bass-foundry/docs/ideas.md

Plugin ideas

A running list of plugin ideas for riddim, dubstep and bass-music sound design. Each entry names the problem first. Entries are grouped as released or idea. Names follow the metalwork theme of the Bass Foundry suite.

Problems seen in bass-music production

1. Every bass patch brings its own sub level and phase, so the weight of a drop jumps between switch-ups, and layered patches cancel in the sub.
2. Patches at the same peak level still differ in perceived loudness and brightness. The usual fix is hand-set gain or clipping per patch.
3. Resampling is repetitive: bounce, chop, trim, declick, normalise, find the root key, name, save, load into a sampler.
4. Writing wobbles: LFO shapes are drawn, synced, copied and edited for every variation. Triplet and dotted patterns plus per-note changes take long.
5. Unison, OTT and stereo distortion widen the mids. Parts disappear in mono or on club systems.
6. Kick and bass clash. Sidechain helps, but the sub phase at each note start changes how the kick hits.
7. Distortion chains build up harsh resonances. (Patina covers this.)
8. Comparing a drop with a reference: the difference is usually sub/mid balance over time, which generic meters do not show clearly.
9. Fixed-frequency tools act differently on every note. An EQ boost at 300 Hz lifts the 4th harmonic of one note and the 7th of another; a distortion with a fixed tone filter sounds brighter on low notes than on high notes.
10. Layered bass patches partly cancel in the mids, not only in the sub.
11. Texture makes a riddim patch: grit, rattle, metal, vowel, noise. It takes long chains (distortion, comb, formant, OTT, resample, repeat), and the result changes from note to note.
12. Grit and tone cannot be shaped separately. More distortion for more grit also changes the body, the level and the low end.
13. Resampled textures sound printed: a loop of one moment, with no movement inside the note.
14. Texture changes inside a note (the switch-up within one held note) need automation lanes or chopping.
15. There is no meter for texture. Two switch-ups can sound "the same" without a clear reason, and generic analysers show only level and spectrum.
16. OTT and multiband compression use fixed band edges, so the same setting squashes different harmonics on every note.
17. Riddim masters rely on clipping. A clipper that also clips the sub turns the low end into distortion.
18. Resampled basslines have no MIDI, so tools that need the notes have nothing to follow.
19. A clean sub carries the note on a club system but vanishes on phones and laptops. Distortion makes the note readable there, but also changes the patch.
20. Detuned layers (reese) beat at a rate that changes with every note and every harmonic, so the movement is never in time with the track.
21. Kick and sub on the same MIDI tick are not heard at the same moment: the heard hit depends on the attack shapes and on the offset between them.

Signal flow and suite audit (2026-11)

How the released plugins fit into one production:

WRITE      Rivet (motif → storyboard) ──MIDI──► synth
DESIGN     synth → Lathe → Temper → Knurl → Patina
RESAMPLE   Crucible (capture → slice → export) ──► sampler / wavetable synth ──► DESIGN
LAYER      layer A + layer B → Weld (mids) → bass bus → Anvil (sub)
MIX        gap: loudness, clipping, OTT, mono check, kick against sub, reference

Writing and sound design are covered. Mixing and mastering have only Anvil and Patina. Findings from the audit:

• Latency adds up. A full chain (Knurl 71 + Temper 67 + Lathe 67 + Patina about 45 + Anvil 80 ms) is about 330 ms. FL Studio compensates it for playback, but a chain this long cannot be played live.
• MIDI mode keeps the lookahead. Tracker::latency() depends on the pitch range only. With MIDI as the source, the note is known at its attack, so the lookahead could be close to zero. This needs a latency change at runtime, which must be checked in FL Studio first.
• Knurl confuses every tracker after it. Measured: Knurl output fed into TrackerConfig::BASS. From C2 (65 Hz) upwards, Double, Fry (the default), Gallop and Triple make the next tracker read the note one octave (Triple: an octave and a fifth) too low for about 85 % of the note. Below C2 the false sub-octave is under 30 Hz, out of range, so low notes are safe. Level and time jitter also cause 5–25 false note starts per second. Anvil on the bus after Knurl then plays the sub an octave low and restarts its phase. Until there is a shared tracker: put Knurl after the other tracked plugins, and use MIDI as the source for any tracked plugin after Knurl. See Signal chain.
• Each tracked plugin needs its own MIDI route. Four plugins in MIDI mode are four FL routing setups.
• Weld and delay compensation are untested together. If the main layer's chain has more latency than the sidechain layer and FL does not compensate the sidechain route, the main layer arrives late, which an allpass cannot fix. Put Weld after every plugin with latency on that track.
• Crucible lost takes on reload. Takes are not saved with the project. Fixed: Keep takes writes every capture's loop to disk at once.
• Settings handling differs. Patina has presets and A/B; the others do not. Lathe's drawn gains and Knurl's drawn pattern cannot be automated.

