into doing this for a living, my five days is roughly as good as my five days is going to get, and the research suggests practicing wouldn’t move it a great deal anyway. But the guess was never what cost me December 4. A document did that, and documents I can actually do something about, mostly by writing the awkward sentence down while I still know why it’s awkward, instead of hoping somebody thinks to ask me for it later.