Notes on the Synth breathe out now instead of being cut off. Lifting a key used to stop the oscillator dead in the same instant — which meant the Release on a patched Envelope was never audible, because the sound it was fading had already been thrown away, and every note ended on a small click. A released note is now let go of rather than destroyed: it keeps sounding through its fade and is cleaned up afterwards, with any LFO or Mod Env still moving through the tail so the release lives instead of freezing. Every preset in the bank gained the release it was always written to have.
Everything shipped here, newest first. No version numbers — just the work, in the order it happened. RSS feed ⟶
The Oscillator has Release and Pan of its own. Release is how long that voice takes to fade after you let go — a fiftieth of a second is a pluck, two seconds is a pad you can play chords on because each note hangs while you find the next. Pan places each oscillator separately in the stereo field, so two detuned saws at −0.6 and +0.6 open the patch out wide, which is a different thing than panning the finished sound.
Modules now join the chain where they belong, not where you clicked. Auto-routing used to splice everything in immediately before the Output, so a Filter added after a Reverb landed after it — filtering the room instead of the sound. Each kind of module now has a place in the signal path — tone, then shape, then level and position, then space, then the glue on the end — and a new one is inserted at its place regardless of what order you built the rack in. Cables you drew by hand are still never touched, and the ◂ ▸ arrows still override all of it.
The Song Analyzer now says whether it's working before you paste a link. Its YouTube path runs on a home server behind a tunnel, and when that box was down, or YouTube was blocking it, visitors got "Failed to fetch" and no idea why. The backend now records the outcome of every YouTube download and reports it from /health, and the page shows a status line under the input on load: online and YouTube working, online but YouTube is blocking downloads — drop a file in, or server offline. Network failures mid-session say the same thing in plain words.
Under the hood: a bot-checked download is retried once through YouTube's embedded and mobile players before giving up, since the check is applied per client as much as per address; the JavaScript runtime yt-dlp needs for YouTube's player challenge now prefers Deno, the one yt-dlp actually supports, with Node kept as a fallback; and every link shape works — Shorts, /live/, YouTube Music, youtu.be share links, or a bare video ID.
The Synth's recordings are WAV files now, not WebM. Browsers only offer WebM to a web recorder, and WebM has one specific failure that matters here: it plays fine in a browser and most DAWs refuse to import it. A jam you can't open in the thing you make music with isn't much of a recording. The capture is now decoded and rewritten as a 16-bit WAV before it reaches your downloads — no encoder library, no external service, just the samples written out properly — so it opens in Logic, Ableton, Reaper or GarageBand and plays on any device. The cost is size, about 10 MB per stereo minute, which is the honest price of lossless and universal. If a browser ever fails to re-encode, the original is saved rather than the take being lost.
The Synth's drums have faders. They bypass the patch on their way to the output, which is what lets them work before you have built a sound, and it also meant they had one fixed level and no way to sit against the notes. There is now a Mix pair at the right of the drum row: Drums for the kit, Synth for everything you play. The Synth fader is the Output module's Master reached from beside the grid, so the module's own slider moves with it and the two views can never drift apart. Every drum row also gained its own fader under its name, so a hat can be tucked under a kick the way it would be on any drum machine. All of it is remembered between visits.
The Synth now has sixty presets, in six moods. Thirty-four are new, and every one has a name that says what it feels like and a line that says when you'd want it: Rhodes at Midnight, Warm Blanket, Choir of Ghosts, Sunrise (hold a chord and wait three seconds for the filter to open), Underwater Cathedral, Moog Walk, Neon Bass, Screaming Saw, Flute in the Fog, Theremin, Harp in the Rain, Berlin School, Wind Chimes, Bubbles, Distant Thunder, Broken Radio. The toolbar's preset strip, which had already outgrown a laptop screen, is now six mood buttons — Keys, Pads, Bass, Leads, Plucks & Arps, Texture & FX — each opening a panel of cards, plus 🎲 Surprise me for a sound you didn't choose. Every preset is built from the same modules you can add by hand, so load one and read the rack.
The theory guide's keys are easier to read. A lit key now wears its role across its whole face — amber root, purple third, green fifth, yellow seventh, cyan scale tone — with the note name set in the same colour, instead of a nine-pixel dot you had to lean in for. Keys outside the pattern go dark, so the scale is the bright thing on the keyboard.
And the Synth fits any window. Its navigation bar and toolbar are fixed and both wrap as the window narrows, but the page padded itself for one assumed height, so a custom-sized window put the top row of modules under the toolbar and the bottom row under the keys. The page now measures the nav, the toolbar and the dock and pads itself to fit, live, as the window changes; on a short window the keyboard dock scrolls inside itself rather than swallowing the canvas.
The Synth has an arpeggiator. Press ⇅ Arp, hold a chord, and it plays one note at a time, in time — up, down, up and down, random, or in the order you pressed the keys, over one to four octaves, at any note length from quarters to thirty-seconds, with a gate for staccato or legato and a Hold that keeps the chord going hands-free. It runs on the same clock as the sequencer, so a programmed beat and a held chord lock together at one BPM. Keys, pads and MIDI all feed it; the sequencer and the theory guide bypass it, since they already know when their notes should land.
And a second envelope. Every subtractive synth since the Minimoog has had one envelope for loudness and one for the filter, and this one only had the first. The new Mod Env module sits outside the audio path like the LFO and points at any parameter from its Drives menu. Aim it at the filter cutoff with a fast decay and every note opens bright and snaps dark, which is what a pluck is; give it a slow attack and a pad blooms; aim it at pitch with a short decay and you have an 808. Oscillators also gained a Coarse tune in semitones — the octave switch on every hardware panel — so a sub-oscillator is −12 and an FM modulator at a 3:1 ratio is +19. And a fix underneath: an LFO or envelope pointed at an oscillator's pitch was only ever wired to the module's silent template, so vibrato did nothing on a played note. Each voice now carries the same wiring.
Eight new presets, and the old Pad rebuilt. Each one is a lesson, and its button says which: Strings and the new Pad are width and slowness (three oscillators, a sub, two envelopes, chorus); Pluck and Trance Pluck are the filter envelope; E-Piano and FM Bell are FM at a harmonic ratio and an inharmonic one; Kalimba is an envelope with no sustain; Chiptune and Arp switch the arpeggiator on with a sound made for it; Reese is wide detune under a driven filter; 808 is the Mod Env driving pitch; Dub Stab is a chord left to a long delay. Load one, change one thing, and listen for what moved.
The Synth now shows you what to play. Knowing what a scale is and knowing which keys it lives on are different things, and the second one is what you need with your hands on a keyboard. Press ♪ Guide in the piano bar, pick a root and a scale or mode, and every key that belongs lights up while the rest fall back. Play the lit ones and nothing is wrong; the amber key is home. Labels switch between note names and scale degrees, and ▶ Hear it runs the scale through whatever you have patched.
