Fluid sandbox
A full-screen ink fluid, with a scripted stroke on a deterministic solver that produces a golden frame and basic field checks.
This is an interactive visual spec on Spekkr, published by nova. It is listed with the other specs. Paste the line to an agent that can open a web address, and the agent fetches the files and builds from them.
- Spec
nova/fluid-sandbox- Version
- 0.2.2
- Published by
- nova
- Updated
- 15 Nov 2026
Build this spec
Build the spec at https://spekkr.com/nova/fluid-sandbox/start.md.
Paste this to your AI agent. It opens the address, fetches the spec's files, records your choices and builds from them. Nothing to install, no account.
Published by nova.
What does the result look like?
What can I customize?
Your choices go into the line you copy. Anything else is yours to edit, but it leaves the tested path.
What do I get?
-
spec.mdRequirements and acceptance checks.Fluid sandbox
What this builds
A full-screen fluid you smear with the pointer. Ink (dye) is advected by a velocity field, stepped by a stable solver at a fixed timestep. A scripted stroke, on a deterministic field, produces a golden frame. Basic checks cover bounds and a divergence or residual bound. This is an original sandbox, not a copy of a famous demo's look.
Read
spekkr.inputs.jsonfirst.titleis a corner label.viscosityis the diffusion amount.resolutionsizes the grid.dyeandboundschange what is drawn and how the edges behave.Out of scope
- A physically certified CFD code.
- Network or a gallery of other people's strokes in the checks.
- A 3D volume in this version.
Requirements
Each requirement has an id. Checks name the id they cover.
R1. The solver
- A fixed timestep of 1/60 s. The chosen resolution is a square grid.
- With
boundson, velocity at the edge cells is 0. Withboundsoff, the field wraps. - After 30 steps of a rest field (no stroke), the L2 of velocity is under 1e-6. After a scripted stroke, no cell is NaN or infinite, and dye stays in [0, 1] per channel.
Accept when: the rest residual, the bounds, the dye range, and the NaN check all pass at 128 and 256.
R2. A scripted stroke
fixtures/fluid-stroke.jsonis a list of pointer samples (x, y, force, colour) at listed ticks.- At tick 40 the dye field, rendered, matches
stroke-mid. At tick 180 it matchesstroke-settled, both at viscosity 0.8, 256, bounds on, seed 0. - A second splat fixture produces
splat-twoat tick 90.
Accept when: the two tapes match the goldens within the tolerance, and a second run of each hashes the same dye checksum.
R3. Files
- One command runs the field checks and the tapes, and writes the four screenshots including
clear. No network. Tests may run the solver on the CPU if a GPU is missing, as long as the checksums match.
Accept when: the command passes at 128 and 256.
R4. Inputs
- The build reads
spekkr.inputs.json. Settings label the UI, options turn features on or off, and a swap must be one of its choices. - A value outside a stated range, or a swap that is not one of its choices, is refused before the app starts its checks, with a message that names the key.
Accept when: the default inputs render, an illegal swap is refused by name, and changing a visible setting (for example the title) shows up in a later check.
Verification
A scripted stroke on a deterministic solver produces a golden frame and a dye checksum. Rest-state residual, closed bounds, dye range, and a no-NaN check are tested. A CPU reference is preferred in CI. No network.
Definition of done
- Every acceptance check above runs from one command, and all pass at the person's inputs.
- A README says how to smear ink, how to run the checks, and how to fall back to a CPU solver.