Producer feedback (2026-11-06)

Feedback from producers who tried the suite: most plugins are confusing to use. One example: "why does it try to detect the key? It switches around fast, because sounds usually aren't just one kind of note." Anvil "was kind of cool, but didn't do what we expected." Patina's feedback (threshold, ratio) is handled in Patina 0.5.0.

What goes wrong

1. Nine of thirteen plugins depend on one hidden guess: the note. Anvil, Lathe, Temper, Knurl, Planish, Braze, Forge, Ballast (Notes mode) and Crucible (file names) all read the pitch from the audio. Patina's Tone guard also does, but it is off by default. Our tests use saws and sines, where tracking is perfect. Real riddim basses are FM growls, formant sweeps, noise, pitch drops, stacked octaves and layers. On those, "the note" is often not one note.
2. The tracker is built to react fast, not to stay put. In common-dsp/src/tracker.rs, a move of 0.7 semitone that holds for two 5 ms hops starts a new note. Smaller moves retune every 5 ms. Frames below clarity 0.5 are skipped, but the last guess stays in use. On a growl, the formant and FM movement reads as a pitch moving every few milliseconds, and octave errors count as new notes. The NOTE readout flickers, and everything tuned to the note moves with it. That is the "switches around fast".
3. A wrong guess is worse than no guess. When tracking is wrong, a tracked plugin does something no normal plugin would do:
   • Anvil plays a sub at the wrong pitch.
   • Lathe moves its bands onto the wrong frequencies.
   • Temper's Floor, Push and Bright filters jump.
   • Planish's band edges jump, so the bands pump.
   • Braze's beating leaves the grid.
   • Knurl's cycle clock runs at the wrong rate.
   • Crucible names a slice after a note it is not. A producer cannot tell this from "the plugin is broken".
4. The plugins explain themselves in DSP terms. Harmonic numbers, cycles, periods, splits, phase and allpass sections. A producer reads the category ("EQ", "distortion", "OTT", "sub") and expects that behaviour. When the tracked behaviour differs, it feels random.
5. The escape route is hidden. MIDI as the pitch source fixes the guess, but in FL Studio it needs the channel's MIDI output port and the effect's input port set to the same number. Few producers know how to do that.
6. Panels show everything at once. Most plugins put 10–20 controls on the front panel, with equal weight. The one or two that matter for the first result do not stand out.

Principles for the next round

• Never act on a guess you are unsure of. Every tracked plugin falls back to something neutral when tracking is unsure: pass the audio through, keep the original sub, or use fixed-Hz settings. It does not fall back to the last guess. Show this state: the NOTE readout dims or shows "—".
• Stable by default. A new note needs a confident pitch held for 40–60 ms. An octave jump needs longer still. Inside a note, the tuning holds and does not retune every hop, unless the user turns on "Follow bends".
• Let the producer say the note. Add a third pitch source next to AUDIO and MIDI: NOTE or KEY. The producer picks the root (for example F), and the plugin uses it with no detection at all. Most riddim drops sit on one or two roots. A scale lock goes in between: detection may only pick notes in the chosen key. This removes the micro-jumps and most wrong notes.
• Name controls by outcome, not by mechanism. Use names like "Sub level" and "Tighten". Show one big main control per plugin, and move the rest under an "Advanced" fold.
• Test on real material. Collect 5–10 short stems: a growl, an FM bass, a reese, an 808 with a slide, a formant bass, a layered drop, a noise riser. Add a tracker metric for note changes per second and the share of time the tracker is wrong or unsure. Every tracker change must lower it on the stems, not only pass the saw tests.

