DanceRudiments

By Kieran Simkin · DanceFlow ecosystem · Vector logo and usage guide.

Deterministic rhythmic position functions with C++, Python, TypeScript/WASM and C# bindings. https://kieransimkin.co.uk/

By Kieran Simkin — My Songs · Part of the DanceFlow motion workflow.

Visualizer · Language bindings · C++ · Python · C# / .NET · TypeScript / JavaScript

Before every new workflow

Check the latest stable official release and the actual installed/imported version before starting; do not infer currency from a directory name or a development checkout. Read the current release notes, tagged capability documentation and relevant Potential problems. Use the DanceFlow release preflight and complete capability references.

People and agents should consider the full ecosystem, use all capabilities that materially serve the task, compose supported files/APIs, and improve a demonstrated reusable gap in its canonical tool. Record why no worthwhile change is justified when that is the result. Keep independent tool identities, compatibility, source clocks, privacy, model/download costs and publication/deployment permissions intact. Store weights and model-specific caches in the owning project's ignored models/.

Find your route

▶ Try the pattern visualizer · 3D dance examples · Which package should I use? · Included rudiments · Default collections · Dancefloor 06 · BPM-synchronised N-dimensional recordings and comparison harness · Rigify character motion · Curve and event authoring · Language bindings · Harness

▶ Try the pattern visualizer

▶ Open the movement visualiser

Download the self-contained HTML for offline use

Save the file and open it in a browser. No install, server, CDN, soundfont or separate WASM file is required. The complete 1,731-movement catalogue includes a shared-tempo beat player: 305 scores, including all 68 dance-music studies, the full four-bar Amen break, four-to-the-floor, garage, drill, grime, jungle, house, techno, Groove MIDI excerpts and the earlier event/sticking studies.

Choose a beat, press Play, and change BPM to retime the sound and animations together. Compare any movements, or use Only matching movements and Use this movement’s beat. Local .mid import and MIDI export are included. Sound is synthesised from note events, not the original Amen recording.

The downloadable file is the checked-in snapshot; the Pattern visualizer workflow also rebuilds and tests a fresh artifact from every pushed native catalogue. The Release demo workflow publishes the complete tagged site, including the 3D examples below, when Pages is enabled. Main-branch visualizer builds do not replace that released site. Beat player, rebuilding, release downloads and optional live hosting.

Captured from the running C++/WebAssembly visualizer at beat 2.75. Audio was not started during this screenshot.

3D dance examples

The release build includes a separate meshes/ gallery beside the pattern visualizer. The production character is Kandi Boom from Bass Ben & Friends, using the real checked-in FBX, 24-bone rig and PBR textures. Two pinned optional source-motion packages can drive her through an explicit source-rig → Kandi retarget adapter:

The source recordings preserve their own skeleton profiles, original timing and calibration frames. The retained converter calibration T-pose establishes a per-control retarget reference pose: Kandi keeps her FBX joint positions and proportions, while each mapped continuation bone is oriented to the corresponding CMU calibration direction before captured rotation deltas are applied. This is especially important for the arms because Kandi's authored bind pose is a steep A-pose while the CMU converter frame is near-T. Kandi's external PBR maps also inherit the FBX diffuse map's image-Y orientation, UV channel, wrapping, scale and translation instead of assuming glTF texture coordinates. The captures have no verified musical beat/downbeat annotations and are not presented as beat-matched to a song. Loading a performance fetches only its optional motion package plus the hash-pinned Kandi target assets; switching performances unloads the previous recording and target scene. Play, pause, seek, restart and preview speed continue to use the existing native C++/WASM sampler.

Playback-rate interpretations for rigged motion

The Kandi Boom gallery offers half time (0.5×), original time (1×) and double time (2×) in addition to the 0.75× and 1.5× preview rates. The TypeScript motionTimeRate and motionSourcePipAtTime helpers convert elapsed presentation time to an original source pip. The motion stays on the native 64-pip source clock, with no implicit audio synchronisation, source-BPM correction, wrapping or resampling.

The gallery also supports local StemLab rhythm-analysis review. After loading Kandi, import the corresponding stemlab.rig-motion-analysis JSON from the complete salsa/ballet package. Source hashes and channel identity are validated before estimated motion beats and selectable loop-preview boundaries appear on the existing timeline. The analysis file stays in the browser; no musical beat/downbeat approval, audio synchronisation or seam/foot-contact repair is implied. See mesh-demo rhythm review.

For newly imported animation, the optional source-first StemLab loop job analyses original BVH frames or evaluated Blender Action captures before resampling to the native 64-pip grid. It ranks source-time beat and loop hypotheses, then creates independent, review-only RigMotion previews; it never changes the full source motions or adds assets to the default catalogue. Run python tools/analyze_source_motion.py original.bvh --output-dir .cache/source-loops/action or analyse all reviewed, locally cached source archives with python tools/analyze_motion_collection.py --sources cmu_dance. See the source-first analysis guide. Musical beat-1, audio alignment, rig deformation and repeat seams still require separate checks.

World/contact timing and bounded loop finishing

Source analysis now evaluates portable skeletal captures in world space before compilation. Root-relative joint motion, body-height changes and inferred foot plants provide separate evidence alongside control-space accents. The default --tempo-strategy consensus retains competing tempo interpretations and abstains when supported families disagree. Unevaluated constrained rigs are explicitly unsupported for this world-space pass.