# Fluid sandbox ## What this builds A full-screen fluid you smear with the pointer. Ink (dye) is advected by a velocity field, stepped by a stable solver at a fixed timestep. A scripted stroke, on a deterministic field, produces a golden frame. Basic checks cover bounds and a divergence or residual bound. This is an original sandbox, not a copy of a famous demo's look. Read `spekkr.inputs.json` first. `title` is a corner label. `viscosity` is the diffusion amount. `resolution` sizes the grid. `dye` and `bounds` change what is drawn and how the edges behave. ## Out of scope - A physically certified CFD code. - Network or a gallery of other people's strokes in the checks. - A 3D volume in this version. ## Requirements Each requirement has an id. Checks name the id they cover. ### R1. The solver - A fixed timestep of 1/60 s. The chosen resolution is a square grid. - With `bounds` on, velocity at the edge cells is 0. With `bounds` off, the field wraps. - After 30 steps of a rest field (no stroke), the L2 of velocity is under 1e-6. After a scripted stroke, no cell is NaN or infinite, and dye stays in [0, 1] per channel. Accept when: the rest residual, the bounds, the dye range, and the NaN check all pass at 128 and 256. ### R2. A scripted stroke - `fixtures/fluid-stroke.json` is a list of pointer samples (x, y, force, colour) at listed ticks. - At tick 40 the dye field, rendered, matches `stroke-mid`. At tick 180 it matches `stroke-settled`, both at viscosity 0.8, 256, bounds on, seed 0. - A second splat fixture produces `splat-two` at tick 90. Accept when: the two tapes match the goldens within the tolerance, and a second run of each hashes the same dye checksum. ### R3. Files - One command runs the field checks and the tapes, and writes the four screenshots including `clear`. No network. Tests may run the solver on the CPU if a GPU is missing, as long as the checksums match. Accept when: the command passes at 128 and 256. ### R4. Inputs - The build reads `spekkr.inputs.json`. Settings label the UI, options turn features on or off, and a swap must be one of its choices. - A value outside a stated range, or a swap that is not one of its choices, is refused before the app starts its checks, with a message that names the key. Accept when: the default inputs render, an illegal swap is refused by name, and changing a visible setting (for example the title) shows up in a later check. ## Verification A scripted stroke on a deterministic solver produces a golden frame and a dye checksum. Rest-state residual, closed bounds, dye range, and a no-NaN check are tested. A CPU reference is preferred in CI. No network. ## Definition of done - Every acceptance check above runs from one command, and all pass at the person's inputs. - A README says how to smear ink, how to run the checks, and how to fall back to a CPU solver.
-
design.mdThe solver checks, the scripted stroke, and the dye checksums.Design
This file fixes the parts that are easy to get wrong. The language and tools are up to you, as long as the numbers here hold.
Stroke tape
Coordinates are in [0, 1] of the field. The file is the authority. Dye checksums (a 64-bit hash of the 256-grid dye, quantised to 8 bits) are
fixtures/fluid-checksums.json.Screenshot tolerance
12 levels, 96% of pixels (fluids smear). The checksum is exact on a CPU reference; a GPU run may use the looser image tolerance only, if a flag says so, but CI should run the CPU reference.
Inputs
Key Label Default Notes titleTitle Smear the ink Shown faintly in a corner. viscosityViscosity 0.8 From 0.1 to 2.0, the solver's diffusion amount. dyeLeave dye on A colour field is advected with the velocity. If off, only the velocity is drawn as lines. boundsClosed bounds on No-slip walls. If off, the field wraps. resolutionField resolution 256 The velocity grid. Tests run 128 and 256. Acceptance at the defaults
The numbers, names and fixture paths already in this file are the authority. Where a later table or a shipped fixture file disagrees with prose, the table or the fixture wins.
Fixture Role fixtures/fluid-checksums.jsonA shipped fixture. The build may move it, but then design.mdmust name the new path.fixtures/fluid-stroke.jsonA shipped fixture. The build may move it, but then design.mdmust name the new path.spekkr.inputs.jsonA shipped fixture. The build may move it, but then design.mdmust name the new path.Golden frames are compared at 1400 by 1000 CSS pixels unless a requirement names another size. Tolerance, unless a tighter figure is given above: 8 levels, 98% of pixels, with motion, cursors and clocks frozen as this file says. A build's goldens are of that build; they do not have to match a shipped reference preview pixel-for-pixel if the tool or the type is different, but they do have to match this file's layout, counts and copy.
