Retrospective · Example deck

Project Aurora: shipping a portal in eleven weeks

A worked example of writing a psi-briefing deck in Markdown

11Weeks
1280Commits
1HTML file
1How the timeline went
2How it was built
3What we learned
Project Aurora · example deck

What this file is

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.

How to read it

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.

Nine of thirteen slides name no component. The shape of the Markdown was the instruction.
Project Aurora · example deck
1

Part one

How the timeline went

What the eleven weeks looked like from the inside, and the numbers behind them.

Project Aurora · example deck

The numbers

1280Commits across the whole project
117Person-days worked by the three of us
34%Of commits before 10 a.m.
3People on the core team
1280 commits over 117 person-days is eleven a day, and a third of them landed before 10 a.m. The mornings were kept free of meetings.
Project Aurora · example deck

Plan against reality

What actually happened
  • Kickoff and scope, agreed in a room, on one page. Nothing was written down that could not be read aloud in a minute.
  • First clickable prototype. Planned for week 3, and a week early because the component library came from an earlier project.
  • The scope argument. Two weeks of low commit counts that the chart on the next slide makes look like a holiday.
  • Feature freeze, planned for week 7. Two features were cut outright rather than deferred, which is the only reason the date held.
  • Load testing found the one real bottleneck: a query that had never been run against production-sized data.
  • Launch, on the day named in week 1, with the two cut features still cut.
Exactly two dates moved, both a week earlier than planned. Week 11 was week 11 from the first day.
Project Aurora · example deck

Activity

Commits per week

Weeks four and five produced 115 commits between them, fewer than week three alone. Both weeks went on the scope argument.
The eleven bars add up to 1280. Merges and generated files are not among them.How this is counted →

How a commit is counted

What is in

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.

What is out

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.

Counted the other way – every commit object in the repository – the same eleven weeks come to 1407. Pick one definition and stay with it.
Project Aurora · example deck
2

Part two

How it was built

Where the 1280 commits went, and the weekly loop that produced them.

Project Aurora · example deck

Four areas of work

Frontend

620 commits, three authors. The guided flow and the component library, the only part of the work users ever see.

Backend

410 commits, three authors. Enrolment logic and the data model; 300 of the 410 landed in the first six weeks.

Content

180 commits, three authors. Copy, translations, and the eleven weekly notes, written by the same people who built the thing.

Infra

70 commits, three authors. Deploy scripts and the font-embedding pipeline, none of it touched after week 3.

All three names appear in all four commit logs. Nobody owned a layer.
Project Aurora · example deck

What changed for the user

Seven separate pages to enrol
One guided flowSame information in 5 clicks instead of 19, answers carried forward
PDF forms, printed, signed and scanned
Fill in the browserThe signature moved to the confirmation step
Enrolment only during office hours
Open around the clock64 % of enrolments now start outside them
Four places to check your status
One page that answers itReached from the same button as everything else
One entry point was asked for in 168 of the 412 support tickets from the year before launch. It shipped in week 3 and now takes 38 % of page views.
Project Aurora · example deck

The weekly rhythm

1

Frame the week.

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.

2

Build in the open.

Small commits, deployed continuously, no staging queue to wait on. Anyone could see the portal as it stood at any hour.

3

Review on Friday.

Look at the real thing running, never at a status document. Bugs found here were fixed before the note was written, not filed.

4

Write it down.

A short note per week, describing what now works that did not before. Eleven notes are where these slides came from.

5

Cut on Monday, not Thursday.

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.

6

Leave it running.

The Friday build stayed deployed over the weekend. Anyone who wanted to look – registrar, IT, a student – could, without asking anyone.

Eleven Fridays, eleven builds left running over the weekend. Nothing was reported as done from a document.
Project Aurora · example deck
3

Part three

What we learned

Eight rules that were on the wall, and the four lessons that outlived them.

Project Aurora · example deck

Two ways of working

Worth keeping
  • Decide in the room. Async is for reporting, not for deciding. Three decisions a week, each taking ten minutes, beat a fortnight of threads.
  • Prefer the boring tool. One HTML file beat a toolchain, twice: once for the portal’s docs and once for this deck.
  • Cut, don’t defer. “Later” is a graveyard with good intentions. The two features cut in week six appear in none of the 96 tickets since launch.
  • Ship on Fridays. A deadline every week is easier to hit than one in eleven: eleven weekly builds, eleven hit.
Worth avoiding
  • Silent scope creep. New work needs old work removed, explicitly and in the same conversation.
  • Status theatre. A slide saying it works is not it working. We stopped accepting screenshots as evidence, and ran all eleven Friday reviews against the deployed build.
  • One owner per layer. It made every handover a negotiation; end-to-end ownership removed the queue.
  • Estimating in hours. Nobody was right once. “Does it fit this week” was right in 9 of the 11 weeks.
Eight rules on one sheet of paper, put up in week 2. Not one of them was edited in the nine weeks that followed.
Project Aurora · example deck

Four lessons

Ship something real every single week.

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.

One page of scope

The page from week 1 held for all eleven weeks: two features came off it in week 6, nothing was added.

Mornings for the hard part

34 % of commits landed before 10 a.m. No meeting was booked before then.

The tooling has to disappear

Infra took 70 of the 1280 commits, and none after week 3.

All four came out of the eleven weekly notes, not out of a retrospective workshop.
Project Aurora · example deck
↓ scroll · → next
📱

Turn your phone

This deck is designed for landscape. Rotate your device, or tap to dismiss.