It also knows the chords in a key — the seven the Theory page builds, with their roman numerals — and any chord on any root, from triads to ninths. Tap one and its notes are coloured by their role: amber root, purple third, green fifth, yellow seventh. Three fingers on the coloured keys is the chord, and the purple one is the note that decides major or minor. Every chord chip plays as you tap it, so a progression is something you can hear and then copy.
The Chord Voicings page is rebuilt. It drew each voicing as a single column of numbers, which is a fret list, not a chord chart. Voicings are now real chord boxes — strings across, frets down, a finger number in every dot — and every dot is coloured by what the note is doing, the same five colours as the Synth and the Theory page. A neck map above the boxes shows every place the chord's notes live in twelve frets, and tapping a voicing lights up which of them it uses, so a shape stops being a diagram and becomes a piece of the neck. Every voicing has a ▶ to hear it, because the whole reason to learn a second one is that it sounds different.
The voicings themselves are no longer three templates. Every playable way to hold the chord is searched from the tuning — four fingers, a barre where one finger can lie flat, the played strings sitting together so a strum works — and the page picks a spread up the neck: an open shape, a low one, a middle one, a high one, and a small three- or four-string shape for playing in a band. Filters narrow it to a region. An In the key of row lists the chords that belong together, so you can tap through I – vi – IV – V and see each chord's voicings in turn. And the three pages now hand chords to each other: a chord on the Theory page opens on the Synth's keys or on a guitar neck, and back.
Seeing the Fretboard's triad section is now diagrams instead of prose. It made a claim about three-note chord shapes and then drew all three of them on one twelve-fret neck, which meant untangling them in your head before you could play any of them. They're now nine chord boxes — the vertical diagram every guitarist already reads — one shape per box, on every string set, with the position marked and each dot labelled by its interval rather than its note name.
And it now says why the shapes work, which is the part that turns nine diagrams into one idea. A triad is three notes and order is the only variable, so every box is the same C, E and G. The inversion picks your melody note, consistently: root position puts the fifth on top, first inversion the root, second inversion the third — on every string set, so "I want the root ringing on top" has an actual answer. And because the third is the only note carrying the chord's quality, moving it down one fret turns any of the nine shapes minor.
The best part is a pair of boxes that settle the B-string question by demonstration rather than explanation. They are the identical three pitches — G, then C, then E — and one is a staircase while the other is a flat barre. The only difference is whether the shape has to cross the G-to-B major third. If you have ever felt a shape "stop working" when you moved it to different strings, that is the whole reason, and it is one fret every time. Every voicing on the page was generated and then read back out of the rendered diagram to confirm the notes are what the text claims.
The Synth's keyboard has a new default, and a much simpler rule. The piano-shaped layout was faithful to a piano and confusing as a keyboard: because the black keys live on their own row, walking along one row skips notes. Pressing 5 then 6 jumps F♯ straight to G♯, past a G sitting on the row below. Same for 9 to 0. Every row had that problem — not one of the four was a chromatic run.
The default is now Chromatic rows: one long chromatic run, left to right, bottom row upwards. Neighbouring keys are neighbouring notes, with no exceptions — no skips, no repeats, nothing to remember. The rows join seamlessly too (/ to a, ' to q, ] to 1 are each one semitone), so the whole keyboard is a single ladder: 45 keys, 45 semitones, three octaves and a major third. Because no note is bound twice, every key on screen now shows exactly one letter instead of two or three.
The piano-shaped layout hasn't gone anywhere — it's Piano shape in the layout menu, and it's still the one to pick when you want chord shapes to look the way they look on a piano. That's the honest trade: one layout is easier to read as chords, the other is easier to play as a keyboard. Also fixed a cosmetic thing the longer keyboard exposed — the layout ends on G♯, and a black key's negative right margin has nothing after it to pull in, so its box hung nine pixels past the strip and put a scrollbar under a piano that already fitted.
A bug worth owning. Anyone who had ever rebound a single key on the Synth was silently locked out of every later improvement to the keyboard layout. Remapping stored your map as a frozen snapshot and the loader handed that snapshot back forever — so the day the layout gained new keys, you kept the map exactly as it stood the moment you last touched ⌨ Remap, with nothing on screen to explain why. The piano would even draw the new keys while they made no sound.
Custom is now your bindings layered over the current default instead of a copy of it. Keys you deliberately bound still win — they're applied last — and keys you never touched follow the standard layout, including anything added to it since. Only the difference is stored, not the whole resolved map, because saving the whole thing would bake that day's default into your save and re-create the same trap the next time the layout improved.
No key on the Synth's keyboard is dead any more. The two-manual layout followed the piano strictly, and a piano has no black key between E and F or between B and C — which left a, f, k, ', 1, 4 and 8 doing nothing at all. Two of those are home-row keys. A dead key under your resting fingers is a worse cost on a computer keyboard than a tidy metaphor is a benefit, so those slots now play the white note they lean against — which isn't a fudge, because E♯ is F and B♯ is C. f plays F. k plays C. Nothing under your hands is silent.
The range grew too. The three keys to the right of P continue the pattern properly — [ and ] carry the white row up to F and G, = is the F♯ between them — taking the layout from two octaves and a third to two octaves and a fifth across 45 keys, with both manuals still exact chromatic runs. Where a piano key now answers to several letters, it shows the main two and lists all of them when you hover it.
Small fix on the Synth: the audio suspended readout sat dead centre at the bottom of the screen, which put it directly over piano keys 12 to 20 — the middle-C region where the two manuals overlap, so precisely the part you play. Nudging it sideways wouldn't have held: the piano is 780px wide on a desktop and 595px on a tablet, so any fixed horizontal offset lands on keys at some width. It now floats above whatever is docked at the bottom, tracking the piano, the touch pads and the sequencer as they open and close. It also stopped intercepting clicks meant for whatever it happens to be floating over — it's a readout, it was never something you click.
The Synth's sequencer had a bug worth naming. On a phone the melodic grid and the drum grid were sized independently — steps at a fixed minimum width, drums stretched to fit — so they were two different scales sitting on top of each other. Measured at 16 steps on a 390px screen, step 15 of the melody sat 151 pixels away from step 15 of the kick, and the playhead lit two columns that had nothing to do with each other. They now share one scroller and identical column tracks; drift is zero at every width and every pattern length, and a step never shrinks below a tappable size.
Pitch now moves by scale degree. Editing a melody meant clicking −/+ a semitone at a time — seven clicks from C to G, per step, with most stops outside the key. Set Key and a scale and every edit lands in the key: drag up or down on a step and it walks the scale. A note already off-scale snaps onto it rather than staying wrong, Chromatic is still there if you want semitones, and Random now generates in your chosen key instead of always C pentatonic.
