Screen-space kaleidoscope — kaleido_order, kaleido_angle, kaleido_center_x, kaleido_center_y, kaleido_edge

Folds the finished frame into kaleido_order mirrored wedges before present. kaleido_order < 2 (default) is a passthrough; >= 2 folds (clamped to 48). kaleido_angle (radians) rotates the fold — ride it on time for a turning kaleidoscope. Works on any scene.

The fold reaches a disc, and kaleido_edge decides what happens outside it. The fold is a polar operation on a rectangular picture, so it can only reach the largest circle the frame contains — radius half the shorter side, centred on the fold axis. At 16:9 the frame’s corner sits at 2.04x that radius and 56 % of the frame lies outside the disc, so this is not corner trim: it is most of the picture, and it is a choice per preset.

kaleido_edge	What it does outside the disc
0 — falloff	Clamps the sample radius at the disc and fades out past it, so the corners are backdrop. Crops.
1 — tile (default)	Continues the picture past the disc by mirroring the source at its own borders, so the frame is filled with related content. No crop and no fade.
2 — squash	Compresses the radius asymptotically into the disc, 1:1 at the fold axis and approaching the rim at the corners. Fills the frame with no crop and no fade, at the cost of bending geometry near the frame edge — and, unlike a clamp, it pulls the disc’s interior inward too.

The default is 1, not 0. A preset that binds nothing fills its frame — the numbering and the default are separate facts here, and a live A/B on a centred figure and a border-filling field chose differently on each.

Two shipped presets carry the instructive contrast. attractor_leviathan (a centred figure that overruns the inscribed disc) binds tile; attractor_clifford (a ribbon whose wings reach past the frame edge) binds squash — and both files carry a header comment beside the binding recording the render that chose it. Read those two before choosing for a preset of your own. Several renaissance field worlds bind their own edge too — fragment_mandala, fragment_supernova, fragment_tunnel, fragment_vitrail — so the default is no longer the common case.

Which to pick is a question about your scene, and the honest answer is to look:

• A border-filling field (a fragment_field, a reaction_diffusion) is the case falloff visibly crops — a frame that filled before the fold becomes a disc with backdrop corners. tile or squash is usually what you want.
• A centred figure with real space around it may prefer falloff, whose fade reads as a vignette, if the fill would only bring in more of a figure that is already the whole subject.
• squash and tile differ in what they do to the interior: tile leaves it exactly as a clamp would, squash compresses all of it. On a figure whose scale you have already tuned, tile is the one that does not move it.

Consequences worth composing around whichever you choose:

• falloff crops. At 16:9 the disc is 56 % of the frame’s width, so a figure that filled the frame before the fold will not fill it after. Scale the figure up (scale, zoom) if you want it to reach the disc’s edge — or pick a fill treatment, which is what they exist for.
• Under falloff the corners are the backdrop’s, so bg_hue / bg_bright / bg_vignette decide what the frame’s edge looks like. Under tile and squash the corners are the scene’s, so the backdrop retreats to whatever the scene leaves transparent.
• The backdrop is underneath the fold, not inside it. It is painted first and the folded scene composites over it, so bg_vignette’s darkening stays centred on the frame however you drive kaleido_center_*, and the backdrop is never chopped into the wedges. A lit backdrop is the way to give a folded preset a frame edge that is not black — and it is worth turning bg_bright up while composing a fold, because the disc’s boundary is much easier to judge against a lit corner than a dark one. That is not a style note: sixteen confirmation captures at bg_bright = 0 once confirmed an edge treatment that two screenshots of the running app then reversed.

kaleido_center_x / kaleido_center_y place the fold axis in the frame, 0..1 in normalized screen coordinates, default 0.5 (the centre). This is what makes pan_x / pan_y and the fold usable together: pan the scene and drive the fold centre with the same expression and the rosette travels with its own axis instead of sliding off it. Both are clamped into the frame; an off-centre axis shrinks the disc on the side it moved toward, which the falloff absorbs.

kaleido_order is a stepped parameter — it is rounded to a whole number. A wedge count has to divide the circle evenly or the frame tears along a horizontal ray to the left (the fold’s angle wrap cannot absorb atan2’s branch cut at a fractional order), so the engine rounds to the nearest integer. Bind or smooth it however you like — an eased kaleido_order still eases, it just snaps at each half-integer instead of sweeping continuously. If you want a continuous kaleidoscopic motion, ride kaleido_angle; that one is fully smooth.

kaleido_edge is stepped too, and this stage is the only one with two such params. It is rounded exactly like kaleido_order above and for a stricter reason: it is a selector, so its values are names rather than amounts and there is nothing between tile and squash to be halfway to. Bind it or ease it and the picture snaps at the midpoint — 1.4 is tile, 1.6 is squash. A binding that evaluates to a non-finite value falls back to the default rather than to a bound, so a broken expression lands on tile, not on falloff.

Two things follow that are easy to get wrong:

• [smoothing] on kaleido_edge buys you nothing. It delays the snap; it does not soften it. If you want a treatment to change during a track, accept that it is a cut.
• A dissolve between two presets with different treatments blends correctly, and that is a different mechanism — a dissolve cross-fades two finished frames, so it is blending pictures rather than interpolating the selector. Two presets that disagree about the edge cross-fade cleanly; one preset easing across the boundary does not.