Use --repair-loops to create separate, bounded corrected derivatives. Pose and tangent closure, retained activity, correction budgets and available foot-contact constraints must pass. An explicit --repair-options JSON can request bounded root stabilization or in-place root travel. Every accepted derivative is checked again after compilation to the existing half-open 64-pip native recording; --verify-native additionally checks C++ playback. Unchanged source Actions and raw loop candidates keep their own identities and catalogues.

python tools/analyze_source_motion.py original.bvh \
  --output-dir .cache/motion-review/action \
  --tempo-strategy consensus --repair-loops --max-repairs 6 --verify-native

The same options work for reviewed archive collections, extracted Blender Action batches and the two-source benchmark. See the complete workflow and measured CMU results, world-space evidence and loop correction limits.

The source job exports observed motion-events.json for compatible StemLab audio/motion alignment. Exact audio/capture hashes and an explicit clock map bind offset, playback-rate and variable-tempo proposals to their sources. Pass a result with tools/build_song_loop_preview.py --alignment-report alignment.json to compare candidates, adjust phase and save a separate timing decision. The browser requires the exact local audio master before saving; beat one and meter remain unconfirmed unless entered. See timing review.

The local collection audition compares a complete reviewed source collection with selected StemLab example songs. It loads bounded, hash-verified motion assets on demand, filters by source Action and plays supported calibrated sources on Kandi through the same native sampler. Original source-speed previews remain available even when no full Action clears the musical-fit gate. An explicit private-audio flag can include the selected example masters; the normal generated artifact remains audio-free. “Lots of movement” ranks measured body-scale-normalised joint activity ahead of tempo/seam preference, while retaining the actual speed ratio and all review limits. Quiet phrases are no longer rewarded simply for returning to a pose.

Development audition: actual native poses and Kandi skin, not an illustration or artist-approved musical match. Motion: CMU / NSF EIA-0196217; BVH: cgspeed; Kandi Boom: Bass Ben & Friends.

Kandi's mesh gallery accepts a local corrected .motion.json and matching .repair.json, verifies their identity and final compiled checks, then uses the same native player and calibrated source-to-Kandi adapter. Explicit repeat, compare-original and restore controls make the derivative reviewable on the actual skin. Numerical closure, functional mesh QA and a user's musical decision remain distinct results.

These changes require the companion StemLab world/event APIs. Source integration CI pins their exact development commit; published package/release dependencies continue to follow the existing stable-release policy. CI uses pytest for both function-style and unittest-style motion tests, builds the native runtime, and checks the timing-review JSON against StemLab's validator. No release or Pages deployment follows automatically from the analysis commands.

Kandi's canonical FBX contains vertices with more than four source influences. The source remains untouched and fully audited; the WebGL renderer uses an explicit, normalised four-weight runtime approximation and records that limit in the build/browser evidence. The target is a custom Kandi rig, not a Rigify rig.

The release workflow browser-tests both performances on the real Kandi mesh, attaches gallery/complete-site archives and deploys the existing visualizer and gallery together when Pages is enabled. See building and validating the Kandi Boom motion demo and the release workflow guide.

DanceRudiments is a C++17 library of deterministic rhythmic position functions. One integer pip is 1/64 of a beat. Built-in and custom patterns each keep their own loop period. Every legacy pattern sampler wraps positive or negative input into its own loop before sampling.

Outputs are dimensionless offsets, normally in [-1, 1]. The caller chooses pixels, CSS units, metres, or another scale. There is deliberately no time interpolation in the public API: renderers advance with integer pips and get one exact sample per pip.

Which package should I use?

GitHub automatically adds “Source code” ZIP and tar.gz links to every release; those are repository snapshots, not precompiled packages. Python wheels belong on PyPI, the C# package on NuGet.org (mirrored to GitHub Packages), and the TypeScript/WASM package on npm, while GitHub release attachments include C++ archives and the separately generated HTML demos, screenshots and checksums.

Included rudiments

The general motions are bounce, sway, circle, figure_eight, step_touch, box_step, helix, and clay_background.

The first drum-derived set is single_stroke_roll, multiple_bounce_roll, double_stroke_roll, single_paradiddle, flam, drag, and five_stroke_roll. The first six mirror the fundamentals in Vic Firth's Tier One learning sequence and collectively cover the roll, diddle, flam, and drag families in the Percussive Arts Society's 40 International Drum Rudiments. A right-hand stroke moves right, a left-hand stroke moves left, and both rise slightly. Every stroke is a raised-cosine (sin²) gesture: zero displacement at the start, smooth attack, a rounded peak, and smooth decay back to zero before the next motion. Accents use greater amplitude rather than an instantaneous position jump.

Source references: Percussive Arts Society International Drum Rudiments and Vic Firth 40 Essential Rudiments, accessed 28 September 2026.

clay_background preserves the translation waypoints from the Clay/Stars ks-particle-dance background treatment as a normalized 256-pip lookup table. The source effect's opacity, rotation, and scale are not position offsets, so they are intentionally excluded.

Default collections

The default catalogue contains 1,731 movements: 15 original routines, 28 Initial 01, 24 Expansion 02, 256 Motion Atlas, 320 Continuum, 144 Club Rhythms 05 movements, and 944 Dancefloor 06 additions. Use the ordinary sample(name, pip_count) API without loading a separate pack.