And every step has velocity. Previously every note fired at a fixed 100 and every drum hit was a boolean — which is awkward on a site whose own Groove & Timing guide argues that velocity is one of three controls separating music that breathes from music that doesn't. Melodic steps get a velocity strip you drag; drum cells have no pitch to set, so dragging up and down on a drum cell sets its velocity, which is where ghost notes live. The Beat preset now lays hats down with alternating accents rather than flat. Old saved patterns migrate to velocity 100 — exactly what they were already playing at — so nothing you had changes how it sounds.
The grid also became reachable from a keyboard for the first time: 80 cells that were plain <div>s with click handlers now take focus and report their state, with ↑/↓ for pitch and ←/→ for velocity.
The Synth now takes external gear. A new Audio In module puts a live signal from anything your computer can hear into the patch as a source — an interface with a guitar, bass or hardware synth on it, a USB mic, a drum machine, a turntable. Everything downstream then treats it exactly like an oscillator, so the filter, delay, reverb, chorus, distortion and compressor all process your instrument, and the LFO can sweep the filter while you play it. Click Enable, allow access, pick your input from the list. Nothing is uploaded or recorded — the signal goes to your speakers and nowhere else.
Two things this had to get right. Echo cancellation, noise suppression and auto gain are all switched off — they're voice-call processing, and on an instrument they gate off reverb tails, eat quiet passages and ride your dynamics away, which is most of why browser audio has a reputation for sounding bad. And feedback: speakers into a live input builds to painful in about a second, so an input that pins the master output for a sustained second has its level cut automatically. Use headphones anyway — the guard is a net, not a plan, because it fires after the noise starts.
The interesting failure was silence. An Envelope is a note gate: it sits shut until you press a key, so a continuous signal routed through one makes no sound at all, and the default patch has one. Every path was correct and the meter read nothing — which is exactly what a broken feature looks like. Rather than quietly routing around it (the way round the envelope also skips the filter, and running gear through the filter is most of the point), the module detects the gate, says so plainly, and offers a one-click fix. Deliberately putting an input through an envelope is a real technique — it lets you play a drone in and out from the keyboard — it's just never what you wanted by accident.
The Synth's computer keyboard now uses the whole keyboard. It followed Logic Pro before — A W S E D F T G Y H U J and a few more — which is seventeen keys and a tenth of range, leaving two entire rows doing nothing. The default is now the tracker layout: your keyboard read as two pianos stacked an octave apart, one per pair of rows. The Z row is white keys with S D G H J as the black keys above them; the Q row is the same shape an octave up with the number row on top. Because the rows are physically staggered, each black key sits where it would sit on a real piano.
That's 34 keys across two octaves and a third — double the range, and enough that both hands can play at once without ever touching the octave control. Octave and velocity moved to the arrow keys so Z X C V are free to be notes. If you had the old layout in your fingers, a Two rows option in the new layout menu restores it exactly, and anything you bound by hand with ⌨ Remap is kept as Custom — so you can flip between a preset and your own map without losing either.
New tool: the Harmony Workbench. Most chord tools hand you symbols. This one shows you the voices — because a progression doesn't sound classical because of which chords it uses, but because of how the inner lines move between them. Build a progression and it draws all four parts (S/A/T/B), colouring each connection by whether that voice held, stepped or leapt. Load the descending lament and you can see the whole thing move in parallel steps, which is the entire reason it sounds the way it does.
The suggestions aren't a lookup table — each candidate chord is actually voice-led and scored on the motion that results, so when it recommends ♭VImaj7 after i and says "3 voices hold," it has computed that Am and Fmaj7 share three notes. The palette is stocked for the darkwave and classically-inflected vocabulary: suspensions, added ninths, minor-major sevenths, the Neapolitan, and the borrowed ♭III/♭VI/♭VII. It spells enharmonically by function too — the third of E major shows as G♯, not A♭, because it's a leading tone.
Two guides alongside it. Writing Harmony explains the moves — why holding notes matters more than choosing chords, what a suspension is actually delaying, the borrowed chords of natural minor, the descending lament bass and the chromatic inner line, pedal points, and what Pieter Nooten specifically does in Clan of Xymox (natural minor with the leading tone absent, added ninths as a default, inner voices carrying the tune, and very few chords held for a very long time).
And The 4AD Sound — how a record label became an adjective. Ivo Watts-Russell's curation as the creative act, Blackwing Studios and John Fryer, and the defining production idea: reverb as an instrument rather than a finishing touch — printed, committed, timed to the tempo, and often louder than the source feeding it. Plus guitar used as texture rather than riff, the voice mixed as one colour among many, and Vaughan Oliver's sleeves as part of the same statement. It only works on sparse arrangements, which is the same argument the restraint guide makes.
The Synth now opens on a sound you can play. An empty rack meant the first thing a visitor did was press a key and hear nothing. It now starts on a Prophet-style patch — two sawtooth oscillators detuned seven cents against each other into a resonant lowpass, shaped by an envelope and widened with a slow chorus. That detuning is where most of the warmth in a classic analogue poly comes from, and it's why that instrument sounds the way it does. Presets are still one click away, and a shared patch link still overrides it.
Fixed: module cards were cut off on phones. Slider value readouts were being pushed clean off the right edge of the card, so you couldn't see what any control was set to. The cause was a flexbox trap — a range input has an intrinsic width of about 129px, and flex: 1 won't shrink a flex item below its automatic minimum width, so the slider held the row open at 220px inside a 132px card. Everything now fits, with the sliders keeping usable travel.
Also refreshed the site's page titles and descriptions to match what's actually here now — the production hub still described itself as a "three-part guide" when it's grown to ten, and the synth's description predated the touch pads, the drum machine and the auto-routing. They're now written around the problems people actually search for rather than as feature lists.
New: DJing on the Prime GO+. First DJ guide on the site, written around the Denon DJ Prime GO+ — a two-deck standalone system with stem separation and a six-hour battery that needs no laptop at all. Half of it is getting the hardware out of your way: preparing a library in Engine DJ, why beatgrids are the single highest-value prep task (a wrong grid drifts mid-mix and there's nothing you can do live), cueing the same four points on every track so your hands stop reading the screen, and rendering stems ahead of time rather than during a set.
The other half is craft that transfers to any gear. Sync solves the easy problem — phrasing is the hard one, and no machine does it for you: dance music is built in fours, and a mix starting on the wrong bar feels wrong even when the beats line up perfectly. Also: pulling the outgoing bass so only one track owns the low end, why a set of nothing but long blends has no punctuation, using fewer effects than you want to, the energy arc and why starting too high is the commonest mistake there is, harmonic mixing as useful-not-sacred, a pre-gig checklist, and recording every set — including practice ones, since the mistakes you miss while playing are painfully obvious the next day.