Screenshots
These pictures are headless-Chrome captures of the working reference in
reference/preview.html, at the default inputs. They are not drawings of a product that has not been built. The first picture is the one Spekkr shows on the spec's card. A build may ship its own goldens; those goldens are of that build and do not have to match these frames pixel-for-pixel.File What it shows shots/smear.pngA scripted ink smear on the deterministic solver, viscosity 0.8. How to verify
- Run the checks at the default inputs, without a network.
- Walk through the scripted playthroughs, fixture lookups or goldens named in the requirements; each must match this file.
- Repeat once at the person's inputs when they differ from the defaults.
- A scripted stroke on a deterministic solver produces a golden frame and a dye checksum. Rest-state residual, closed bounds, dye range, and a no-NaN check are tested. A CPU reference is preferred in CI. No network.
# Design This file fixes the parts that are easy to get wrong. The language and tools are up to you, as long as the numbers here hold. ## Stroke tape Coordinates are in [0, 1] of the field. The file is the authority. Dye checksums (a 64-bit hash of the 256-grid dye, quantised to 8 bits) are `fixtures/fluid-checksums.json`. ## Screenshot tolerance 12 levels, 96% of pixels (fluids smear). The checksum is exact on a CPU reference; a GPU run may use the looser image tolerance only, if a flag says so, but CI should run the CPU reference. ## Inputs | Key | Label | Default | Notes | | --- | --- | --- | --- | | `title` | Title | Smear the ink | Shown faintly in a corner. | | `viscosity` | Viscosity | 0.8 | From 0.1 to 2.0, the solver's diffusion amount. | | `dye` | Leave dye | on | A colour field is advected with the velocity. If off, only the velocity is drawn as lines. | | `bounds` | Closed bounds | on | No-slip walls. If off, the field wraps. | | `resolution` | Field resolution | 256 | The velocity grid. Tests run 128 and 256. | ## Acceptance at the defaults The numbers, names and fixture paths already in this file are the authority. Where a later table or a shipped fixture file disagrees with prose, the table or the fixture wins. | Fixture | Role | | --- | --- | | `fixtures/fluid-checksums.json` | A shipped fixture. The build may move it, but then `design.md` must name the new path. | | `fixtures/fluid-stroke.json` | A shipped fixture. The build may move it, but then `design.md` must name the new path. | | `spekkr.inputs.json` | A shipped fixture. The build may move it, but then `design.md` must name the new path. | Golden frames are compared at 1400 by 1000 CSS pixels unless a requirement names another size. Tolerance, unless a tighter figure is given above: 8 levels, 98% of pixels, with motion, cursors and clocks frozen as this file says. A build's goldens are of that build; they do not have to match a shipped reference preview pixel-for-pixel if the tool or the type is different, but they do have to match this file's layout, counts and copy. ## Screenshots These pictures are headless-Chrome captures of the working reference in `reference/preview.html`, at the default inputs. They are not drawings of a product that has not been built. The first picture is the one Spekkr shows on the spec's card. A build may ship its own goldens; those goldens are of that build and do not have to match these frames pixel-for-pixel. | File | What it shows | | --- | --- | | `shots/smear.png` | A scripted ink smear on the deterministic solver, viscosity 0.8. | ## How to verify 1. Run the checks at the default inputs, without a network. 2. Walk through the scripted playthroughs, fixture lookups or goldens named in the requirements; each must match this file. 3. Repeat once at the person's inputs when they differ from the defaults. 4. A scripted stroke on a deterministic solver produces a golden frame and a dye checksum. Rest-state residual, closed bounds, dye range, and a no-NaN check are tested. A CPU reference is preferred in CI. No network.
-
tasks.mdBuild tasks in order, each with a check.Tasks
Work in order. Finish a task's check before starting the next.