Club Rhythms 05 interprets 36 beat patterns, including the four-bar Amen, four-to-the-floor, UK garage, drill, grime, jungle, house and techno, using four different movement mappings. Its offline demo includes a synthesised drum player, exact onset grid, and C++/WASM movement animations.

python -m pip install .
python tools/build_club_demo.py --output dist/club-demo.html

See the default library guide for all collections. Release demos enumerate the actual tagged native catalogue and also render the focused Club Rhythms page. No original recordings are bundled.

Dancefloor 06

The default catalogue now contains 1,731 movements, including 944 new Dancefloor 06 entries. All 68 beat studies have sixteen movement interpretations. The previous Club 05 scores and their original 144 movements remain unchanged. Research, rhythm-specific notes, and exact timing semantics are in the Dancefloor guide and complete mapping catalogue.

python tools/build_dancefloor_collection.py --check
python tools/build_default_catalogue.py --check

The release workflow builds the complete native catalogue demo plus an offline, audio-enabled 68-rhythm / 16-mapping comparison page. All live motion comes from C++/WASM.

BPM-synchronised N-dimensional recordings and comparison harness

The additive dancerudiments.recording format stores any number of named float64 dimensions at 64 pips per beat, with a mandatory BPM estimate or confirmed value and separate beat-phase provenance. SampledRecording provides native non-looping C++/Python playback, including captures beyond the old 65,535-pip loop length; existing XYZ patterns and their wrapping rules are unchanged.

from dancerudiments_authoring import load_recording
from dancerudiments_authoring.harness import build_recording_harness
r = load_recording("output/recordings/audio_recording.json")
values = r.to_native().sample(r.clock.start_pip)  # every dimension, real C++
build_recording_harness([r], "output/index.html")

The exported page uses the same packaged visualizer assets as catalogue/release pages, with native WASM sampling, shared transport, four-card comparison and review import/export. Recording mode adds arbitrary-axis X/Y/Z selection, locked BPM/origin and non-looping completion. The tiny generic WASM sampler is bundled; no compiler or CDN is needed at export time. StemLab's record-rudiment and generate-batch produce individual or varied stem/filter combinations using this API without copying it. Music is not bundled unless explicitly requested.

See recording format, harness, native API and compatibility. Build the updated source to use this unreleased API; earlier releases lack it.

Rigify character motion

Typed rig motions now bind the native N-dimensional recording sampler to Blender Rigify character controls. The pinned Blender 5.2.2 LTS catalogue covers all eight built-in metarig presets and 28 rig component types, with numbered channel manifests generated from actual rigs. Each channel identifies its bone, property, component, physical units, range, neutral value and interpolation. The catalogue guide records exact counts and any component context; custom/generated characters can export their own profiles.

Model types use a hierarchy such as syn://type/danceflow/motion/rigify/metarig/human and syn://type/danceflow/motion/rigify/component/limbs/arm. synny:// is an equivalent spelling, not another type. Channel numbers persist across explicit profile extensions; profile hashes prevent applying a recording to an incompatible skeleton, rest basis or control layout.

python tools/rig_motion.py catalogue
python tools/rig_motion.py profile human --output generated/human.profile.json
python tools/rig_motion.py channels generated/human.profile.json --output generated/human.channels.csv
python examples/rig-motion/quickstart.py --output generated/rig-motion --native

The installed CLI is dancerudiments-rig. The optional Blender bridge imports local BVH, FBX and glTF/GLB, captures animation already retargeted/baked onto Rigify controls, and applies or bakes motion using the actual native sampler. dancerudiments.rig-motion combines an exact profile with the unchanged normalized Recording format: float64 [-1,1], 64 pips/beat, explicit bounds and non-looping playback. Scalar controls interpolate linearly during authoring, quaternion groups use shortest-path SLERP, and discrete switches hold their values.

Source BPM/phase or measured beat anchors control compilation. A separate target beat map permits variable-tempo playback, absolute seeking and explicit half/double-time without accumulating frame-clock drift. The existing XYZ catalogue and C++ playback engine remain the movement authority.

Read the complete rig-motion guide for capture, compile, native playback, Blender baking, actual model mappings and compatibility. The motion-source research records motion libraries and their format/licence/access constraints. The Blender import guide covers multiple Actions, source-skeleton reconstruction and explicit Rigify FK mappings. Musical beat detection and foot-contact/transition repair remain separate authoring steps. The dedicated Rigify motion CI checks native playback and an actual skinned mesh using the pinned Blender runtime. These APIs require this development revision until a new stable package is released.

Optional model-specific motion packages

Large character recordings are independent optional packages. They are not added to the 1,731 default rudiments, fetched at startup or loaded site-wide. Each package shares one exact numbered rig profile and stores individual compressed recordings with hashes, timing, source attribution and licence. Only the selected page or scene loads its requested clips.

The new TypeScript ./motion-packages subpath provides a lazy, scene-owned loader and the existing native C++/WASM recording accessor. The page lifecycle example follows the host's integer beat clock, keeps one recording chunk loaded, handles seeking and releases requests/memory on route exit. Rig application remains the consuming scene's responsibility; the existing XYZ visualizer does not become a mesh renderer.

Python provides MotionPackage, MotionPackageBuilder and the explicit dancerudiments-motion-pack install/inspect/export CLI. Offline BVH import uses the optional motion-import extra; the base package does not acquire NumPy or Blender dependencies. Public source plans and reproducible import tools produce external data under ignored motion-packs/; the base npm/Python/native packages exclude those assets.