On stems specifically, the honest version: separation is an estimate, not surgery. It's brilliant for killing an incoming track's drums so you can blend over the outgoing beat, or removing a clashing bassline so two records coexist — and the best stem move is almost always removing something so two tracks fit, not building a live mashup. Subtraction sounds intentional; addition usually sounds cluttered.
New: Seeing the Fretboard. The Fretboard Explorer could always show you notes and intervals across tunings, but it never explained how to read any of it. This does. The premise: most guitarists hit a wall they can't name — they can play what they've memorised and nothing else — and the wall is that the fretboard gets taught as a set of positions to remember when it's really a map of relationships. A 22-fret guitar has 138 note positions for 44 distinct pitches, so memorising positions means memorising the same information three times over.
The most useful section is octave shapes, which most players are never taught at all: from any note, the same note an octave up is two strings and two frets away — or three frets, once the B string is involved. Learn those and you only ever have to memorise two strings, because the shapes give you every other note for free. That's an afternoon's work, not a year's. Also in there: why the B string is the only irregularity on the instrument and how it explains every shape that "stops working" higher up, position as a tone choice rather than a lookup, escaping the pentatonic box by playing along a string instead of across, triads on three-string sets (how professionals actually comp), targeting chord tones instead of running scales, and what an alternate tuning really changes. Every fretboard diagram is generated from the tuning itself, so the notes on them can't be wrong.
The Synth now has a real instrument on phones. Musical typing needs a computer keyboard, and a phone hasn't got one — which left the piano strip, whose keys are narrower than a fingertip. Open the synth on a phone or tablet now and you get touch pads instead: a grid of large multi-touch targets, locked to a scale so there is no wrong pad. Pick a root and a scale (it starts on C minor pentatonic, the most forgiving there is), and the amber pads mark the root so you always know where you are.
The layout does something a single row of keys can't: each row starts two scale steps above the one below it, so three pads stacked vertically are a chord. Put three fingers down in a column and you get one. Slide a finger across the grid and it retriggers every pad it crosses, which is how you play a run with one thumb — and hitting the top of a pad plays harder than the bottom, so there are dynamics on a surface that otherwise has none. Root and scale are remembered between visits, and ◇ Pads switches back to the piano whenever you want it.
New: Space & Restraint. Everyone repeats less is more and almost nobody can act on it, because it's advice about a feeling rather than a method. This is the method — starting with why sparse music actually hits harder, which turns out to be three facts about hearing rather than a matter of taste: people can consciously track only three or four simultaneous streams, masking is physical (two sounds in the same range literally reduce each other's audibility inside your ear, which is not something EQ can undo), and contrast needs a baseline, so if every section is maximal then none of them is. The practical consequence is sharp: a part you can't clearly hear isn't making the track bigger, it's making everything else smaller.
Then silence as something you play rather than the default state music fills — the full stop before a chorus, the drop-out, reggae's missing downbeat, the unresolved phrase. Where restraint comes from, since it was arrived at independently several times: minimalism proper, ambient, modal jazz, techno, neo-soul, and dub, which gets its own section as the clearest case study in music — engineers making versions by taking things away live at the desk, which is where the breakdown, the delay throw and the filter sweep all come from. The transferable habit: perform your mutes instead of drawing them.
Eight concrete techniques for doing less — the mute test (mute each track for eight bars; if nothing gets worse, delete it, don't mute it), capping the track count, one idea at a time, stripping the verse down instead of building the chorus up, the reverse build. And an honest counterweight the slogan usually omits: when minimalism fails. Empty and spacious are different things, and the difference is entirely in what remains — the fewer elements you have, the better each one must be. A dense arrangement can carry a mediocre snare; a four-element track gives the listener nothing else to look at. That's why restraint is expensive rather than easy.
New: Rhythm & Beats. Melody gets written about endlessly and rhythm gets treated as the thing underneath it, which is backwards for most of the music anyone actually listens to. This is the note-level half — which hits go where, and why. What makes a beat good (you can find the one; three bars of repetition and a fourth that breaks it; space; two-bar question-and-answer; only one layer is busy), and the diagnostic that catches most failures: can you sing it? Real rhythms are memorable the way melodies are. The most common beginner mistake isn't simplicity — it's uniform density.
The heart of it is ten load-bearing beat patterns drawn as sixteen-box grids — the backbeat, four-on-the-floor, boom bap, trap, the funk beat with its ghost notes, reggae's one drop, tresillo, dembow, bossa nova and afrobeat. They're written in exactly the layout the synth's drum machine uses, deliberately, so you can read one here and type it in there in twenty seconds. Notice how few notes most of them contain: the ones that feel most complex are complex in their accents, not their note count.
Then the instruments, framed as jobs rather than objects — a beat is three roles, not three drums. Something marks the pulse, something marks the backbeat, something keeps the subdivision; fill them with a stomp, a clap and a shaker and you still have a beat. Ten percussion instruments and what each one actually adds (why a real shaker part has an alternating accent, the three conga tones, why a stacked clap sounds human and a single one sounds thin), the techniques that make a part sound played — ghost notes, rimshot versus cross-stick, flams, buzz rolls, heel-toe — and the four rudiments actually worth your time. If you program rather than play, the rudiments still matter: they tell you what's physically possible, and a fill nobody could stick sounds synthetic even when every note is in the right place.
It also goes somewhere most rhythm writing stops short of: clave, tresillo and polyrhythm. Clave isn't a drum part, it's a rule — an asymmetric two-bar reference the whole arrangement agrees with, and an idea that's portable to any genre. Tresillo (3-3-2) is probably the most widespread rhythmic figure in popular music worldwide and the shortest path to making a four-square beat lean. And the cheapest trick on the page: put a three-note cycle on a percussion layer over a 4/4 beat, and it returns to the downbeat every three bars instead of every one — four bars of an otherwise static loop suddenly have a shape. Closes with nine rhythm exercises, none of which need an instrument.
The Music Education menu is now grouped rather than being one list of fifteen links — The Path, Craft, Reference, Trainers. Same links, four short lists instead of a wall.
New: The Mixing Guide — the missing part of the path. The production series went Record → Arrange → Effects → Master, with Effects standing in for mixing. But knowing what a compressor does isn't the same skill as mixing, and it showed: the word panning did not appear anywhere on this site. Mixing now has its own guide, and it starts where a mix actually starts — the static mix, built with nothing but faders and the pan pot before a single plugin opens. Then gain staging and headroom, panning and the stereo field, the low end (who owns the sub, keeping the bottom mono, when to high-pass and when not to), subtraction as the alternative to EQ, and automation as the honest alternative to over-compression.
