The last job is taking the scaffolding down. The folder the report was built in wasn't fit to publish. This is how it became the clean public kit, including two slips on the way and the rule they left behind.
with Elizabeth Cecilia. Tap any underlined word for a quick definition.
Have the report open. The lessons point at real pages and tiles. The .pbix opens in Power BI Desktop; the PDF works on any device, but things like View as and the i buttons need Desktop.
The folder the report was built in had grown for weeks. It held a long diary of decisions, working notes, names from the project it was first made for, personal paths, and a git of 77 commits, all carrying a personal email address. None of that belonged in public.
Predict first: What's the safest way to turn a working folder into something you can publish?
01
Copy from the last commit
The copy was made with , from the last commit, leaving out a list of files agreed in advance: the notes, the old handover, the original readme, a second, larger mock-up, an old theme and a .pbix.
git archive HEAD → C:\conducting-the-agent-kit
(no history, no dropped files)
The working folder also had uncommitted changes: two test measures and a copied slicer, left over from trying something in the seed.
Predict first: Why copy from the last commit, not from the folder as it stands?
02
Rename, scrub, and prove nothing else moved
In the copy, every file name with the old project's name in it was renamed to match the course. In the workbook, exactly 22 cells were changed. Then the copy was compared with the original:
sheets 33, tables 33 names and ranges identical
cells compared 9,847
cells different 22 (exactly the listed ones)
Around a hundred code comments that named a person became neutral. Comments that recorded only a date kept it where it explained something.
Predict first: Why compare all 9,847 cells after changing only 22?
03
A check that tests itself
Before anything was published, a searched every file: names, text, every cell, sheet name and property in the workbook. It also failed on any .pbix, any cache file, and any history longer than one commit.
It passed first time. That wasn't the end of it.
Predict first: The scan says PASS on its first run. What should happen next?
04
Two slips
Two things went wrong. Both times the agent caught it, stopped and reported, and nothing was lost.
Slip 1. To compare the generators' output, the agent patched a scratch copy of them to find their own folder. The patch matched nothing. The run was ed straight after it, so the generators ran against the original working folder and rewrote 265 files. Three came out different. git put them back, with a yes.
Slip 2. An edit to a template silently changed nothing, because of the file's . The commit chained after it ran anyway. The result had a new id and the same content.
Predict first: What did the two slips have in common?
05
The rule they left behind
Fixing the slips wasn't enough. The lesson went into the kit's , where every future session reads it first:
Check each edit before the next command. Never chain an edit to a commit, a run or another edit. After every edit, confirm the change is in the file before doing anything else. A command can report success and change nothing.
That's the rule lesson 1 showed you. The very next edit after it was added was made, checked and committed as separate steps.
Predict first: Why put the rule in the brief, rather than just fixing the two slips?
06
One clean commit, then public
The finished copy got a fresh git history: a single commit under a GitHub no-reply address. A file of line-ending rules made sure that a fresh clone matches the folder byte for byte. Then came a licence, a push, and a with the .pbix and a PDF attached.
Check it in the report: this is the release the report buttons at the top of every lesson link to: v1.0, with sample.pbix and sample.pdf.
Predict first: Why start with one fresh commit, instead of keeping the 77 old ones?
07
The scaffolding comes down
That's the series. The brief, the specs, the workbook, the generators, the phases and the register put the report up, and a careful copy took the scaffolding down. The kit you can download is what was left standing. Next: Reading the Tiles, the finished report read tile by tile.
Publish from a clean copy, never the original. Prove each change touched only what it should. Test your checks by making them fail.
And when something slips, turn it into a rule in the brief, where the next session will read it.
Terms in this lesson
Earn 5 XP for every term you know.
Get the project files
Everything is free on GitHub. Open the report next to the lessons and check each step for yourself.
The whole project
The Power BI project, the seed project, the specs and the tools: the whole kit. On the GitHub page, click the green Code button, then Download ZIP, and unzip it anywhere.
The Critical Raw Materials Intelligence report as a single .pbix file to open in Power BI Desktop. Open the latest release under Releases, then Assets, on the right of the GitHub page.