Reproducible source recipes cover ACCAD human actions, 100STYLE locomotion, Lifelike Agility dog captures, Space-XR dance/movement recordings, a deduplicated CMU dance subset and all 14 Actions of Quaternius' CC0 robot. The source plan index records separate data terms and acquisition commands. A collection can be large without increasing the base package or its startup work.

python -m pip install '.[motion-import]'
python tools/import_motion_collection.py download --sources accad
python tools/import_motion_collection.py build --sources accad
python tools/motion_package.py list --root ./motion-packs

Complete source actions can span contiguous bounded chunks. Unknown musical timing is labelled time-only: a 60 BPM seconds coordinate is not an observed music tempo. Supply source tempo and beat phase or measured beat anchors before claiming musical alignment. The optional-package guide describes formats, hosting, exact profile compatibility, source conventions, native playback and separate package licences.

Review, finish and export imported loops

The local loop-review workspace joins original Action captures, StemLab timing decisions and the actual Kandi mesh. It supports source in/out and beat-count edits, bounded re-finishing, phase/manual timing anchors, audio-clock audition, original/corrected comparison, separate seam/feet/skin verdicts, saved drafts and review reload. Changed motion, timing or target identity invalidates the earlier approval. Explicit source bone maps and saved calibration poses support evaluated portable humanoid Actions beyond the two public demo scenes; unevaluated constrained controls remain unsupported. For imports with a neutral object parent, bounded foot stabilization accepts an explicit moving ancestor such as root_bone="Hips" when the saved hierarchy supports it. Correction uses the actual parent/rest transform and preserves the original profile and all local/world correction limits.

python tools/review_motion_loop.py prepare --help documents the source-checkout workflow; serve WORKSPACE opens its local Kandi panel. Accepted results export through MotionPackageBuilder with the exact native samples, source provenance and bounded audio mapping. load_reviewed_loop replays that mapping through the existing native player. Audio, original capture samples and target assets are excluded from the optional package. Automated test reviews remain explicitly unapproved. These are development capabilities pending stable publication.

Source-first CI requires native/StemLab checks and Python/JavaScript source-rig parity, then builds a deterministic full-character workspace with tools/build_loop_review_test_fixture.py. The dedicated tools/test_loop_review_browser.py gate exercises finishing, draft/review reload, audio-clock playback, stale writes, package ZIP/reload and resource disposal on desktop and mobile. Its movement, WAV and timing annotations are synthetic; saved decisions and exports remain labelled automated test fixtures.

Curve and event authoring

The optional Python authoring tools compile LFOs, segmented curves, sampled waveforms, seeded repeating randomness and rational-beat event/gesture scores into JSON packs or C++17 headers. All movement playback remains C++, including Python and TypeScript/WASM consumers. The original fifteen motion tables are preserved; additional custom packs use independent PatternLibrary objects rather than global overrides.

python tools/compile_patterns.py compile examples/patterns/starter.json --json generated/starter.json --cpp generated/starter.hpp

The compiler runs from a checkout without third-party Python dependencies or a native build. See the authoring and native API guide for curve formats, event anchoring, loop validation, generated functions, Python playback, WASM loading, provenance and testing. These are original infrastructure examples, not imported third-party preset banks.

Language bindings

All four language interfaces sample the same C++ runtime. Choose the interface, not another movement implementation. The examples below are checked into examples/bindings/ so they can be built and tested.

One beat is 64 pips. Supply a signed 32-bit integer, not milliseconds or a floating-point beat number. Every legacy pattern sampler wraps negative as well as positive positions into its own loop. Offsets are {x, y, z}, dimensionless and bounded to [-1, 1]; choose the pixel/metre scale in your application. There is no automatic interpolation, BPM detection or audio playback inside a binding.

For an absolute playback time, use floor((playback_seconds - beat_zero_seconds) * BPM / 60 * 64). Use floor, not integer truncation, for negative positions. Do not add a rounded pip increment on every frame: that accumulates drift. For a tempo map, integrate the elapsed beats; do not multiply the entire song time by the latest tempo. For long-running playback, wrap at the selected movement's period before passing an integer that could overflow the API. Different movements have different periods.

Checkout versus published version: these examples describe the current repository. pip install / npm install retrieve the latest published package, which may predate this catalogue. Inspect catalogue() / catalogue rather than assuming a fixed count. Build the checkout when trying unreleased patterns such as the expanded Amen interpretations.

C++ binding

Requirements: CMake 3.20+ and a C++17 compiler. Python is not needed to build or play the committed movement tables. Build and install the core from the repository root; choose a prefix writable by your user:

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX="$PWD/install"
cmake --build build --config Release
ctest --test-dir build -C Release --output-on-failure
cmake --install build --config Release

In PowerShell, use an absolute prefix such as -DCMAKE_INSTALL_PREFIX="${PWD}/install". Multi-configuration generators put executables under Release/; the --config Release arguments handle that.

In your application's CMakeLists.txt, create the executable before linking it:

cmake_minimum_required(VERSION 3.20)
project(MyAnimation LANGUAGES CXX)
find_package(DanceRudiments CONFIG REQUIRED)
add_executable(my_animation main.cpp)
target_link_libraries(my_animation PRIVATE DanceRudiments::DanceRudiments)

#include <dancerudiments/dance_rudiments.hpp>
#include <dancerudiments/sampled_pattern.hpp>
#include <iostream>

int main() {
    const auto& items = dancerudiments::catalogue();
    std::cout << items.size() << " movements\n";

    const auto position = dancerudiments::sample("beat_amen_four_bar_bounce", 48);
    std::cout << position.x << ", " << position.y << ", " << position.z << '\n';

    // Negative pips wrap; the original named C++ functions are also retained.
    const auto previous = dancerudiments::sample("circle", -1);
    const auto direct = dancerudiments::circle(-1);
    return previous.x == direct.x ? 0 : 1;
}

Configure the consumer with -DCMAKE_PREFIX_PATH=/absolute/path/to/install. Alternatively, point that prefix at an extracted, compatible native release archive. A static archive is specific to its OS, architecture, compiler and runtime-library settings; use a source build or the Conan recipe for another combination. The exported target handles the Windows DanceRudimentsCore.lib filename, so do not hard-code an archive name.

