markov.vst3 is a Markov-chain melody generator (MIDI in and out, transport-synced).
The melody is a random walk over the twelve pitch classes, counted from the Root note. A 12 x 12 transition matrix gives, for the note just played, how likely each note is to follow. The matrix is drawn on screen: brightness is each move's real chance (after the scale, anything learned and Chaos), the digit is the weight you set (0-8), and clicking a cell changes it. Eight styles fill the matrix; the plugin can also learn it from a melodic line on its MIDI input.
Full description, controls and the learning workflow: docs/pages/_products/markov.md
(published as the Markov page). Reference entry: docs/pages/plugins-summary.md.
include/markov_core.hpp,src/markov_core.cpp: the portable core. The model (base matrix plus learned first- and second-order counts), styles, the effective matrix, the walk, playback, MIDI handling (learning and the Conductor CCs) and the state text format. No framework code.src/dpf/MarkovPlugin.cpp: thin DPF wrapper. Settings are parameters; the model is themodelstate, and the editor sends matrix edits through a separateeditstate so it cannot overwrite what is being learned.src/dpf/MarkovUI.cpp: the editor, in the Magneto look, built on the shared kitinclude/downspout/dpf/MagnetoKit.hpp(also used by retune, sprout and damiano).tests/markov_core_tests.cpp: deterministic tests: scales againstdocs/scales.md, the walk and its statistics, order 2, learning, locate and block-size independence, the Conductor CCs and restart, and the state format.
- Determinism. Each phrase restarts the walk from the tonic, seeded by (Seed, a re-roll
counter, phrase number), so note
kof phrasepdepends only on those and the model. The walk costs O(k) per step; phrases are at most 8 bars. - Chaos is the exponent
2^(1 - 2 chaos)on the weights, so 50% is the identity. - Scale mask before sampling. Out-of-scale notes get weight 0; a row left empty falls back to every allowed note.
- Order 2 uses learned second-order counts for the (two back, one back) pair when they hold at least two observations, and the order-1 row otherwise. With nothing learned, Order 2 equals Order 1.
- Clear learned is a counter parameter (any change clears), so the editor can trigger it reliably; the first value seen after load is adopted without clearing.
- Scale tables are Sprout's, with an independent copy of the semitone sets in the tests.
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DDOWNSPOUT_BUILD_MARKOV=ON
cmake --build build --target downspout_markov_core_tests markov-vst3 --parallel "$(nproc)"
ctest --test-dir build -R downspout_markov_core_tests --output-on-failureCore, tests, wrapper and panel are complete. Host validation in REAPER is pending.