# A room that another learner can continue

Begin with the eight-box room. Work in an isolated learner checkout or an assigned
sandbox module. Keep production framework code unchanged. Your contribution should
be small enough for a partner to understand during one session.

## Make and test a prediction

Choose one change: widen the roof slot or change the coloured wall from red to
blue. Before editing, name one expected effect in Direct and one in Combined,
which surfaces you will inspect, and one result that would disprove your account.
Keep camera, exposure, sun and other settings fixed during the comparison.

For the colour investigation, use azimuth 60 degrees and elevation 27 degrees for
both variants. First inspect whether the coloured wall receives direct sun. The
starter lighting at 240/44 produced only a subtle receiver change in the technical
rehearsal. A weak effect is an observation to keep, not a failed student. Compare a
neutral receiver; seeing the changed wall itself proves only its material changed.

## Implement the room

Create a fresh scene factory with the required group, finite camera and target,
and owned-resource disposal. Add a named second camera looking into a shaded part
of the room. Return valid nonempty bounds or allow the registry to derive them.
Shared geometry and materials must be released once each.

Register the room under a custom ID in a small host using the public framework.
Supply complete settings and your own validation; the lab validator intentionally
restricts its scene IDs. Add an exposure adjustment that preserves all current
settings rather than replacing the state with a partial patch.

Demonstrate startup, the authored cameras, a direct/combined comparison when
WebGPU is available, actual movement and disposal. On direct fallback, identify
that GI is unavailable instead of presenting a substitute as voxel lighting.

## Diagnose a seeded mistake

In an isolated copy of the host, replace one full settings update with a partial
object. Before executing, predict the failure mode or uncertainty. Reproduce the
failure, identify which required fields are missing as well as the reported error, restore the full state update and
repeat the check. Do not introduce the defect into the shared production lab.

The exact visible failure can depend on which settings are omitted. Evidence that
required fields are missing is a contract failure even if the canvas has not yet
crashed. A previous screenshot alone does not prove subsequent updates are valid.

## Practice the expansion calculation

The recipient asks for a future extension from a longest-axis reference span of
20 units to 30 units at reference resolution 80. Predict and then verify the new
requested resolution and spacing using the real quality resolver. Explain why
this is not an exact GPU-memory prediction. Provide a contrasting case that exceeds
the 256 request budget and record its explicit rejection.

Propose a bounded next region with a framed distant view. Identify separately
what needs detailed reachable geometry, distant silhouettes, navigation containment
and light-transport coverage. Do not claim collision, streaming or lighting zones
are already implemented.

## Independent transfer

After practice, the facilitator supplies a new room and expansion requirement.
Complete the first attempt individually with normal references, but without a
teacher solution. Record any help. The worked calculation above is practice and
does not by itself establish transfer.

## Hand off and teach back

Give a recipient only your submission, normal project references and run instructions.
Ask them to start it, explain ownership and add a third camera without your solution.
Record their questions and first attempt. If you provide code or an approach, log
that as level 3 assistance. Preserve both the initial and corrected attempt.

Submit the initial prediction, source changes, exact check commands and results,
backend/device labels, diagnosis, limits and recipient observations. A working
room without these records is incomplete evidence for the teaching pilot.