Per plugin

• Anvil. It does not do what producers expected. The candidates are below; ask the tester which one they meant.
  • (a) They expected a sub enhancer: add weight under the existing sound, not replace it.
  • (b) They expected it to follow their growl, and it played wrong sub notes. The wrong sub notes are the worst failure in the suite.
  • (c) They expected a simple "fix my low end" (mono, tight, level), and the front panel offers Phase, Retrigger, Kick lock, Ghost and Audition. Fixes:
  • Keep the original sub (Keep 100 %) whenever tracking is unsure, and switch to the clean sine only on a confident note.
  • Add the NOTE source with a root picker.
  • Add an Enhance mode that adds the sine under the original instead of replacing it.
  • Keep three main controls: Split, Sub level, Mode. Fold the rest.
• Lathe. A harmonic EQ is a new idea for most producers. Show the band centres in Hz on the plot, next to the harmonic numbers. When tracking is unsure, pass the audio through.
• Temper. Producers judge it as distortion first. Floor, Push and Bright should keep their last confident tuning, or fall back to fixed Hz, instead of jumping. Consider Floor/Bright with fixed-Hz values as the default, and tracked values as an option.
• Knurl. It needs the cycle clock, so it needs a pitch. NOTE mode and fade-out on unsure tracking matter most here.
• Planish. When tracking is unsure, fall back to fixed band edges (like OTT) instead of jumping. Crossfade the edges when the note changes.
• Braze. It needs a pitch. Use NOTE mode, hold the detune through unsure stretches, and do not jump.
• Ballast. Make Grid the default for Parts from. A switch-up lands on the grid, and Grid needs no tracking.
• Crucible. Name a slice by its note only when one note is confident over most of the slice. Otherwise leave the note out (or write "?"). An optional song key field avoids enharmonic names that do not fit the key.
• Forge. It inherits everything above from the shared tracker.
• Weld, Quench, Patina. These use no tracking by default. Weld's display is dense: lead with Recovered dB and one ALIGN button, and fold the plots.

Order

1. Shared tracker: confidence, stability and hold, plus the stem corpus and metric. One change in common-dsp fixes the flicker in nine plugins.
2. NOTE/KEY pitch source in the shared tracker settings and in each tracked plugin's header.
3. "Unsure means neutral" fallback in each tracked plugin, Anvil first.
4. Anvil Enhance mode and a reduced front panel. Ask the tester what they expected first.
5. Front-panel pass on every plugin: a main control, outcome names, and an Advanced fold.

Status (2026-11-06)

Done in Anvil 0.4.0, Lathe 0.3.0, Temper 0.3.0, Knurl 0.3.0, Planish 0.2.0, Braze 0.2.0, Ballast 0.2.0, Forge 0.2.0, Crucible 0.4.0 and Weld 0.4.0. Not tested in FL Studio yet, and the tester has not been asked which Anvil they expected.

1. Done. The tracker needs a steady pitch for 40 ms to start a note (80 ms an octave down, 200 ms up), holds within a quarter semitone, ends a note after 20 ms of silence, and reports unsure after about 40 ms of disagreeing reads. Follow bends keeps the old fast behaviour. The corpus is synthetic (common-dsp/src/stems.rs, examples/tracker_report.rs): extra note starts per second on formant, wobble and reese fell from 28.5, 11.8 and 1.3 to 0. Wobble is still unsure 37 % of the time and the hard growl 4.5 %.
2. Done: NOTE source with a note picker, Key lock (minor, major, root only) and Follow bends, shared through common-params. Ballast has no NOTE: its parts need note starts from the audio or MIDI.
3. Done. Anvil plays the original sub, Lathe passes the audio, Knurl fades out, Temper, Planish and Braze hold the last sure note. Planish slides its edges over 6 ms on a new note. Crucible names slices only on a sure note and spells them for a song key.
4. Done: Replace / Enhance mode, and a MAIN page with Mode, Split and Sub level.
5. Done for Anvil, Lathe, Temper, Knurl, Planish, Braze, Forge and Weld: a MAIN page and an ADVANCED switch. Lathe labels harmonics in Hz. Ballast defaults to Grid.

Still open:

• Real stems. Score the tracker on 5–10 real drops with tracker_report and tune on those, not only on the synthetic corpus.
• Outcome names. Controls keep their old names; only the pages changed.
• Temper with fixed-Hz Floor and Bright as the default.
• Weld's single ALIGN button.
• Panel pass for Quench, Patina, Crucible, Rivet and Ballast. Their panels were already short, so they were left as they are.
• tools/editor-check has a clippy chunks_exact lint outside the plugin checks.

Research findings (2026-11)

From the papers in research/ (local, not committed). Each finding names where it goes in the suite. They are starting points to test by ear, not rules.