- Set up. Set up a web app (or the stack you have chosen) and a test runner that can open a page, click, and compare screenshots. Read
spekkr.inputs.jsonfirst. Check: a blank or heading page loads, and an empty test run passes. - The solver (R1). Build what
R1. The solverasks for, to the numbers indesign.md. In particular: A fixed timestep of 1/60 s. The chosen resolution is a square grid. Check: theR1. The solveracceptance checks pass. - A scripted stroke (R2). Build what
R2. A scripted strokeasks for, to the numbers indesign.md. In particular:fixtures/fluid-stroke.jsonis a list of pointer samples (x, y, force, colour) at listed ticks. Check: theR2. A scripted strokeacceptance checks pass. - Files (R3). Build what
R3. Filesasks for, to the numbers indesign.md. In particular: One command runs the field checks and the tapes, and writes the four screenshots includingclear. No network. Tests may run the solver on the CPU if a GPU is missing, as long as the checksums match. Check: theR3. Filesacceptance checks pass. - Inputs (R4). Build what
R4. Inputsasks for, to the numbers indesign.md. In particular: The build readsspekkr.inputs.json. Settings label the UI, options turn features on or off, and a swap must be one of its choices. Check: theR4. Inputsacceptance checks pass. - Goldens. Capture the frames named in
design.mdat the defaults, with motion frozen where the file says so. Compare them within the stated tolerance. Check: the frames match, and a pixel-level check fails if a documented element is removed on purpose. - Finish. Write a README that says how to install what is needed, how to change the inputs, and how to run the checks. Run the one command from a clean folder. Check: every item under Definition of done in
spec.mdis true.
# Tasks Work in order. Finish a task's check before starting the next. 1. **Set up.** Set up a web app (or the stack you have chosen) and a test runner that can open a page, click, and compare screenshots. Read `spekkr.inputs.json` first. Check: a blank or heading page loads, and an empty test run passes. 2. **The solver (R1).** Build what `R1. The solver` asks for, to the numbers in `design.md`. In particular: A fixed timestep of 1/60 s. The chosen resolution is a square grid. Check: the `R1. The solver` acceptance checks pass. 3. **A scripted stroke (R2).** Build what `R2. A scripted stroke` asks for, to the numbers in `design.md`. In particular: `fixtures/fluid-stroke.json` is a list of pointer samples (x, y, force, colour) at listed ticks. Check: the `R2. A scripted stroke` acceptance checks pass. 4. **Files (R3).** Build what `R3. Files` asks for, to the numbers in `design.md`. In particular: One command runs the field checks and the tapes, and writes the four screenshots including `clear`. No network. Tests may run the solver on the CPU if a GPU is missing, as long as the checksums match. Check: the `R3. Files` acceptance checks pass. 5. **Inputs (R4).** Build what `R4. Inputs` asks for, to the numbers in `design.md`. In particular: The build reads `spekkr.inputs.json`. Settings label the UI, options turn features on or off, and a swap must be one of its choices. Check: the `R4. Inputs` acceptance checks pass. 6. **Goldens.** Capture the frames named in `design.md` at the defaults, with motion frozen where the file says so. Compare them within the stated tolerance. Check: the frames match, and a pixel-level check fails if a documented element is removed on purpose. 7. **Finish.** Write a README that says how to install what is needed, how to change the inputs, and how to run the checks. Run the one command from a clean folder. Check: every item under Definition of done in `spec.md` is true.
- Set up. Set up a web app (or the stack you have chosen) and a test runner that can open a page, click, and compare screenshots. Read
-
reference/preview.htmlA working reference preview of the default view, used to make the listing pictures. A build does not have to start from this file. -
reference/fonts/fonts.cssWebfont faces for the reference preview. -
reference/fonts/Geist.ttfGeist variable typeface for the reference preview. -
reference/fonts/Fraunces.ttfFraunces variable typeface for the reference preview. -
reference/fonts/Fraunces-Italic.ttfFraunces italic variable typeface for the reference preview.
Is it for me?
- Platforms
- web
- Tools
- TypeScript, WebGL
What does it need?
- Agent permissions
- shell
- Commands it runs
install a test runnerinstall a tool that compares screenshots
What changed?
- 0.2.215 Nov 2026Names what it builds.
- 0.2.114 Nov 2026Add animated loop to the listing shots.
- 0.2.014 Nov 2026Polished reference preview and new listing shots, including a short loop.
- 0.1.014 Nov 2026First version.
For agents: start.md says how to build this spec, spec.json holds everything on this page, and bundle.tar.gz holds its files.