That guide closes with the thing underneath every other skill: how to listen. Your room, your speakers, your volume and your fatigue are all part of the signal chain, and unlike your plugins they lie to you constantly. Why untreated rooms make you bake in the opposite of the error you're hearing, what headphones are good and bad at, why you should pick one monitoring level and keep it, and how to use a reference track properly — level-matched, because a mastered record is far louder than your mix and louder always wins an A/B for reasons that have nothing to do with mixing. The same rule applies to your own plugins, and it explains a great deal of what people believe processing does for them.
New: Groove & Timing. You can write the right chords, choose the right sounds and mix them beautifully, and the track can still feel like nothing. Groove isn't a mystical quality — it's velocity, note placement and note length, three controls already sitting in your piano roll. Why 100% quantize sounds dead (and why layered hits landing on the same sample sound smaller), velocity accents and ghost notes, playing ahead of and behind the beat and what each feels like, note length as a rhythm control, how much swing is actually used in modern records, and how kick, snare and bass lock into a pocket.
New: Choosing Sounds, on the least glamorous and highest-leverage idea in production — the right sound needs almost no processing, and the wrong one can't be processed into it. Always audition with the track playing rather than in solo, because what matters isn't how a snare sounds but what it sounds like in the space left over by everything else. What you're actually judging (frequency footprint, transient, length, character, tuning, width), how to layer so each part contributes one thing, why a layered kick sometimes gets thinner and how flipping polarity fixes it, and how to stop a 4,000-sample library from turning selection into procrastination.
New: Finishing — the skill nobody teaches. Almost everyone has a folder of unfinished eight-bar loops, and not for lack of knowledge. Six reasons tracks stall and the different remedy each one needs, how to escape loop purgatory (copy the loop across three minutes first — a bad complete song can be improved anywhere, a perfect sixteen bars can only be extended), what to do when you're stuck, a workable definition of done, and how to ask for feedback that isn't "yeah, it's cool." Listeners are excellent detectors and poor diagnosticians; take the problem, ignore the prescription.
The Arrangement guide gained a section on register — arranging in the vertical. Everything else on that page is about what happens when; this is about what frequency range each part occupies and whether two of them are in the same place at the same time. It's the real cause of most muddy mixes, and no EQ can separate two parts genuinely playing the same notes in the same octave. Masking is an arrangement problem, and it's far cheaper to solve there.
The Synth now wires itself. Click Osc and you have a working instrument — an Output appears and the cable is drawn for you. Everything you add after that joins the signal path on its own: sources stack in parallel at the front, effects land at the end where they belong. Delete a module and the chain closes over the gap instead of going silent. A modular synth is a wonderful thing to understand and a miserable thing to be handed cold, and dragging four cables before you hear a note was the wrong first five minutes.
Because the order is now the interesting decision rather than the tedious one, every module in the chain got ◂ ▸ arrows to move itself along the signal path. Filter before delay is a different instrument than delay before filter, and that comparison is now two clicks instead of a re-patch. ⌗ Tidy lays the rack out left-to-right in signal order so the picture matches the sound. Drag a module anywhere you like and the auto-layout backs off for good.
The LFO can finally be pointed at things. Modulation isn't audio, so no cable could ever carry it — which meant until now the only way to get an LFO onto a filter cutoff was to load a preset that already had one. It now has a Drives menu listing every parameter in the rack, in signal order, and picking one sets a depth range that actually suits it: choose Filter · Cutoff and the sweep becomes 200–4000 Hz instead of the ±100 that would have done nothing. A dashed violet line shows what's driving what. Vibrato, wah, tremolo and auto-pan are each one menu pick.
Cables are still real and still yours — tap or drag to build parallel paths and feedback loops the automatic routing won't invent — and a cable you drew by hand is never re-routed by the automatic wiring. Two more fixes along the way: a note held while you delete or reorder a module now re-articulates through the new path instead of dropping to near-silence, and + Channel Strip splices into the rack you already have rather than standing up a second, disconnected one beside it.
The Synth's computer keyboard no longer traps you in one octave. Press Z and X — or the new −/+ buttons beside the octave readout — to move the whole keyboard across the full MIDI range, and C/V to ride velocity. The default layout also grew from an octave to a tenth (A W S E D F T G Y H U J K O L P ;), so full chords and wide voicings fit under your hands.
And because no two keyboard layouts agree, you can now bind your own keys. Click ⌨ Remap, click a piano key, press whatever key should play it — repeat as long as you like, then Esc. Your bindings save in the browser and come back on your next visit; right-click Remap to restore the default. The on-screen keys always show the letter that actually plays them, so the layout is never a guess. Notes held through an octave change now release correctly too — the note you started is the note that stops.
New: a drum machine under the Synth's sequencer. Four programmable voices — kick, snare, hat, clap — on their own grid, running in lock-step with the melodic sequence and sharing its tempo, swing, and length. Click cells to program a pattern, or hit Beat for an instant four-on-the-floor to jam over. Drum hits are scheduled directly on the audio clock rather than the animation frame, so they stay sample-accurate even when the tab is busy, and patterns save and share along with the rest of your patch.
Split the site's music menu in two. Music Tools now holds the seven things you use — Synth, Song Analyzer, Tuner, Tap BPM, Fretboard, Chord Voicings, Songwriter — while Music Education holds the ten things you read: the production, recording, arrangement, effects, mastering and Logic Pro guides, plus theory, the jam guide, ear training and the chord-tone trainer. One seventeen-item dropdown had become a wall; making instruments and lessons separate ideas means you find either one without scanning past the other.
The Song Analyzer got a much deeper brain. Feed it a YouTube link or your own audio file and it now returns a genuinely comprehensive read: precise BPM with the half/double-time alternative and a 3/4-vs-4/4 meter estimate (waltzes finally get 3-beat bars), the key with a confidence score and closest alternative, the song's structure mapped into sections you can click to jump around, its energy arc drawn over time, and the mix's tonal brightness and dynamic range. YouTube failures now explain themselves too — if a video is age-restricted or bot-checked, it tells you plainly and points you to the file-drop path, which always works.
New: Sound Dispatches — the music counterpart to AI Dispatches. A live, searchable feed pulling from ten sources across audio research (arXiv sound & audio-processing preprints), gear & software (Synthtopia, Bedroom Producers Blog, Gearnews — free plugins turn up constantly), and production analysis (CDM, Attack, MusicTech, MusicRadar, Sound on Sound). Refreshes itself automatically, filterable by category and source — so the music side of this site stays as current as the AI side, with no hand-updating.
Self-hosted the site's fonts. Cinzel, EB Garamond, and JetBrains Mono now load straight from our own server instead of Google's — so pages paint faster (no round-trip to a font CDN, no render-blocking external stylesheet, two fewer domains to connect to) and it's more private: your browser no longer pings Google just to read the site. Same type, ~136 KB, cached hard.