• A published bass fatness model. Schmedeke's dissertation (Hamburg 2021, "BASS DROP!", section 5.4.3) models momentary fatness of dance-music excerpts over 50 ms windows from two inputs only, with adjusted R² 0.773: l = 10 / (1 + 4.5487·exp(0.13·(62.9366 − SPL<200))), h = 10 / (1 + 1.4317·exp(2.5575·(0.3299 − HR<200))), Fat = 0.14689·l·h (0 to 14.7). SPL<200 is the C-weighted level below 200 Hz in dB(C) with the material normalised to −23 LUFS; HR<200 is the MPEG-7 harmonic ratio (how periodic) of the same band. Chapter 4's earlier linear model on synthetic tones (R² 0.597) points the same way: a tonal low band is fatter, a noisy one is not. Anvil's clean sine maximises HR; Quench keeps it when the top is clipped. The thesis does not state the dB(C) of its reference, so the dBFS-to-dB(C) offset is an assumption. Measured with scripts/fatness.py on the seven Devath references (offset 100 dB): drops sit at 4–6, breakdowns near 0, kick gaps show as 50 ms dips; SPL<200 medians are 67–69 dB(C), a few dB under the thesis's "mostly 70–80", so the offset may be 2–5 dB low. Goes into Assay as one lane, never as an overall score.
• Virtual bass from known harmonics. Moliner, Rämö and Välimäki (DAFx 2020) boost the harmonics that exist with weights that follow the sound's spectral envelope (limits 10 dB/oct below, 3 dB/oct above) and generate the missing ones, so a small speaker suggests the fundamental. Their nonlinear stage fails intermodulation tests, so they use it on transients only. The suite knows f0, so it can add harmonics at exactly k·f0 with no intermodulation (problem 19). Anvil's sub is a pure sine, so Lathe has no harmonics there to boost.
• Group-delay audibility. Liski, Mäkivirta and Välimäki (2021): between 500 Hz and 4 kHz, listeners hear a group-delay peak from about 0.64 ms (positive) or 0.56 ms (negative) on clicks over headphones; a fast-decaying hi-hat hides it. Weld's allpass sections add positive group delay there. Done: Weld shows the delay spread (released in 0.3.0). Measured: common fits from Weld's own tests spread 0.82–1.03 ms, above the click threshold, so the display gives a note, not a warning.
• Timbral tension. Farbood and Price (2017): more roughness, inharmonicity and spectral flatness raise judged tension; brightness (centroid) gives no clear result. These are the measures planned for Loupe, and the controls a tension macro would move.
• Coupled one-knob controls. A 2026 distance renderer derives five distance cues from one control, so they always agree; in its listening test it roughly halved placement error and every participant preferred it. A pattern for macros in Forge.
• Shared motion binds layers. Temporal coherence (Elhilali et al. 2009) is a grouping cue: sub, body and top sound like one sound when their main movements line up. One modulation source inside Forge gives that by construction; across separate .clap files it needs the shared memory rejected below.
• Heard timing of layered hits. Danielsen et al. (2026): attack shape and the offset between two layers together set when the compound hit is heard and whether it fuses. Hove et al. (2014): timing deviations are heard more clearly in lower voices, so the bass sets the grid (problem 21).
• Loopback FM. Smyth (DAFx 2026): an oscillator that frequency-modulates itself shifts its own sounding pitch as feedback rises. Her closed-form solution gives the sounding frequency, so a growl can stay on the tracked note at any feedback.
• Beating depends on note and harmonic. For two layers detuned by ratio r, harmonic k beats at k·f0·(r − 1) Hz (Plomp and Levelt 1965 for the roughness context). Fixed cents beat faster up the keyboard; fixed Hz beats the same on every harmonic (problem 20).
• Do not encode as rules: a best sub frequency, crossover or kick/bass offset; one number for loudness, roughness, groove and preference; random jitter sold as humanisation; more harmonics sold as more physical bass; largest correlation sold as the best alignment.

Released

Anvil — sub anchor (problems 1, 6)

Replaces the band below a split frequency with a clean mono sine that follows the bass pitch, at one level and one start phase for every note. Pitch comes from detection with 80 ms lookahead, from MIDI, or from a set note. Enhance mode adds the sine under the original sub instead. See Anvil.

Crucible — resampler (problem 3)

Records the input continuously. One click captures the last bars on the host bar grid, slices on a grid or on hits, trims, declicks, normalises, detects the root key and exports labelled WAVs, a full loop and an optional Serum/Vital wavetable. See Crucible.

Lathe — harmonic EQ (problem 9)

EQ bands placed on harmonics 1–32 of the detected note instead of fixed frequencies. "+4 dB on harmonic 3, −6 dB on harmonic 7" sounds the same on F1 and on C2. Odd/even balance, tilt over harmonic number, and a Focus control that lifts or cuts everything between the harmonics. Uses Anvil's pitch tracker, or MIDI. See Lathe.

Temper — note-tracked distortion (problem 9)

A waveshaper whose tone filters before and after the shaper follow the note, so every note distorts with the same character and the same harmonic content. Input level is normalised before the shaper, so drive does not depend on how loud the patch is, and the output is gain-matched. Oversampled to keep aliasing (which is never harmonic) out. See Temper.

