// problem.md
Problem   Saturday slots clash, and the
          wood runs out mid-week.
Who       14 households, one sauna.
It works  no double bookings and no
when      cold Saturdays for a month.

// problem.md
Problem   listing takes too long, so
          outgrown gear sits in the attic.
Who       parents of kids under 6.
It works  a pram is listed in under
when      2 minutes.

// problem.md
Problem   stuck at 23:00, nobody to ask.
Who       people studying from notes.
It works  9 of 10 test questions get a
when      right answer with its source.

// scope.md
IN      book a slot
IN      wood rota
IN      reminder the day before
LATER   payments, chat, heat sensor
Estimate: 3 weeks, team of 3

// scope.md
IN      list with photos
IN      search by size
IN      pay safely
LATER   ratings: needs users first
LATER   delivery: pickup works for now

// scope.md
IN      upload notes (PDF, text)
IN      ask a question
IN      answer + source paragraph
LATER   voice, quizzes, flashcards
Estimate: 4 weeks, team of 3

CREATE TABLE slots (
  id         INTEGER PRIMARY KEY,
  starts_at  TIMESTAMP NOT NULL UNIQUE,
  booked_by  INTEGER REFERENCES people
);
-- API: POST /bookings  GET /rota

shop-api   listings, search, users
payments   talks to the bank
queue      order.paid  -> email seller
           order.paid  -> mark as sold
photos     stored outside the database

upload     -> split into chunks
chunk      -> embedding -> vector DB
question   -> 3 closest chunks
chunks     + question -> language model
answer     + source -> you

@Test
void sameSlotCantBeBookedTwice() {
  book(SAT_19, "Mari");
  assertThrows(SlotTaken.class,
    () -> book(SAT_19, "Jaan"));
}

test("paid order marks item sold",
  async () => {
    await pay(order);
    expect(await item.status())
      .toBe("sold");
  });

for await (const part of answer) {
  res.write(part.text);
}
res.end(sourceFooter(chunks));

AI drafted  findFreeSlots()
tests       11 passed, 1 failed
failed      28.03.2027 shifts an hour
            (summer time starts)
you         fix it, keep the test,
            explain it in the PR

spec        list an item, price in €
agent       PR #42, all tests pass
you         2 fixes, 1 question
missed      price can be negative
next        a test for it before merge

system  Answer only from the notes.
        If the notes don't say,
        reply "Your notes don't
        cover this."
tested  46 of 50 right, with source

push      main
tests     48 passed
build     docker image sauna:1.0
deploy    production
watch     bookings_failed  0

staging   payments end to end  passed
release   v1.0 -> production
alert     orders_failed above 1%
          for 5 min -> team chat

cache     repeated questions  38%
fallback  provider down -> show the
          matching notes only
limit     20 € a day, then queue

demo      village council, 10 min
review    tests, code, accepted
feedback  360, everyone to everyone
next      payments, if they ask

pitch     3 features, cost and value
decision  delivery in month 2
review    passed on the 2nd round
feedback  "explain the queue sooner"

right     46 of 50, with source
cost      0.4 cents per answer
never     grade exams, guess when
          the notes say nothing
review    passed, 360 feedback done