Retrospective · Example deck
A worked example of writing a psi-briefing deck in Markdown
This deck is one Markdown document, examples/example-deck.md, converted with node tools/md-to-deck.mjs examples/example-deck.md. Thirteen slides, no HTML written by hand, and everything on them is placeholder content. The three numbers on the title slide come from the front matter, and each # heading became one of the three numbered dividers.
The file’s order is the deck’s order: front matter, then one # per part and one ## per slide, with the blockquote at the end of a slide becoming the band across its bottom.
Count the directives in the .md: five of them (::: chart, ::: detail, ::: delta, ::: principles, ::: editorial) on four of the thirteen slides. The other nine name no layout at all: this one, the title, three dividers and four content slides. A table became the plan/reality list, a numbered list became the steps, four ### blocks became the 2×2 grid – and the report the converter prints names the rule it used for every slide.
Part one
What the eleven weeks looked like from the inside, and the numbers behind them.
Activity
Commits on main, and on any branch that eventually reached main – a branch that was abandoned still cost the time it took, but it did not ship, so it is not in this number. One commit per author per push, which means a pair programming session shows up once rather than twice.
Reverts count as commits. Undoing work is work, and a week with three reverts in it was a real week – flattening that would make week five look calmer than it was.
Merge commits, because a merge is bookkeeping rather than a change. Anything touching only dist/ or a lockfile, for the same reason. And the two history rewrites in week six, which would otherwise have shown up as 300 commits on a single afternoon.
Together that is roughly 9 % of the raw count – enough to matter if you compare this chart against a number someone else counted differently, and the reason this panel exists at all rather than a footnote nobody reads.
Part two
Where the 1280 commits went, and the weekly loop that produced them.
620 commits, three authors. The guided flow and the component library, the only part of the work users ever see.
410 commits, three authors. Enrolment logic and the data model; 300 of the 410 landed in the first six weeks.
180 commits, three authors. Copy, translations, and the eleven weekly notes, written by the same people who built the thing.
70 commits, three authors. Deploy scripts and the font-embedding pipeline, none of it touched after week 3.
One outcome, written down on Monday morning, in a single sentence. If it took two sentences it was two weeks of work and got split.
Small commits, deployed continuously, no staging queue to wait on. Anyone could see the portal as it stood at any hour.
Look at the real thing running, never at a status document. Bugs found here were fixed before the note was written, not filed.
A short note per week, describing what now works that did not before. Eleven notes are where these slides came from.
Scope came off at the start of a week, while there was still time to reshape it, never in a panic two days before the review.
The Friday build stayed deployed over the weekend. Anyone who wanted to look – registrar, IT, a student – could, without asking anyone.
Part three
Eight rules that were on the wall, and the four lessons that outlived them.
Nothing aligned the team like a working artifact on a Friday afternoon. Eleven weeks produced eleven builds and eleven notes, each note written from the build it describes.
The page from week 1 held for all eleven weeks: two features came off it in week 6, nothing was added.
34 % of commits landed before 10 a.m. No meeting was booked before then.
Infra took 70 of the 1280 commits, and none after week 3.
This deck is designed for landscape. Rotate your device, or tap to dismiss.