Weld — layer phase alignment (problem 10)

Takes a second bass layer through the sidechain. It fits polarity and a chain of up to four allpass sections to the measured cross-spectrum (no added delay), so the layers sum instead of cancel. A display shows how much level the cancellation cost and how much Weld recovered. Anvil solves the sub band; Weld solves the mids. See Weld.

Knurl — cycle texture (problems 11, 13)

Follows the played note and treats every cycle on its own: a repeating level pattern (in-tune subharmonics), random level and timing per cycle, and a noise burst per cycle tuned to the note. Amounts are counted in cycles, so the texture is the same on every note. Slow per-harmonic drift is postponed. See Knurl.

Planish — harmonic OTT (problem 16)

Upward and downward compression in four bands whose edges are harmonic numbers of the tracked note (default 1, 2–4, 5–16, 17 and up), so the same setting squashes the same harmonics on every note. Each band's level is measured over one period of the note, so heavy settings do not ripple at f0. See Planish.

Next for released plugins

• Anvil: Kick lock and the offset audition are done (Anvil 0.3.0, see Anvil): the kick through the sidechain ducks the sub and restarts its phase at an offset (problem 6); the audition steps −30 to +30 ms, 4 kicks each, Keep takes the offset heard, and the low-band sum per offset is shown apart from that choice (problem 21). Measured on a 55 Hz kick burst against a 55 Hz sub: +2.4 dB at 0 ms, −4.6 dB at −10 ms, −5.9 dB at +10 ms, and +1.8 to +2.2 dB again one period away (±20 ms), so the sum alone cannot pick the hit. Open: a finer second sweep around the kept offset; a gate or sensitivity for kicks with long tails or a kick bus with other drums. A sub shape from sine to rounded triangle. A note attack counted in cycles. Ghost harmonics and the Phone listen are done (Anvil 0.3.0, see Anvil): harmonics 2–10 above a "from" frequency, filled only where the patch lacks them, with the sub level unchanged (problem 19). Open for Ghost: weights that follow the patch's own harmonic slope instead of a fixed 1/k; measuring a new note's harmonics from the lookahead, so its first 60 ms do not count them as missing; a listening check on an 808 over a real phone.
• Lathe: A/B gain snapshots with one automatable Morph parameter. A keyswitch note that selects the drawn shape (EQ switch-ups without automation). A dynamic mode per harmonic.
• Temper: a harmonic split: harmonics 1–2 clean, 3 and up driven. More than one Push band. A bias that changes per cycle from Knurl's cycle clock.
• Knurl: velocity or note selects the pattern. Stereo grit above a split only. Drift (planned for 0.3).
• Patina: Tone guard (done, Patina 0.4.0). MIDI as its pitch source. M/S processing. Threshold, Ratio and Soft/Hard knee from tester feedback (done, Patina 0.5.0). Next, from the same feedback round:
  • An absolute level gate (dBFS) so quiet passages are never cut, if the tester meant a loudness threshold rather than a prominence threshold.
  • Narrow rings on dense or noisy material are under-read: a +12 dB ring (Q 25) on noise reads only 3–6 dB above the floor, so it is cut by 1–3 dB. Two causes, measured: the joined-tops envelope lifts the floor to the noise peaks (about 5 dB above the noise mean), and envelope smoothing loses about 3 dB on a narrow ring. A floor that follows Sharpness (1/3–2 octaves) was tried and rejected: it lost a resonant lowpass, absorbed broad bumps and cut plain harmonic tones. Better candidates: join tops only across deep gaps (harmonic spacing), or a mean-based floor for noise-like regions.
  • A finer analysis grid (1024 cells) for narrow rings at high frequencies.
• Weld: delay spread display (done, Weld 0.3.0). A gentle fit option that keeps the delay spread under a limit, for plucky layers. Alignment slots per note or per switch-up. The External sender/receiver fallback if FL's sidechain does not arrive.
• Crucible: a heard-onset marker per slice next to the technical cut, set by ear and written to the WAV (problem 21). A switch-up slicer that cuts where the spectrum changes, for legato basses. Optional retune of slices to one root. Reopening saved takes from the takes folder in the take list.
• Rivet: more than two sound roles, one per switch-up patch. Its MIDI output as the one note source for the tracked plugins.
• Suite: presets and A/B in common-ui for every plugin.

