v0.1.0

The raster as a PCM stream through audio effects, as an FFGL effect for Resolume Arena and Avenue, shown as SW Databend. Each frame is read the way an audio editor reads a raw file — one long stream of samples in the file's own layout, Planar or Interleaved, with a row stride that may carry padding — and real delay-line effects run on that stream in the order the editor would: echo, flanger, phaser, pitch shift, then an export back to pixels through a format that wraps or clips past full scale. An echo comes back lines down and samples across and marches on a slant with feedback; a flanger swept in stream time combs across the scanlines; a phaser smears smooth content rightwards with ringing at unity gain; in an interleaved file a delay that is not a multiple of three lands one channel's samples in another's slot and every effect rotates hue. None of it is drawn.

What is verified

By tools/verify.sh on an M4 Max, macOS 26.4, on 2026-09-24, at two rasters (320×180 and 1280×720):

• An echo of one line + 7 samples ghosts at (+1, +7); the same delay in samples lands there too; with 5 samples of padding it lands at (+1, +2), as the stride predicts; the whole picture matches the closed form with worst 0 against a 9.7×10⁻⁶ tolerance, four cases.
• Fourteen echoes at (j, 7j) with amplitude m·gʲ, worst 5.7×10⁻⁸ of signal; the tap count is the one a float-file code predicts.
• In an interleaved stream a delay of 4 puts R's echo in G one pixel on, G's in B, B's in the next pixel's R; a delay of 3 keeps every colour; a delay of L + 1 lands a line down in G; whole-picture worst 0, six cases.
• On 179 of 180 and 720 of 720 lines the flanger's ghost sits at the LFO's value at that line's stream time, 16.1 to 48.0 samples across the picture, worst 0.0014 of a 0.0015-sample tolerance (the LFO's own slope over one sample).
• With 4 and 12 all-pass stages sweeping 0..0.9 a noise picture's energy is kept to 10⁻⁵ and 4×10⁻⁴; every sample matches the serial double cascade to 1.6×10⁻⁷ / 7×10⁻⁷ against a proved bound of 4×10⁻⁵; the restart window (283, 660) is the stated one, and at a quarter of it the error is hundreds of tolerances.
• 1.4375 of full scale exports as 111/255 (8-bit Wrap), 1 (8-bit Clip), 0.21875 (16-bit Wrap), 32767/32768 (16-bit Clip) and 1 (float), and 0.6289 quantises exactly.
• Bars of period 16 come out with period 32 at Ratio ½ and 8 at Ratio 2.
• Seven perturbed plugins each fail their check, and one character changed in the shipped echo shader is caught by 17 of 25 checks.
• All 25 controls change the picture; all 8 shaders compile through glslc; oxbow loads it as SW Databend / DB01 / effect and renders 120 frames.
• Render cost at the defaults: 2.9 ms a frame at 720p, 5.8 ms at 1080p, 18.7 ms at 4K, so 4K is not real time with the phaser on. macOS figures only.
• On Windows, in Resolume Arena 7.27.1 (win-lab, Mesa llvmpipe, no GPU, 2026-09-24): this release's DLL loads from Extra Effects, registers as SW Databend / DB01 / effect, all 31 host controls match the declaration, it renders and Arena's log stays clean: 9 of the fleet gate's 9 checks, with 21 of the 26 controls moving the picture (ten under a precondition) and five inconclusive on the gate's single frames (Rate, Depth, Flanger On, Phaser Rate and Line Padding: the LFOs' phase advances every frame, so the gate's noise floor covers them; the harness measures each against its closed form). Software rendering says nothing about a GPU or about speed.

What is not

• Never loaded into Resolume on macOS.
• Each frame is its own file: the delay memories do not cross frames, only the LFOs' phase does.
• No phaser feedback, on purpose: a swept phaser with feedback has no bound of the kind its windowed GPU form rests on, and the exact alternative is a serial pass over millions of samples. The control is left out rather than shipped unproved.
• In a signed format (16-bit, float) black is full negative, so an echo sum of a black background overshoots the bottom: pastels with Wrap, black with Clip. The defaults (8-bit unsigned, Planar, Clip) were judged on twelve of Resolume's bundled demo clips.
• No OpenFX port and no presets. There is a user guide at https://stoatworks-labs.com/software/databend/guide/ and a browser demo at https://databend-demo.stoatworks-labs.com/.

The macOS build is Developer ID-signed and notarised by the release pipeline after publication; if macOS refuses a download, it predates the signing, so download it again. The Windows build is unsigned, so SmartScreen warns once.