The L.A.B. simulator is live — 12 auto-graded labs on a prerequisite DAG

[akash-coded]

Look · Attribute · Build.

Twelve labs, eight tracks, 15–50 minutes each. Fill in a function, run the checks, read the
debrief. Start here.

python scripts/lab.py next          # what you can start right now
python scripts/lab.py run L01       # public checks
python scripts/lab.py run L01 --hidden
python scripts/lab.py status        # how far through you are

Why not just more exercises

The 22 exercises are open-ended — go
measure this on your own corpus. They are unbounded and they are how you find out what you do
not know.

Labs are the missing rung below them: bounded, auto-graded, 15–50 minutes, and they tell you
immediately whether you are right. You need both. Do a lab when you want to know if you
understand a mechanism; do an exercise when you want to know if you can find something.

What makes a lab different from a coding problem

Every lab puts a decision in front of you before it lets you type — 3–4 options, each
defensible under some constraint, and the checks encode why one is right here. That is the part
most practice material skips and the part interviews are actually about.

And every lab has hidden checks that cover what the brief deliberately did not mention.
Passing the public set and failing the hidden one is the normal experience. The gap between
them is where the brief's assumptions were doing work you did not notice.

The lifecycle is the prerequisite graph

Each track's capstone produces the artefact that stage of a product cycle hands to the next
person — a corpus spec, an ADR, an implementation with a measurement, an eval set and its noise
band, a quality gate, a cost model, a trace, a decision record.

L12 cannot be completed without T4's noise band, T5's calibrated judge, T6's cost model and
T7's trace. The DAG enforces it, because that dependency is the lesson.

Submitting

Fork, fill in starter.py, open a PR. CI posts your results and edits the same comment on each
push. It will not fail the build on red checks.

Track threads

One per track, in Q&A, prefixed [lab · T1] … [lab · T8]. Ask there rather than opening a new
thread.

If a check message is unhelpful

That is a bug. The messages are meant to be actionable — post it in the track thread and it gets
fixed.