Ideas

Order

1. Small fixes from the audit. Done: chain-order guide (Signal chain), Crucible Keep takes. Open: Weld test with delay compensation (in FL), MIDI mode without lookahead (needs an FL check that a latency change at runtime is accepted).
2. Patina pitch-aware mode. Done as Tone guard (Patina 0.4.0): bumps that follow the note are protected, bumps fixed in Hz are cut. Measured first: Patina already cut fixed formants on every note, but it also cut a boosted harmonic as hard (5.7 dB), at a position that moved with the note. Open: MIDI as the pitch source.
3. Planish (done, released 0.1.0), then Quench and Ballast (both done, released 0.1.0): they fill the mix and master gap.
   • In parallel, from the research findings, in this order: Weld delay spread (done, Weld 0.3.0); the fatness lane, as the first Assay piece (prototype scripts/fatness.py); Anvil Ghost harmonics and Phone listen (done, Anvil 0.3.0); Braze (done, 0.1.0); Anvil Kick lock with offset audition (done, Anvil 0.3.0).
4. Forge (done, released 0.1.0): Lathe, Temper and Knurl behind one tracker.
5. Assay, one plugin for the display-only ideas.
6. Texture plugins (Anneal, Swarf, Etch, Bellows) and the note-keyed sequencer.

Merges: Gauge, Loupe and Drop compare share one analysis core and one display, so they become one plugin, Assay. The wobble sequencer, Damascus and Chase are all "step grid, note selects the pattern", so they become one sequencer.

Forge — one tracked channel strip (problems 9, 11, latency)

One tracker, one latency, and Lathe, Temper, Knurl and later Anneal as modules in an order you choose. Two further reasons from the research findings: one modulation source for all modules moves sub, body and top together, which binds them into one sound; and coupled macros become possible, for example a Tension knob that raises roughness (Knurl), inharmonicity and noise (Anneal) together at matched loudness, measured by the same descriptors as Loupe. It removes the stacked lookahead, the trackers that disagree (Knurl's sub-octave cannot confuse a tracker that reads the input before Knurl) and the repeated MIDI routing. The standalone plugins stay. Sharing one tracker between separate .clap files would need named shared memory across DLLs, which is fragile; one strip is simpler.

Built (released 0.1.0, see Forge). Lathe, Temper and Knurl became modules (a Tracked trait in common-dsp: ahead, event, process); each standalone plugin is now its tracker plus its module, Lathe and Temper bit-exact against the old engines, Knurl one sample later (its interpolation needs one newer sample) and equal within float rounding. In the strip one tracker reads the input; a per-sample record of the tracker's events, note state and input is replayed to each module, delayed by the latency of the modules in front of it. Findings:

• One module alone in Forge matches its standalone plugin sample for sample, in all six orders (tested).
• Knurl's Double-style pattern (0, −12 dB) in front of Lathe: standalone Lathe after Knurl read the note an octave low in 149 of 149 frames on an A2 saw; Lathe inside Forge read A2 in 149 of 149.
• Latency 72 ms against 206 ms for the three in a row (48 kHz).
• Temper's Normalize looks one lookahead ahead; in the strip it reads the strip input, not the modules in front of it. Fine while those change every note's level alike.

Next for Forge: the Tension macro over Knurl (and later Anneal) at matched loudness; Planish and Braze as modules (both are tracked, and Braze after Knurl is the next tracker that hears the sub-octave); a short crossfade when the order changes; one shared modulation source (per-note envelope or LFO in cycles) that every module can read.

Braze — tempo-locked beating (problem 20)

Brazing joins two metals with a filler. Braze adds a detuned copy of the bass whose beating is in time with the track on every note. With the tracked f0, it sets the detune ratio for each note so that a chosen harmonic beats at a tempo-synced rate: r = 1 + B/(k·f0), for example an eighth at 140 BPM is B = 4.67 Hz. A second mode adds a frequency-shifted copy, so every harmonic beats at the same rate. Controls: rate (synced), target harmonic, depth, spread across harmonics, mode. It reuses the tracker, and Weld should not "fix" this beating: put Weld before Braze.

Built (released 0.1.0, see Braze). The detune is a delay line whose delay falls one period per beat cycle and jumps back one period, which is seamless on a periodic note, and the delay follows the host's beat position, so the peaks land on the grid. Harmonic n then beats at n/k · B: every harmonic is on the grid, so the planned spread control was not needed. A Low cut (LR4 on the copy, the bass split by the same crossover) keeps the sub out. Open: the frequency-shift mode. Mixed with the dry bass, a plain single-sideband shift gives every harmonic the same envelope at the same moment, which is tremolo; it needs a dispersive allpass on the copy to stagger the harmonics before it sounds like movement. Also open: a listening check on a real reese, and how Weld reads a layer after Braze.