Run the complete installed-package example (it also checks custom-table ownership):

cmake -S examples/bindings/cpp -B build-example -DCMAKE_PREFIX_PATH="$PWD/install"
cmake --build build-example --config Release
ctest --test-dir build-example -C Release --output-on-failure

Custom data: SampledPattern copies and validates an std::vector<Offset3>; PatternLibrary owns its custom patterns and also exposes all defaults. Construct it once, not on every frame. catalogue() metadata contains string_views, so a custom library must outlive those views. Unknown names and invalid custom tables raise C++ exceptions. See the complete C++ source and authoring guide.

Python binding

Install a released package, or build the checkout to use unreleased additions:

python -m pip install --upgrade dancerudiments
# Alternative, from this checkout (requires a C++17 compiler):
python -m pip install .

A matching prebuilt wheel does not require a local C++ compiler. A source build does; pip's isolated build installs the declared scikit-build-core/pybind11 build dependencies. Use the same Python interpreter for -m pip and your application. The package metadata supports Python 3.9+; release wheel availability is narrower than every possible Python/OS/CPU combination.

Python wheels retain the native extension, full catalogue, authoring APIs, binding examples and illustrated documentation. C++ headers, static libraries and CMake package files belong to the separate C++ distribution. The Python source archive retains rebuild inputs and source rig metadata; large generated browser pages and Kandi's mesh/texture payloads stay in the repository and demo downloads. See Python distribution boundaries.

import dancerudiments as d

items = d.catalogue()
info = next(p for p in items if p["name"] == "beat_amen_four_bar_bounce")
print(len(items), "movements")
print(info["period_pips"] / d.PIPS_PER_BEAT, "beats per loop")

position = d.sample(info["name"], 48)
print(position.x, position.y, position.z)
print(position.as_tuple())

# Original named functions are also exposed. Generated names use sample().
assert d.circle(-1).as_tuple() == d.sample("circle", -1).as_tuple()

Offset3 exposes read-only x, y, z and as_tuple(); it is not a dictionary or NumPy array. catalogue() returns dictionaries with name, description, period_pips and integer dimensions. Sample values come from C++, not a Python reimplementation. Native invalid-argument errors are exposed as ValueError; use integers and catch errors for untrusted names or data.

Run the complete example, including validated time-to-pip conversion:

python examples/bindings/python/quickstart.py
# Export the complete loop as pip/beat/XYZ rows:
python examples/bindings/python/sample_motion.py --name beat_amen_four_bar_bounce --csv amen.csv

Author and load a private pattern: compilation is a Python build-time step; .to_native() transfers the result to a C++ bank for runtime playback.

python tools/compile_patterns.py compile examples/bindings/pulse.score.json --json generated/tutorial.json --cpp generated/tutorial.hpp
python examples/bindings/python/quickstart.py --pack generated/tutorial.json

from dancerudiments_authoring import load_pack

compiled = load_pack("generated/tutorial.json")
library = compiled.to_native()
print(library.sample("tutorial_pulse", 0).as_tuple())
print(library.sample("circle", -1).as_tuple())  # Defaults remain available.

The generated C++ header can instead be included by a native application. Existing collections are already defaults: loading initial_pack() or dancefloor_pack() is for inspection/subsetting, not a prerequisite for ordinary sampling. An exact default reload is idempotent; changing a default under the same name is rejected. Give an edited movement a new identifier.

C# / .NET binding

The DanceRudiments NuGet package wraps the original C++ library through a versioned, exception-safe C ABI. It targets .NET 8+ desktop/server applications; CI tests .NET 8 and .NET 10. Native Windows, Linux and macOS libraries are included for x64 and ARM64, with automatic NuGet/.NET runtime selection. This does not claim support for .NET Framework, Unity, mobile or browser-WASM hosts.

After the first release containing this binding has been published:

dotnet add package DanceRudiments

using DanceRudiments;

Console.WriteLine(Rudiments.NativeVersion);
var amen = Rudiments.Catalogue.Single(p => p.Name == "beat_amen_four_bar_bounce");
int pip = Rudiments.PipAtTime(1.25, bpm: 140, periodPips: amen.PeriodPips);
Offset3 position = Rudiments.Sample(amen.Name, pip);
Console.WriteLine($"{position.X}, {position.Y}, {position.Z}");

// Full-loop export: one interop call; every sample is still calculated in C++.
Offset3[] loop = Rudiments.SampleMany(amen.Name, 0, amen.PeriodPips);

Offset3 has read-only double-precision X, Y, Z fields. Negative pips wrap normally. SampleInto fills an existing span; batches handle int32-boundary crossings safely. PatternLibrary.Load / FromJson read the same compiled JSON packs as the other bindings; a custom bank owns copies of its tables and should be enclosed in using. SafeHandle protects native lifetime, including concurrent sampling. No movements are reimplemented in C# and there is no Python dependency at playback time.

