Spell port · firewhip · from the 101-spell sandbox

Firewhip

A spell from the reference sandbox, rebuilt in Godot. Engine output.

Eight frames of the firewhip spell
Render Eight frames of the cast: wind-up, release, the lash travelling, the crack, then the fall. This is the reference sandbox's firewhip rebuilt in Godot 4.7.

What this proves

Shader #include works — with a negative control

The sandbox shares GLSL chunks via #include. If Godot could not do that, every shader would need its chunks transcribed by hand — the difference between a port and a rewrite. This was verified rather than assumed.

Include probe passing
ProbeWith the include — the chunk's function is reachable and renders.
Include control failing
ControlWithout the include — fails with the exact predicted error. The test is not passing vacuously.

One-card campfire — and its failure

The single-card campfire from four azimuths
Render One billboarded card, four camera angles. Panel 2 is blank, and panels 3–4 collapse.

Measured: the card collapses at two angles

CameraFire pixelsvs az000
0°410,137—
45°435,261+6%
90°15,660−96%
135°7,549−98%

The single card loses 96–98% of its pixels at 90° and 135°. A yaw billboard should always face the camera and never do this, so the billboard is either not applied or the card is being culled. Not diagnosed. Reported because it is the reason panel 2 is empty and panels 3–4 are nearly empty.

Campfire — the craft stack

Real logs, rocks, coals, ground pool, embers. The layer the judges named as the worst failure — a dead-straight base — was fixed here.

The campfire from four azimuths
Render Four camera angles. Crossed logs, rock ring, coals, ground pool, embers.

Before and after the stack

Before: fire with no fuel bed
Before No fuel bed. The card's bottom edge cuts a dead-straight line across the fire, and the base is a black void.
After: fire grounded on logs
After The fire now sits on something. The base is occluded by fuel rather than cut by the ground.

One full cycle

16 poses across one cycle
Render 16 poses. The logs hold still while the flame changes shape above them — which is what makes it read as a fire on fuel.

Measured

MeasureResultTarget
Straightest span at the base4.3%<8%
… before the logs7.5%—
Base bottom-row spread156 px IQR>50
Ground pool reach0.52 m~0.5 m
Shape change per frame24.2%>4%
Frozen control0.9%~0

The base is no longer a straight line: the bottom row varies across a 156 px spread instead of clustering on one row. And 24.2% shape change against a 0.9% frozen control is 27× separation — the instrument is measuring the fire, not the embers.

What the stack added

Three honest problems

1. Too much orange

Measured against the approved sheet's own palette, the build runs +14.1 points heavy on orange and 7–8 points light on yellow.

RED 38%
ORANGE 26% → 40%
YELLOW 23% → 15%
WHITE 13% → 8%

The build is warmer and less bright than the target. Fixable by shifting the ramp, but it is a real deviation and it is why the fire looks slightly heavier than the approved sheet.

2. The second sheet disappears at one angle

The under-sheet sits at 30°, so at a camera angle near 135° it goes nearly edge-on and falls below the alpha scissor — its contribution drops to 0.38%, effectively nothing. The loop-hider is missing for about a quarter of the orbit, which is the one layer whose entire job is to be present at every angle.

3. The pool's tail is short

Its brightest 10% is about 2× darker than the target's (25–28 vs 66), so the falloff ends more abruptly than the reference. The reach is right; the tail is not.

Your call

It is grounded and lit now. Does it read as fire?

1. Is this the right direction?

YesNo

2. Fix the orange/yellow balance and the vanishing sheet?

YesNo

3. Anything else wrong?

Reply in chat
The frozen control — proof the numbers are real
The frozen control
Control Same build with the pose pinned. Scores 0.9% against the live 24.2%, so the instrument is reading the fire and not noise.