Arc — self-FM growl (research spike)

An arc welds with a sustained spark. Arc runs loopback FM (Smyth, DAFx 2026) on the tracked note: a time-varying delay line modulated by its own output. The closed-form sounding frequency lets Arc retune the carrier, so the growl stays in key while feedback sweeps from clean to torn. Needs listening and an aliasing check before it becomes a plugin.

Quench — sub-safe clipper (problem 17)

Quenching hardens hot metal in one fast step. Quench clips only above a split frequency and adds the clean sub back with its own headroom, so the master can be clipped hard without distorting the low end. Pairs with Anvil's split.

Built (released 0.1.0, see Quench). Design checks, made in a prototype before the plugin:

• Splitting and giving the top the instantaneous room, clamp(top, −C − sub, C − sub), is exactly a plain clip of the sum: the error is the same signal. The sub-safe part has to come from somewhere else.
• Feeding the clip error back through 1 − lowpass diverges: the complement of an LR4 lowpass peaks at 1.5× near the split.
• What works: clip the top symmetrically at the ceiling minus the sub's held peak (no pumping at the sub's rate, the sum stays under the ceiling), then high-pass the clip error at the split so its products below the split go. Two clip-and-high-pass passes and an oversampled hard clip hold the ceiling. Measured: sub level +0.04 dB against −2.2 dB for a plain clip; low-band products −35 dB against −16 dB. Cost: 0.5–1.2 dB less RMS at the same ceiling.
• Where it does not help: a top whose fundamental lies within an octave of the split is partly in the sub band (saw at 98 Hz over a 49 Hz sub: about the same as a plain clip).
• Next ideas: link the channels' sub peak and offer mono below the split (with Gauge); give the top the room on the sub's far side too (asymmetric) with the error still high-passed, and measure whether that wins back the loudness without pumping; a true-peak ceiling.

Scribe — audio to MIDI (problem 18)

A scribe marks lines on metal. Scribe runs the note tracker on a bass and outputs MIDI notes and a .mid file. A resample from Crucible then gets its notes back, so the other plugins can use MIDI as their source.

Bevel — mono-safe harmonic widener (problem 5)

Pans harmonics of the note alternately (for example odd left, even right) with complementary filters, so the mono sum stays flat. Pairs with Assay.

Peen — note-start transient shaper

Peening hardens a surface with hammer blows. Attack and sustain shaping on every note start, with times counted in cycles, so each switch-up gets the same punch on every note. Could become part of Ballast.

Ballast — switch-up loudness leveller (problem 2)

Measures short-term perceived loudness per note or per switch-up against a target or a reference patch and adjusts gain only, so each switch-up lands at the same loudness without compression changing its character. Earlier working name: Hit Leveler. Pairs with Anvil: Anvil levels the sub, Ballast the whole patch.

Built (released 0.1.0, see Ballast): one gain per part, parts from note starts or the host grid (a patch change on the same note is not a new note to the tracker), K-weighted loudness of the first 100 ms, Learn for the reference. Found while building it:

• A note start from a pitch change comes about 20 ms early, in the gap before the note; a plain mean over the window then reads the part ~0.9 dB quiet. Fixed with 10 ms blocks and a relative gate (−20 dB under the loudest block).
• The K-weighting filters ring the last part into the next one: a near-silent part after a loud one read −18 LUFS instead of under −60. Fixed by weighting every part from rest.
• Next ideas: a brightness match next to loudness (problem 2 names both; with Lathe or Patina doing the EQ), a "follow" option for long notes that swell, and switch-up detection from timbre change instead of the grid.

Bellows — pitch-tracked resonator

A comb or Karplus-style resonator tuned to the detected note. It turns noise, distortion or drums into pitched metallic tone that is in key on every note. Damping, harmonic spread, single or multiple combs.

Gauge — low-end mono check (problem 5)

Shows phase correlation and mono-sum loss per band and per note in the bass range, and marks the notes where widening breaks the mono sum. Optional fixes: mono below a chosen frequency and phase rotation on the mids. Display-first, so it is a small build. Earlier working name: Bass Mono Check.

Wobble sequencer (problem 4)

Step patterns for volume, filter, formant and pitch with triplet and dotted grids built in. Each MIDI note selects a different pattern, so a switch-up is a different note in the piano roll. ShaperBox overlaps, but it has no note-keyed patterns. No name yet.

Fatness lane (problems 2, 8)

Momentary fatness per 50 ms from Schmedeke's model (see Research findings), drawn as one lane next to sub/mid energy, for a patch and a reference. It explains why two drops at the same LUFS feel different in weight: a quieter or noisier low band lowers the value. Prototype: scripts/fatness.py (offline, on rendered files). The first piece of Assay. Open: the dB(C) offset, and how the lane reads on a single bass patch rather than a full mix.