For a checkout build, enable -DDANCERUDIMENTS_BUILD_C_ABI=ON in CMake, then build bindings/csharp/DanceRudiments/DanceRudiments.csproj with dotnet. The runnable C# example and complete C# guide cover native loading, custom packs, supported platform baselines, source development and one-time registry configuration. The NuGet README is included in the output package along with your usage guide, screenshots and Kieran Simkin — My Songs.

.github/workflows/csharp.yml builds/tests all six native targets, creates one complete nupkg, and tests installing it on each target. A published GitHub release then uploads to NuGet.org and GitHub Packages, plus release attachments. Ordinary pushes and pull requests only build/test and retain CI artifacts. NuGet.org requires a configured trusted publisher and NUGET_USER; see release setup. Until publication, use the checkout or the csharp-nuget CI artifact; merging this code alone does not publish.

TypeScript/WASM binding

The npm package contains a typed ES-module wrapper and a separate Emscripten factory and .wasm binary. JavaScript uses the same API; omit the TypeScript import type statements. Initialise the factory once before sampling:

npm install @kieransimkin/dance-rudiments

import { bindNative, catalogue, sample } from "@kieransimkin/dance-rudiments";
import type { RudimentName } from "@kieransimkin/dance-rudiments";
import createDanceRudiments from "@kieransimkin/dance-rudiments/wasm";

const native = await createDanceRudiments();
bindNative(native);

const name: RudimentName = "beat_amen_four_bar_bounce";
const info = catalogue.find(item => item.name === name);
if (!info) throw new Error("This release does not contain the requested movement");
console.log(catalogue.length, info.periodPips / 64);
console.log(sample(name, 48));  // { x, y, z }, sampled in C++/WASM

The currently untyped factory subpath has a small local declaration in examples/bindings/typescript/wasm.d.ts. Include that file in a strict TypeScript project; it derives the module type from bindNative rather than inventing another motion API. The wrapper itself already ships its declarations.

Node.js: use ESM (.mjs or "type": "module") and a current supported Node version. With the package installed, run node examples/bindings/javascript/node.mjs. The factory loads its adjacent WASM binary. The source-build route below avoids relying on an older published package.

Browser: serve the loader and its binary over HTTP(S). Keep them adjacent or supply locateFile: (file, prefix) => ... to the factory when your bundler moves assets. The server must return the WASM bytes rather than an HTML fallback page; application/wasm is the appropriate MIME type. Bare package imports require a bundler or import map. The checked-in browser example supplies a map for a local source build, starts paused, and preserves beat position when BPM changes.

To build from source, activate an Emscripten SDK environment first, then run from the repository root:

npm ci
npm run build:harness
emcmake cmake -S . -B build-wasm -DCMAKE_BUILD_TYPE=Release -DDANCERUDIMENTS_BUILD_WASM=ON -DDANCERUDIMENTS_BUILD_TESTS=OFF
cmake --build build-wasm --config Release
node examples/bindings/javascript/node.mjs ./dist/typescript/index.js ./build-wasm/dancerudiments.js
npx tsc -p examples/bindings/typescript/tsconfig.json
python -m http.server 4173

Open http://localhost:4173/examples/bindings/typescript/. This small consumer is not self-contained; the main visualizer linked at the top of the README is. The visualizer's compact snapshot-WASM interface is not an Embind module and must not be passed to bindNative(). No JavaScript fallback computes movement here.

Custom compiled packs and cleanup: fetch or import the JSON, then explicitly release its native memory when finished:

import { createPatternLibrary } from "@kieransimkin/dance-rudiments";

const response = await fetch("/generated/tutorial.json");
if (!response.ok) throw new Error(`Pack request failed: ${response.status}`);
const bank = createPatternLibrary(await response.json());
try {
    console.log(bank.sample("tutorial_pulse", -1));
    console.log(bank.catalogue());
} finally {
    bank.dispose();
}

Create the bank after binding the native module. A disposed bank must not be sampled. dispose() is idempotent; the wrapper releases its temporary native objects automatically. The global default catalogue does not need disposal. Keep a library alive for the lifetime of an animation rather than constructing it inside requestAnimationFrame.

Limits, attribution and troubleshooting

A private pack accepts at most 1,024 patterns and 1,048,576 total samples; a single pattern is 1–65,535 pips long. These custom-pack validation limits are separate from the larger built-in catalogue. Imported content retains its original licences and notices: copying generated tables does not remove attribution.

The author/project link is Kieran Simkin — My Songs. It is retained in package metadata, the packaged READMEs, AUTHORS.md, CMake and Conan descriptions, and the C++ release's PACKAGE-INFO.txt. Original upstream licences are not modified to insert project branding.

Binding examples and screenshot notes · Complete authoring format · Historical build troubleshooting.

Harness

Use the self-contained HTML link at the top. No server is required for that page. Its 305 beat/event scores include the complete four-bar Amen study; all 1,731 movements remain available. Choose a score, press Play, set BPM, and optionally restrict the catalogue to matching movements. The transport coordinates synthesized MIDI notes and integer-pip C++/WASM motion.

The score view includes ghost-note strengths, fractional event positions, lane mutes, MIDI import and export. It is a timing aid, not a General MIDI soundfont emulator or the original Amen recording. View the mobile screenshot or the full harness guide.