The Synth can now record your jam to a file. Hit Record, play, and hit it again — your performance downloads as an audio file you can keep, post, or drop into a project. Together with shareable patch links, everything you make on the synth now leaves the screen: share the sound, or capture the performance.
You can now share a Synth patch as a link. Hit Share Patch and the entire rack — every module, knob, cable, and hidden modulation — is packed into a URL and copied to your clipboard. Send it to anyone and it rebuilds your exact sound in their browser, ready to play. Sound design stops being something trapped on your screen and becomes something you can pass around.
Made the Synth easier to patch and richer to play. You can now drag a cable straight from one port to another — or keep tapping — and while you're patching, every compatible port glows and the incompatible ones dim, so a wrong connection is basically impossible. Added five new presets built around the new filter — Acid, Sub Bass, Supersaw, Growl, and Brass — bringing the one-click patch library to eighteen. Under the hood, modules now clean up every audio node they create when removed (no more slow resource creep in long sessions), and the built-in guide was updated to teach the new drive, slope, and drag-to-shape envelope.
Brought the AI Evolution timeline up to the present. It now names the 2026 generation turnover — Anthropic's Claude 5 family, OpenAI's GPT-5, Google's Gemini 3 — the arrival of routine million-token context windows, and the moment open-weight models closed to within a few months of the frontier, led mostly by Chinese labs (DeepSeek, Qwen, Kimi, GLM) while Meta stepped back from open Llama. The model comparisons throughout the page were refreshed to match. (The live news feed already keeps this-week's headlines current on its own, from ten sources.)
Gave the Synth real sound-design teeth. The Filter now has a drive stage — pre-filter saturation that lets high resonance scream instead of thinning out — a selectable slope (−12 / −24 / −48 dB per octave, gentle to brutal), a notch mode, and a wider resonance range that self-oscillates at the top. And the Envelope now draws its ADSR as a live curve you can shape by dragging the handles — attack, decay-into-sustain, release — instead of guessing four sliders; drag the graph or the sliders and the other follows. First step of a deeper synth pass: easier patching, a mod matrix, and save-your-patch next.
Fixed the typography on four pages — the AI News feed, Chord-Tone Trainer, Chord Voicings, and Ear Training — that were never loading the site's fonts, so they'd been quietly rendering in the browser's default serif instead of Cinzel and EB Garamond. They now match the rest of the site. While there, warmed up the connection to the font-file server across every page so text paints a touch faster on first load.
Tuned the site's caching so repeat visits are near-instant. Every piece of the site's own JavaScript now carries a version stamp, which lets the browser and Cloudflare's edge cache it permanently instead of re-downloading it on every page — so once you've loaded the site, moving between pages and coming back pulls almost nothing over the network. Pinned third-party libraries are cached for a month. Fewer requests to the origin, faster loads, a higher cache-hit rate.
Fixed a quiet but real bug on the Song Analyzer: a naming clash between the page's own audio element and the site's shared script left the shared script dead on that page, so navigating to the analyzer always forced a slow full-page reload instead of an instant transition. The page's script is now properly scoped, so it coexists with the shared one — smoother navigation, and the analyzer is now a first-class citizen of the site's fast page router.
Folded the Music Theory tool's note-playback onto the shared audio-core.js engine — one source of truth for turning a note into a frequency, and a reliable resume-on-tap so the very first chord you play always sounds (some browsers start audio muted until you interact). Same ongoing pass: every interactive feature standing on the same solid foundation.
Smoothed out the Song Analyzer's play-along. The chord and bar highlighting followed the music on a coarse 200 ms timer that re-scanned and re-scrolled the whole chart every tick — now it tracks the audio clock frame-accurately and only touches the page when the current chord, bar, or second actually changes, so the highlight glides instead of stepping. Re-analyzing a song also cleanly cancels the previous request instead of leaving it running in the background. Part of the ongoing pass to make every interactive feature more stable and more fluid.
Made Tap BPM smarter and steadier. Tap detection now uses a sliding window with median-based outlier rejection — an accidental double-tap or a skipped beat no longer wrecks the reading. And the metronome was rebuilt from a drifting setInterval to lookahead scheduling on the audio clock, so it's sample-accurate and dead-on instead of slowly sliding out of time.
Rebuilt the Tuner's engine for stability and fluidity. Pitch detection moved from a naive O(n²) autocorrelation recomputed on every animation frame to a fast FFT-based McLeod pitch method (~0.4 ms per read, verified accurate to a fraction of a cent), decoupled from the paint loop, median-smoothed so the needle sits still, and paused when the tab is hidden. The DSP now lives in a shared audio-core.js — one fast, correct place for pitch/note math and audio lifecycle that every tool can build on. First step of an ongoing pass to make every interactive feature more stable and more fluid.
Taught buffer size & latency properly in the Logic Pro reference — what the I/O buffer actually is, the latency-vs-CPU trade-off (drawn), and why you record on a small buffer but mix and especially master on a large one (heavy plugins, and latency never touches the offline bounce). Cross-linked from the mastering prep.
Added stylized Logic Pro interface recreations to the guides — an on-brand Channel EQ (analyzer, curve & band handles), Compressor (gain-reduction meter & knobs), and Loudness Meter (integrated LUFS & true peak), drawn as SVG so they match the site and stay crisp at any size.
New Mastering Guide — the final stage, added as Part 4 of the production path: what mastering is (and isn't), how to prep the mix, the mastering chain in order, gentle EQ & glue, and loudness done right — LUFS targets, true-peak limiting, and why crushing a track flat just gets it turned back down. With metering, references, the dithered bounce, and Logic's tools (Adaptive Limiter, Multipressor, Linear Phase EQ, Loudness Meter, Mastering Assistant).
Tied the production guides into one Learn Production hub — a visual, free path from a raw idea to a finished master: record → arrange → mix → master, each part cross-linked with a next-step trail so you can read straight through or jump to what you need.
New Logic Pro reference — the shortcuts, tools, and features that make music: the main window laid out, the key commands worth memorizing (transport, editing, windows, navigation), the toolbox, and the feature set (Drummer, Flex Time & Pitch, Smart Tempo, quantize, Apple Loops, automation, Track Stacks, Live Loops). The machine behind the production series.
Housekeeping for reliability & reach: old removed pages now 301-redirect instead of returning 404 (clearing a Search Console indexing error), and static assets are cached at Cloudflare's edge for faster repeat loads and less origin traffic.
New Recording Guide — a visual reference for capturing sound: microphone types and polar patterns (drawn, not just described), mic placement and the proximity effect, stereo miking (XY / ORTF / spaced / Mid-Side), DI & reamping signal flow, how compression actually works, and punch-in / overdub / comping — with Logic Pro X workflow throughout.