Drop compare (problem 8)

Loudness-matched A/B of a drop against a reference, with sub, low-mid, mid and high energy over time: for example "your sub is 3 dB under the reference in bars 2–4". No name yet.

Texture ideas

These ideas target problems 11–15; Knurl, the first of them, is released above. Most of them reuse the note tracker in common-dsp, so a texture setting means the same thing on every note, the same way Lathe and Temper work.

Anneal — tone and grit split (problem 12)

Annealing changes how a metal behaves without changing its shape. Anneal splits the signal into three parts: the harmonic part (energy on the harmonics of the tracked note), the residual noise and grit between the harmonics, and the transients. Each part has a level fader, a drive and a tilt.

• Turn the grit up 6 dB and keep the body and the sub unchanged.
• Take all the grit out of a layer and leave a clean tone for further design.
• Listen to each part solo, or send the grit to a second output for its own chain.

DSP: an STFT with the note tracker; harmonic masks at k·f0, with a width that follows the pitch; a transient mask from spectral flux. Latency is one FFT frame. The display shows the three parts in different colours on Lathe's harmonic strip.

Swarf — in-key granular shavings (problems 11, 13)

Swarf is the shavings that machining leaves. Swarf records the bass itself and sprays short grains of it back as a layer. Grain lengths are whole pitch periods, and every grain is retuned to the note playing now, so the shredded layer stays in key across a riddim line. Controls are density, spray (position spread back in time), grain size in cycles, reverse probability, and a harmonic-only filter for the layer. It is useful for the "shredded metal" top layer, and works on a resample from Crucible as well.

Etch — texture imprint (problem 11)

Etching transfers a pattern onto metal with acid. Etch takes a texture source through the sidechain or from a loaded sample (a metal scrape, a vocal, a field recording) and imprints its fine structure on the bass. The bass keeps its pitch and its harmonic levels; between and on the harmonics, the source's noise, roughness and movement come through. A "harmonic lock" mode samples the source's spectral envelope only at the bass harmonics, so the imprint is always in key. It differs from a vocoder: a vocoder transfers the envelope, and Etch transfers the surface.

Chase — talking formants that hit harmonics (problems 11, 14)

Chasing shapes metal by hammering it from the front. Chase is a formant filter with a 2D vowel pad (a, e, i, o, u, plus "yoi" and "wub" shapes). You draw a path over the pad and sync it to the tempo, or let each MIDI note select a path. The formants stay in Hz, so a vowel stays a vowel on every note. A snap control moves each resonant peak onto the nearest harmonic of the tracked note, so a narrow formant never falls into the gap between two harmonics and vanishes on some notes. Gain-matched, so a vowel change does not change the level.

Damascus — texture step sequencer (problem 14)

Damascus steel is folded from layers into a visible pattern. Damascus holds up to four texture slots (for example a formant, a comb, a sample-rate reducer and a Knurl-type rattle) and a tempo-synced step grid that chooses a slot, or a mix of two, per step. Steps have their own crossfade time, so changes can be hard cuts or smears. Each slot is gain-matched, so the steps change texture and not level. Each MIDI note can select its own pattern, which is a texture switch-up inside one held note. This overlaps with the wobble sequencer above; both could become one plugin.

Loupe — texture meter (problem 15)

A loupe is the lens used to inspect engraving. Loupe measures texture per note over time:

• roughness and fluctuation strength (psychoacoustic models);
• harmonic-to-noise ratio, which shows how gritty a sound is;
• inharmonicity, which shows how metallic a sound is;
• brightness relative to the note (harmonic centroid, not centroid in Hz).

Roughness, inharmonicity and flatness are also the three timbre measures that raise judged tension (Farbood and Price 2017), so the same lanes show a build-up getting tenser without a note change.

It places every switch-up of a drop as a point on a texture map. Two patches that sit on the same point are why a drop sounds repetitive. A reference track can be loaded for comparison. Display only, so it is a small build, and it fits with Gauge and Drop compare.

Further leads

• Creative extrema stretch for Crucible: Nielsen (DAFx 2026) drives an overlap-add stretch by the spacing of local extrema, a free measure of detail. Cheap and gritty, not transparent; a bass has few extrema per cycle, so the stretch stays close to cycle-synchronous.
• Reference to preset: an offline search over a bounded set of Lathe and Temper parameters that matches a reference switch-up (ST-ITO, TAPE). Large; needs learned audio features.