For rebuilding rather than simply viewing, use python tools/build_harness.py after installing this checkout's native extension. The builder also needs clang++ and wasm-ld; screenshot automation needs Playwright/Chromium. Those are build-time requirements, not dependencies of the downloaded HTML.

DanceFlow ecosystem

DanceRudiments is the reusable motion-vocabulary layer in Kieran Simkin's DanceFlow BPM and motion-response ecosystem:

• StemLab is the music-understanding layer. It analyses audio to produce BPM, beat, structure, stem, lyric, harmony, and related timing evidence.
• DanceRudiments turns integer musical positions at 64 pips per beat into deterministic 1D, 2D, or 3D position offsets.
• DanceMoves is the WordPress EPK motion runtime. It schedules and applies BPM-, cue-, and lyric-timed effects on public pages.

The three components can also be used independently. They exchange explicit timing and analysis data rather than depending directly on one another. DanceRudiments' 64-pips-per-beat sampling convention is intentionally higher resolution and is not the same unit as DanceMoves' 16-ticks-per-beat runtime clock.

Motion safety

Apply an output to a bounded subject. DanceRudiments changes position only; it must not be used to add a repetitive full-frame tint, brightness, flash, or colour-grade effect. Under prefers-reduced-motion, do not autoplay the harness or production motion.

Troubleshooting history

See build and release troubleshooting for prior failures, causes and fixes.

Switch the shared visualizer between trajectories and bars

Use Movement view → X–Y plot / Bar graph. Bar view uses the same native C++/WebAssembly sample and the existing clock, shows every dimension as a separate signed bar, and keeps comparison cards on one common scale. Switching views does not interrupt playback or change the underlying recording. The selection is local to your browser; long dimension sets scroll horizontally. This applies to catalogue/release pages and recording harnesses exported by StemLab. Rebuild old generated pages to include the control. See recording visualizer controls.

Continuous integration and releases

Run python tools/check_package_docs.py before packaging. Its source preflight checks nested first-party README.md files in the package input trees, including rig-motion examples and optional source-plan documentation, for the canonical author music link. Third-party sources/ documentation retains its original attribution. Release jobs also check the actual archives; passing source checks does not establish package publication or registry availability.

The Initial collection and Curve/event jobs verify the installed base wheel before adding the optional motion-import extra, pytest and Pillow, then run tests/python through pytest so function-style tests are discovered. Source integration pins must identify published compatible commits. The PyPI publisher checks out its exact release source before applying distribution-size budgets. Archive checks also reject duplicate C++ development files in Python wheels and generated browser/character payloads in Python source archives. This packaging boundary is checked independently from the measured size policy and registry quota. The complete 1,731-movement package uses reviewed 32 MiB wheel, 40 MiB source and 620 MiB release ceilings. Publication additionally verifies the complete eighteen-wheel matrix and its exact release version; PyPI storage must be available independently. See distribution policy. An official GitHub/PyPI mismatch remains a publication failure; preserve existing releases and follow the partial-publication recovery guide.

Tagged Pages deployments require an allowed Tag rule in the existing github-pages environment, in addition to any allowed main branch rule. A maintainer must explicitly authorize the intended release tags; the workflow does not change environment protections. See release demo recovery for retrying an already built site without rebuilding its immutable release assets.

The additional .github/workflows/csharp.yml builds the shared C ABI and C# wrapper, tests actual installed NuGet packages, and publishes .NET packages only for release tags. Its NuGet.org setup is documented in the C# guide.

.github/workflows/ci.yml builds and tests the C++ core on Linux, Windows, and macOS; builds and imports the Python package; compiles/tests the TypeScript and browser runtime; and creates a Conan package on every push to main and every pull request.

A pushed vMAJOR.MINOR.PATCH tag is the CD trigger. Use the release helper so all five version locations stay in sync:

python tools/release_version.py prepare 0.2.0
git diff
git add CMakeLists.txt pyproject.toml package.json package-lock.json bindings/csharp/DanceRudiments/DanceRudiments.csproj
git commit -m "Prepare v0.2.0 release"
git push
# Wait for the normal main-branch CI to pass, then:
python tools/release_version.py tag 0.2.0 --push

prepare updates CMake, Python/PyPI, npm + its lockfile and C#/NuGet together. tag refuses a dirty worktree, a non-main branch, mismatched manifests or an existing tag. The pushed tag runs .github/workflows/release-tag.yml, which revalidates the exact tagged checkout and creates the GitHub Release. The package and demo workflows are also tag-triggered; they do not depend on a GitHub Release event generated by another workflow.

.github/workflows/release.yml rejects mismatches before it publishes anything, then:

• builds platform Python wheels and a source distribution and publishes them to PyPI using OIDC Trusted Publishing;
• builds the TypeScript wrapper and C++ WebAssembly module and publishes @kieransimkin/dance-rudiments to npm using the configured publishing credentials and provenance;
• builds installable C++ archives for Linux, Windows, and macOS and attaches them to the GitHub release; and
• builds a Conan package and uploads it only when a separate Conan remote has been configured. ConanCenter packages are submitted through conan-center-index; they are not directly uploaded by this repository.