New Effects Guide — the companion to Recording: what every audio effect actually does (EQ, dynamics, reverb & delay, modulation, saturation), the order they belong in the signal chain, which effects pair together, and a real method for practicing mixing and training your ears. Fifteen generated diagrams — EQ curves, a reverb impulse, delay trains, comb-filter notches, harmonic spectra — with Logic Pro X workflow throughout.
New Arrangement Guide — how to structure a song and build that structure in Logic Pro X: the anatomy of sections and the energy arc (mapped visually), blocking it out with the Arrangement track, moving whole sections as blocks, arranging energy by adding and subtracting parts, transitions (fills, risers, filter sweeps, cutouts), and a workflow of habits and key commands.
Taught the Song Analyzer to stop tripping YouTube's rate limits: fresh downloads are now deliberately spaced out and capped per hour, so a burst of requests queues into a polite trickle instead of burning the server's download budget — and when the budget runs low, you get an honest message instead of a mysterious failure.
Every song ever analyzed is now a visible Library on the Song Analyzer — one tap replays it instantly with chords, chart, and audio, at zero YouTube cost. The cache was always permanent; now it's browsable.
Added a no-YouTube path to the Song Analyzer: drop any audio file (MP3, M4A, WAV, FLAC) straight onto the page and it's analyzed by the same engine — no downloads, no rate limits, and identical files are recognized and served from cache no matter who uploads them.
Full site hygiene pass: the sitemap now covers all 35 indexable pages (it listed 16 — every chronicle, the Fretboard, Jam Guide, Ear Training, Chord-Tone Trainer, Chord Voicings, and AI pages were invisible to search engines) with real last-modified dates; every page carries a canonical URL; and five tool pages that had no social-preview cards now render properly when shared.
Retired two orphaned early versions of the Song Analyzer that nothing linked to but search engines could still find — both now redirect to the real Song Analyzer instead of competing with it under the same title.
Gave the Song Analyzer a live progress bar — the backend now streams real stage-by-stage updates (downloading %, decoding, detecting) instead of one opaque spinner, so you can watch it work.
Made chord detection sharper and more musical: it's now beat-synchronous (chords land on the beat, not arbitrary 2-second windows), adds dominant- and minor-7th colors on top of major/minor triads when the 7th is genuinely played, and smooths over half-bars so the chart tracks the real progression instead of flickering.
Added shareable ?v= permalinks and a "copy share link" button to the Song Analyzer — send someone a link and it reopens that exact analysis instantly from cache. Plus EBU R128 loudness normalization so every song plays back at a consistent volume.
Replaced the Song Analyzer's YouTube embed with self-hosted playback — the embed kept hitting YouTube's viewer-side "confirm you're not a bot" prompt, which no server-side setting can fix. The backend now serves the audio it already downloads, and the page plays it with a native audio player synced to the chord timeline. No embed, no bot-check, no ads.
Added a Windows-Media-Player-style spectrum visualizer to the Song Analyzer's play-along panel — now that playback runs through our own audio element, a live frequency display dances with the music (and gracefully sits out on browsers that don't support Web Audio).
Fixed the Song Analyzer failing through the live service while working in every manual test — the root cause was the worker's output piping into journald, whose rate-limiting blocked yt-dlp's progress writes and froze the download. Suppressed the progress output and routed the service's logs to a file instead.
Made the Song Analyzer hold up under simultaneous use: a cross-process concurrency limit now lets overlapping requests from multiple devices queue and run cleanly one at a time instead of dogpiling the box into timeouts.
Sped up a fresh analysis from ~20s to ~12s — the backend now pulls a smaller audio stream, has ffmpeg emit the exact 22 kHz mono the analysis needs, and computes the chroma transform once instead of twice. Repeats of the same song still return instantly from cache.
Made the Song Analyzer's embedded player start loading the moment you hit analyze — in parallel with the analysis instead of only afterward — and switched it to YouTube's no-cookie host to cut down on the embed's "confirm you're not a bot" prompts.
Added optional cookie support to the Song Analyzer backend so it can ride past YouTube's "Sign in to confirm you're not a bot" block — when a real browser session's cookies are present, yt-dlp uses them instead of looking like a bare script.
Closed the gap between "switched to gunicorn" and "gunicorn is actually installed" on the physical box running Song Analyzer — the systemd service was pointed at a binary that had never made it into the venv, which is why restarts were silently dying instead of serving requests.
Made the Song Analyzer host survive a real reboot, not just a restart: WSL2 now auto-launches at Windows startup instead of waiting for someone to open a terminal, and the machine no longer drops to sleep and looks indistinguishable from powered-off.
Added Song Analyzer and Fretboard to the Music nav dropdown — both tools existed and worked, they just weren't actually linked from the menu.
Moved Chronicles in the nav so it sits below Changelog instead of right after Home.
Hardened the Song Analyzer backend for real traffic: results are now cached per video (repeat lookups are instant), the service runs under gunicorn with multiple workers instead of a single-request dev server, and concurrent requests for the same uncached song share one analysis instead of duplicating the work.
Song Analyzer now runs real audio analysis instead of mock data — a self-hosted backend downloads the track and uses actual chord/key/tempo detection (librosa) rather than generating placeholder chords.
Rebuilt Song Analyzer's look and feel to match the rest of the site instead of its old orange fantasy theme, and added a real play-along mode: hit play and the YouTube video plays right on the page with the current chord lighting up in sync, Chordify-style. Click any chord in the timeline to jump straight to that moment.
Moved the "suite of tools" byline from beneath the header epigraph to sit directly under the homepage's intro paragraph, where it reads as a quiet footnote to that line rather than a second tagline crowding the title.
Moved Network out of the Systems nav dropdown sitewide — like Studio, it describes real personal infrastructure rather than a generic tool, so it's now linked as "My Network" from the About page instead.
Moved the Studio gear list out of the homepage's Build section — it's a personal rundown, not an interactive tool, so it now lives as "My Studio" under About instead. Filled in studio.html's nav, which had shipped empty and left direct visits to the page with no way back.
Toned the "Methodictruth" wordmark down from grandiose to sleek: tightened the letter-spacing, pulled back the big diffuse purple halo to a tighter glow, and ran the fill through the site's existing stone-grain filter so it reads textured rather than glassy. Moved the "suite of tools" byline below the epigraph and faded it further, so it sits as a quiet footnote instead of competing with the title.
Re-set the "Methodictruth" wordmark in Cinzel, the same classical face already used for the nav and headings, for a more proper, less novelty-font feel. Folded the standalone Directory page into the homepage — every tool and page already lived there, so the directory link now redirects straight to the homepage; the directory's search/filter box is retired along with it.
Decluttered the homepage sigil: pulled the rotating cyan diamond/triangle tick-marks, the Roman-numeral compass labels, and the time/computation/knowledge icons that were crowding the hands. Re-accented the "Methodictruth" wordmark from amber to a parchment-to-deep-purple gradient.
