Demos / Side-by-side build

Mock

Build A from a blank prompt, Build B from a saved SOP pack

A scripted comparison, not a live run. Both builds try to draw Sample Kitchen 14, the same made-up U-shape studio kitchen used in the designer mock. The difference is what each one already knows on the first pass — and what is left for the next build.

Build A

Blank one-shot prompt

“Make me a small kitchen layout app for a U-shape studio kitchen.”

Cycle 1

The first pass draws a straight run with no scale and no window wall. Cabinets float. Nothing is sourced.

Cycle 2 — correction

Asked to make it a U-shape with a window on the north wall. The plan appears, but the north run is averaged to 3570 mm, hiding a 140 mm disagreement between the laser, the cut list, and the photo scale.

Cycle 3 — correction

Asked not to average conflicts. Estimate badges are still missing, and no confidence or source is shown. The peninsula catalog depth is used as if it were measured.

3 correction cycles
0 leftover SOPs
— source / confidence

What we saved

Nothing reusable. The next one-shot starts from the same blank prompt.

Build B

Saved SOP pack

Same sample kitchen, started from a human-approved SOP pack instead of a blank prompt.

SOP pack on the first pass

  • Reconcile layout measurements
    Source: human-approved skill · Confidence 0.86
  • Badge the conflict, never silent-average
    Source: prior sample project · Confidence 0.91 · Approved by a person

Cycle 1

The first pass already draws the U-shape, keeps each source on the north wall, and badges the 140 mm spread. Planning length stays at 3600 mm from the laser tape.

Cycle 2 — one correction

A label typo had three stools. Corrected to two. The catalog-versus-floor-tape depth conflict is then written up as a new SOP.

1 correction cycle
1 new SOP saved
Yes source / confidence

New SOP saved

When a catalog depth and a floor tape disagree, keep the floor tape as measured and mark the catalog as an estimate. Source: this sample run · Confidence 0.80 · Waiting on a person to approve it for the next pack.

The point of the comparison

Build B does not just finish sooner. It leaves a scored, sourced lesson that Build C can start with. That is the compounding layer — the product sitting next to the app.

Open the Sample Kitchen 14 walkthrough to see the badges in place.