If you have a bunch of neutrons and put them in a box, and look again 10 minutes later, you will see that you have only half of them.
If pick all the neutrons that survived for 3 minutes, and put them in a box, and look again 10 minutes later, you will see that you have only half of the neutrons that survived for 3 minutes.
If pick all the neutrons that survived for 7 minutes, and put them in a box, and look again 10 minutes later, you will see that you have only half of the neutrons that survived for 7 minutes.
If pick all the neutrons that survived for 42 minutes, and put them in a box, and look again 10 minutes later, you will see that you have only half of the neutrons that survived for 42 minutes.
When the time is too long, you need to create a really big number of neutrons initially, so enough survive until you start the experiment.
So ... the waiting time until the experiment start doesn't matter.
It's easier to understand with a discrete model with coins. You have perfectly balanced coins with 50% chance of head and 50% chance of tail. Each minute you flip all the coins at the same time and remove all the "heads". So you can repeat this, and each time you have less coins. You start the experiment, flip the coins 10 times (and remove the heads), and you get 1000 coins that survived. How many additional times should you flip them to remove half of them and have only 500?
These are thermalized neutrons, that means they have bounced a few times in random directions. You can model this and have some kind of convolution of the result, and then deconvolute the experimental data or fit it. But they are trying to measure 1/100 of seconds, and this is possible but very noisy, so it's strange. Wait a minute, it's a exponential decay ... so it doesn't matter ...