Swapped the sigil's hand-coded SVG hands for the real reference artwork: cropped and alpha-keyed the robot/human hands into a transparent sigil-hands.webp, layering the time/computation/knowledge icons on top instead of drawing the hands by hand. Moved the knowledge icon clear of the new hand geometry so nothing overlaps.
Redrew the sigil's two hands so they actually read as hands: the human side is now a closed fist with one finger extended toward the spark, and the AI side got visible knuckle joints down each segmented finger. Warmed the "Methodictruth" title from its old parchment-to-cyan gradient to parchment-to-amber, so the wordmark glows the same warm tone as the ancient hand.
Tucked three small cyan icons into the gap around the homepage sigil's reaching hands — a sand timer, a graphics-card chip, and an open book — standing in for time, computation, and knowledge, drawn in the same line-art language as the AI hand so they read as quiet detail rather than clutter.
Replaced the homepage sigil's old solid hourglass shape with a small scene: a cave-stencil hand reaching for an angular AI circuit-claw, the gap between their fingertips lit by the sigil's glowing core — same spirit as Michelangelo's reaching hands, redrawn in the site's own ancient-meets-circuit language, grain and all.
Archived Live Room — pulled it from the nav, the homepage, and the directory, and de-listed it from the sitemap, since it wasn't fully debugged. The page itself is untouched and the bring-back is a one-line revert whenever it's ready. Renamed the homepage's now-Live-Room-less "Connect" section to "Build" since it was left holding just the Studio card.
Removed Topology from the site nav — the page is still up, just no longer linked from the menu.
Redid the homepage hero: the heading now spells out the full Methodictruth, set in a gothic display face instead of the generic site font, and the old temple-quote epigraph and intro line gave way to new copy about the rush to understand and keep pace with AI before it outpaces you. The empty spot left at the center of the rotating sigil since the crystal-ball emoji was removed now holds a small pulsing diamond core — ancient sigil, futuristic glyph.
Redesigned Synth into one cohesive rack: every module is now color-coded by role (source, shape, mod, mix, space FX, out) across both the toolbar and the canvas card borders, a one-click + Channel Strip button drops a pre-cabled Pan → EQ → Compressor → Limiter → Output chain, every Output module shows a live VU meter, and the canvas got a subtle engineering-grid backdrop — turning "synth + mixer + signal" into a single, legible sound creation engine.
Dropped the "hacker-familiar · born among the ruins" footer line site-wide for something more solemn and to the point, and rewrote two first-person lines on Network that had the agent speaking of itself as "I" instead of describing it plainly.
Fixed the Synth's Envelope (ADSR) module — it used to trigger once when added and sit at its sustain level forever, so patching it in had no real effect. It now retriggers on every note, from the piano keys, MIDI-in, and the step sequencer alike, so a short Decay actually gives plucky, percussive steps.
Added a section to the vault on what building this site would take without an AI collaborator — skill domains, what each dependency actually demands to understand, and solo-vs-team time estimates.
Mobile pass across the whole site: added theme-color and safe-area-aware viewport meta to every page so the nav and notch-area content respect the iPhone's Dynamic Island and home indicator, bumped three iOS-zoom-triggering input/select font sizes up to 16px, enlarged a sub-44px close button to a real tap target, added :active states to the nav for touch feedback, and removed pinch-zoom-blocking viewport flags on Guitar and the vault gate.
Added three sections to AI Evolution with hand-drawn diagrams: why GPUs beat CPUs for AI and where VRAM actually goes during training vs. inference, how MCP "context servers" let agents reach real tools and data, and how corporations are routing work between frontier APIs and self-hosted models through a governed layer.
Added a Cadence & Resolution Guide to Theory: pick any chord in your key and see where it naturally wants to resolve next, with named cadences (Authentic, Half, Plagal, Deceptive, Leading-Tone), audio playback, glowing pulsing targets and arrows showing the pull, and a click-through progression builder with undo/clear and play-back.
Made Songwriter teach craft, not just terms: a step-by-step "How to Actually Write the Lyrics" method, a new Harmony & Dynamics panel on chord moves for lift and how to engineer a real crescendo, and an Energy Arc that now reads your own arrangement — drop a verse-to-chorus jump or fall by enough and it flags the transition right in the timeline with a concrete move to try.
Added MIDI output to Synth — pick a virtual MIDI port (e.g. macOS IAC Driver) from the new MIDI Out dropdown and every note played here, from the piano keys to the step sequencer, gets sent out live to Logic Pro X or any other DAW listening on that port.
Took down the Guestbook and the Salon (and its rooms — the Agora, the Gallery, the Library, the Workshop). Good ideas, but empty rooms — nobody was signing the wall or sitting in the salon. Logged here, not deployed.
Folded the standalone Mixer and Signal pages into Synth — they're now patchable Audio File, Pan, EQ, Compressor, Chorus, and Limiter modules on the same canvas as the oscillators and filters, so you can build a synth patch and a multi-track mix in one place. The old pages now redirect there.
Added AI Evolution: a history of AI from the Dartmouth workshop to now, a breakdown of how closed/open-weight/local models actually differ, a local-hosting-vs-API-key tradeoff table, explainers on DeepSeek, Ollama, and OpenClaw, and a live feed of advances, policy, and market news.
Added a searchable directory of every tool and page, a skip-to-content link and better keyboard support across the site, and this changelog. Filled in missing page descriptions and gave the site a proper 404 page.
Moved the step sequencer out of its own page and into the synth itself — it now plays through your actual patch instead of a separate fixed voice.
Made the site's ambient atmosphere calmer and more mysterious: a slow breathing nebula, drifting motes of light, a softer vignette.
Fixed the mobile hamburger nav dropdown, which had silently broken on several pages.
Added a Jam Guide: circle of fifths, Nashville numbers, capo charts, and jam etiquette.
Added FM synthesis to the modular synth, and a comprehensive AI Dispatches feed — live from 10 sources, searchable and filterable.
Added an Ear Training tool, generalized the Chord-Tone Trainer with more progressions and streak tracking, added a working rhyme finder and syllable counter to the Songwriter, and a Rhythm Trainer drill to Tap BPM.
Generated chord voicings algorithmically for every root and quality, and improved piano key contrast and label legibility in Theory.
Stopped the generative ambient music from autoplaying and hid the music player bar site-wide. Fixed the SPA router breaking onclick handlers on navigated pages.
Connected a second GPU node to the AI agent network with live inference. General site cleanup — fixed broken asset references, added a favicon.
Published The Network — a live map of the distributed AI infrastructure: remote access mesh, OS icons, character avatars for each node. Moved Network into the Systems menu.
Added the Song Analyzer — chord, key, and tempo detection from a YouTube link — connected it to a Cloudflare Worker backend, and gave it a dark, Byzantine aesthetic. Stabilized the synth's audio engine.