Step inside a full cabin. Sit, look around, open a book, or visit the river.
Three things to try +
Step inside. Use the menu to sit on the sofa, or visit the porch and river.
Hold and drag the cup or throw. Release to let gravity take over. Reset visit restores everything.
Tap the book to hold it. Open the cover, then pull a page corner across the spine. Let go drops the book.
An evolving prototype with scanned furniture, simulated objects and fabric. Physical phone and VR-headset tests are still pending.
IN THE WOODLAND RETREAT
Make yourself comfortable.
The scene fills the screen. The little menu holds the extra controls.
On a computer
Drag empty space or scroll a trackpad to look. W A S D walks; Q / E turns. Space jumps and C crouches. You can still look while sitting or lying down.
Mouse look follows the pointer where the browser allows it. If unavailable, drag empty space or use Q/E. Escape releases mouse look. Hold an object to drag it; scrolling while holding changes its distance.
On a phone
Use the left joystick to walk. Drag the scene with your other finger to look. Hold and drag a book, cup or blanket corner, then release. Landscape gives you more room.
The phone layout was checked in a resized desktop browser. A real phone test is still needed; that is a different check.
With a VR headset
Open this site’s HTTPS link in a compatible headset browser. Enter VR appears in the menu when immersive WebXR is available.
Hold the trigger over an object to grab, then release. Hold over clear ground and release on the ring to move. Small controls below your view operate the lights, return you home and exit VR.
Headset testing is pending. This is implemented support, not a verified claim that every headset works. Camera-driven jumping and resting are disabled in VR; your own head controls the view.
THE HONEST REVIEW
The good parts. The rough edges.
These are judgments from a reviewer who knew which model made each room. Later rooms also had more instructions, feedback, assets and time. This is not a blind or controlled test.
Show the room grades +
01 · SPARK, ORIGINAL DELIVERY38/100
Too much was broken.
Some controls worked, but dragging failed, geometry was misplaced and opening it needed help. The repaired copy does not raise the original score.
02 · ASTRA, ORIGINAL DELIVERY89/100
A strong decorating demo.
Selection, colour, movement and rotation worked well. Small screens needed scrolling. This was a quick-demo grade, not a commercial-game grade.
03 · HOLLYHOCK, REVISED BRIEF63/100
Warmer, but still a dollhouse.
Familiar objects and colours add character. The stylized geometry, cutaway setting and limited interaction miss the requested immersive realism.
04 · RETREAT, STRICT GAME RUBRIC66/100
Progress, with real limits.
Walking, grabbing, a hinged book and flexible pages are a substantial step. Contact quality, polish and real-device verification still fall short of AA/AAA expectations.
Two different standards: the first three scores use the demo rubric below. Room four uses the tougher game rubric requested later. Its 66 cannot be ranked directly against the original Astra’s 89.
How the original demo scores were counted
Working behaviour / brief / beginner use / appearance / delivery
Room
Working /40
Brief /20
Beginner /20
Visual /15
Delivery /5
Spark
18
12
5
3
0
Astra original
38
20
15
12
4
Hollyhock
29
9
12
10
3
The retreat’s tougher rubric
Visual quality17/25
Scanned furniture and a complete interior help. The landscape, small props and lighting still lack the consistency of a finished game.
Controls & interactions20/25
Seated drag and scroll, joystick movement, lamps, grabbing and page turning were exercised in the browser. Actual multitouch and headset handling remain unverified.
Physical behaviour10/20
Objects have weight and collision; the throw and pages bend. Paper contacts and fabric friction are approximations. There is no full fabric-to-object force exchange.
Performance & stability9/15
Contact calculations were optimized after a slowdown. Short desktop samples are not a sustained phone or headset performance test.
Beginner experience & delivery10/15
One collection and short tutorials help. Small objects need precision; headset testing and independent visitor access still need verification.
AA and AAA are production categories, not standardized quality certificates. Here, 90+ would require exceptional visual consistency, reliable interactions and sustained performance on the promised devices. Unknown results do not receive full credit. This build has not met that bar.
Grade the guides too +
FIRST GUIDE66 / 100
Readable, but mostly a styled essay. It lacked room pictures, clear visual evidence and a dependable single entry point.
Actual room pictures, working room links and short tutorials are together. Desktop and phone-width layouts were checked. The page is still long, and independent visitor access needs verification.
Clarity 18/20 · accuracy 23/25 · visual explanation 18/20 · beginner use 17/20 · delivery 8/15. This remains a self-review.
WHAT CHANGED BETWEEN THE ATTEMPTS?
The settings are only part of the story.
Different model
Spark and Astra are different AI models. Both are AI, just as free chatbots are AI too.
Different effort
Light requests lighter reasoning. Ultra requests deeper work and can use AI helpers. More effort gives opportunities to improve; it does not guarantee a good result.
Fast is separate
Fast is a processing-speed option. It doesn’t mean a large project finishes immediately or turns a sketch into a finished game.
The first two started with the same decorating request. The later cabins had a richer brief, repeated corrections and imported models and textures. We cannot isolate which setting caused each difference. Models & effort · Fast mode