One-time registry configuration is required before the first release:

1. On PyPI, create a pending Trusted Publisher for owner kieransimkin, repository DanceRudiments, workflow release.yml, environment pypi, and project name dancerudiments.
2. On npm, configure the package's GitHub Actions Trusted Publisher for kieransimkin/DanceRudiments, workflow release.yml, environment npm, with direct publishing allowed. Because npm Trusted Publishers are configured from an existing package's settings, the first scoped-package registration may require a granular NPM_TOKEN secret in the npm environment. Remove that bootstrap token after Trusted Publishing is configured.
3. Create GitHub environments named pypi, npm, and conan; add required reviewers if desired.
4. For an optional private or organisational Conan repository, set environment variable CONAN_REMOTE_URL and secrets CONAN_LOGIN_USERNAME and CONAN_PASSWORD in the conan environment. If they are absent, the recipe is built and verified but not uploaded.
5. For C#, configure the nuget environment and NuGet.org Trusted Publisher as described in the C# guide.

The PyPI publishing job requests short-lived OIDC identity. npm can use Trusted Publishing when configured; the workflow retains NPM_TOKEN as a bootstrap/fallback credential. See the official PyPI Trusted Publisher and npm Trusted Publishing documentation.

Coordinated releases and latest-release dependencies

Keep DanceRudiments, react-timeline-sequence, and stemlab as sibling checkouts. The release order is React Timeline Sequence → DanceRudiments → StemLab: publish and verify the companions before updating the consuming app. StemLab must use their latest stable official releases, not an unreleased commit, sibling source package or vendored npm archive. Each build records exact released versions/commits/integrity after latest-release resolution so all its jobs use the same bytes; CI fails clearly when that snapshot is stale.

From the parent folder, preview and then explicitly execute:

.\stemlab\scripts\Release-DanceFlow.ps1 -ProjectsRoot .
.\stemlab\scripts\Release-DanceFlow.ps1 -ProjectsRoot . -Execute

For the first migration use the same commands with .\release-kit\Release-DanceFlow.ps1 after extracting the complete kit. It applies the three repository patches, updates versions, performs scoped commits and normal pushes, waits for exact-commit CI, then creates new release tags. The existing GitHub Actions workflows publish packages; no protection rules, credentials, old tags or assets are changed. The initial expected next-minor versions are only a preview proposal, never a claim that they are published.

Use Node.js 24, Python 3.11+, PowerShell 5.1+, Git and authenticated GitHub CLI. Required main/registry/Pages jobs must pass; missing approvals and partial publication are blockers, not success. The default script releases all three projects, is opt-in via -Execute, and supports a saved exact-source -Resume session. A changed checkout or pre-commit build failure requires review rather than an automatic stash/reset or guessed recovery.

StemLab's helper updates npm manifests/lock, optional native dependency minimum, release/build metadata and the genuinely rebuilt JS/CSS together. Its daily release-update workflow proposes subsequent updates by PR without auto-merging. A tagged StemLab release generates all four Arcadians pages and validates actual Pages deployment, live HTML and seekable audio. Do not start a duplicate manual Arcadians analysis during that tagged build. Read the full release, resume, verification and troubleshooting guide before running it. Keep this README and AGENTS.md synchronized with future changes.

Agents and contributors

Agents using DanceRudiments should inspect its available capabilities, use them fully where they fit the task, and look for useful improvements wherever possible. When a required feature is missing, or a reusable reliability, usability, performance or documentation gap is found, improve this tool rather than maintaining a separate workaround. If no worthwhile change is justified, explain why.

Submit every improvement as a pull request to DanceRudiments, with relevant tests, documentation and a reproducible example. Follow CONTRIBUTING.md and AGENTS.md for the contribution and validation workflow.

Project website: https://kieransimkin.co.uk/danceflow/.

Source-first whole-motion BPM confidence and loop catalogues

The source-first analysis job scores each provenance-tracked original BVH or Blender-extracted rig Action for an estimated musical beat grid. It records the leading source BPM, alternative half/double-time hypotheses and a 0–100 kinematic fit confidence, where the default 70% threshold controls automatic full-length conversion to a standard native 64-pip RigMotion. Confidence is a transparent engineering score, not a calibrated probability or proof of musical phase, downbeat or exact audio synchronisation.

Whole Actions remain non-looping and retain original file SHA-256, archive and licence provenance, numbered rig identity and estimated timing metadata. Independently discovered repeating excerpts are compiled separately and listed in loopable-clips.json; they are suggestions until full seam/contact checks. The batch CLI outputs independent full_motion_shortlist and loopable_shortlist indices. Run python tools/analyze_motion_collection.py --sources cmu_dance for reviewed cached BVH archives, or python tools/analyze_blender_capture_batch.py --index extracted/index.json --output-root results for provenance-tracked FBX/GLB/BLEND Action captures. The audio-free StemLab preview distinguishes full Actions from loopable clips and does not embed or redistribute masters.

Reliable Python tests on Windows

Run focused tests through pwsh ./scripts/Invoke-Tests.ps1 <test-path> [...]. The wrapper probes create/read/atomic-rename/delete access before collection, uses a unique per-run root beneath %LOCALAPPDATA%/<project>/pytest-tmp on Windows, redirects Python process temporary files, pytest's --basetemp and cache, then removes only that run's directory and restores the caller's environment. This avoids stale or sandbox-altered %TEMP%\pytest-of-* ACLs and prevents parallel checkouts from sharing a pytest root.

Repository text is stored and checked out as LF through .gitattributes; declared binary media, archives, fonts and model formats remain byte-exact.

An omitted test selection fails closed. Use -FullSuite only for an explicit release or whole-environment diagnostic; routine local work should select the affected tests, while GitHub Actions owns the complete platform matrix.

This is a command-line environment-isolation contract, so screenshots or video would not add reliable evidence. The probe messages, selected test output and process exit code are the verification evidence.