Unit 6 · Capstone
Build the thing you named
The principle
A build is finished when someone else can run it from your written instruction and get a result you would put your own name on.
By the end: You can take the recurring task you named in Unit 0 and produce a working, reusable build for it without a walkthrough, then score it against a fixed rubric.
The situation
It is the Thursday from Unit 0. Same report, same four numbers out of one system and three out of another, same two paragraphs of commentary that are the only part anyone reads. Since then you have built a daily brief that works, configured a workspace that holds your standing instructions, drawn a one-page tool map, made something for your own role's work, and written down what data you are allowed to paste. Thursday still takes ninety minutes.
That is not a failure of the course. Four of those five come into this hour with you: the workspace holds your context, the data line sets what you may paste, the baseline is the number to beat, and the versioning habit is how you prove it repeats. The tool map and your track's page are team and role artifacts — not graded here, and never meant to be. What nobody hands you is the brief itself. The task is yours, the inputs are yours, and the only person who knows what good looks like on Thursday is you.
The principle
A build is finished when someone else can run it from your written instruction and get a result you would put your own name on.
Three things make that sentence true. The instruction lives somewhere other than a chat window, or nobody else can run it. The output holds its shape across two different weeks of real inputs, or you have a one-off that happened to work. And you can say what you check before it leaves you, or you are trusting an output rather than verifying one.
Notice what is not on that list: how clever the wording is, how many features the tool has, how impressed you were the first time it ran. The bar is a colleague, a repeat, and your signature.
Try it
Sixty minutes, one sitting, on the task in your Unit 0 sentence: "Every [interval] I turn [inputs] into [output] for [who reads it]."
The brief
Build the thing that does that task. Run it on real inputs from the last two weeks, then again on a different week's inputs without editing the instruction. Keep the build and both outputs.
The constraints
- Your task, the one you named. Not the cleaner one you thought of since. If the original fails one of the four tests from Unit 0, write one line saying which test, then take the nearest task that passes.
- Real inputs. From the system they actually come out of, inside the data line you wrote in Unit 5. Sanitise first. If the task cannot be done inside that line at all, stop and write down why — that finding is worth more than the build.
- Unaided. No walkthrough, no starter template, nobody building it next to you. Rereading any lesson is fine; being talked through it is not.
- One tool. The workspace you set up in Unit 2. Do not open a second one to rescue the first.
- Sixty minutes, once. If it is not working at sixty minutes, stop and record the step where it broke. An honest failure at a named step is a result. An unbounded afternoon is not.
The rubric
Seven rows. This is what the graded track marks against, unchanged.
| Criterion | It passes when |
|---|---|
| Written instruction | The job, the inputs, the output shape and the standard exist as text you could paste into a fresh account |
| Standing context | Anything true every week sits in the workspace instead of being retyped each run |
| Real inputs | It ran on your own data from the last two weeks, not a sample and not an invented example |
| Stable output | Two runs on two different weeks produced the same structure with no edit to the instruction |
| Named check | You can state, in order, what you verify before the output leaves you, starting with whatever is most likely to be wrong |
| Data line | You can say what you deliberately kept out, and why |
| Beats the baseline | It is faster than the number you recorded in Unit 0, and the result is one you would send under your own name |
All seven, or it is a draft. Six is a build with one known hole, which is fine as long as you write the hole down. Four or five means you have a good conversation, not a build — the missing rows are almost always written instruction and stable output, both fixable in twenty minutes.
Grading
This is the free version, so you are the grader — treat that as a real limitation, not a formality. The row people pass themselves on without evidence is stable output: a second run on a second week is the step everyone skips and the only one that separates a build from a lucky afternoon. Do it before you score yourself.
The certificate is not issued for time spent or lessons read. It is issued for a submitted build, marked row by row by someone who did not make it. If that matters where you work, take the graded route. If not, you still have a working build and ninety minutes back every Thursday, which was always the point.