How it works

Grid Radio is a radio station whose only artist is the Gridcoin blockchain. Every block the network stakes is compiled into a short piece of ambient music, using nothing but its hash, and put on air. This page explains the machinery: how 32 bytes become a tune, why the tune is never ugly, and why it will play the same tune in twenty years.

One artist, no playlist

A block hash is 32 bytes of publicly verifiable noise. It was produced by consensus, nobody chose it, nobody can predict the next one, and everybody on the network agrees on it forever. As raw material for generative music this is close to ideal: an inexhaustible, incorruptible, slightly pompous composer who publishes a new work every 90 seconds and never takes requests.

The chain has been staking blocks since 2013: several million works, all composed before this station existed. Grid Radio did not add music to the blockchain. It added a receiver.

The broadcast standard

The mapping from hash to score is a frozen pure function, versioned as gridradio:v1. Freezing it is the whole point: change one constant and every block in history changes its tune, which would make the station a liar. Spec versions activate at a block height, the way the chain versions its own consensus rules: a block is always compiled under the spec of its era, so a later spec can only ever govern blocks staked after its activation height. Nothing before that height can be retuned.

The sketch and the streams

The hash is read two ways. A handful of identity decisions come straight from fixed bytes, and they are the same for every piece a block could play:

b0        key       b0 % 12 semitones above A
b1        mode      one of nine scales: five diatonic church
                    modes (two thirds of blocks), four
                    pentatonics (one third)
b2        tempo     30 + (b2 % 61) BPM for the band and the
                    solo piano; every other piece sets its own
b3        drone     detune in cents; top bit adds the fifth
b16       form      3, 4, 6, or 8 bars for the band and the
                    solo piano; every other piece runs 8 bars
b29..b31  title     b29 picks one of six grammars, b30 the
                    adjective and its contrast class, b31 the
                    noun; verb, gerund and preposition come
                    from b8..b12

Everything else is drawn from named streams. A stream is a small random generator seeded by the hash and a label, so piece, harmony, casting, tuning and the rest are each their own sequence of draws, and a draw added to one can never shift another. The first draw decides which of eight pieces the block plays (about one block in eleven is a solo piano by fixed bytes instead). Each piece then draws its own material from its own stream, in a fixed order. Two pieces read their phrases from the bytes, the band (its bass line and bells too) and the canon’s theme; everything around them, the chords, the density, the canon’s voices and their delay, comes from streams. In the performance the casting, the bells, the tuning, the synthesised timbres and the time feel are drawn the same way, while a few performance decisions read bytes directly: which drone build, which bass seat, whether the bells or the floor show up at all, the room, the sky and the dial. Byte or stream, it is all the hash and nothing else.

The nine scales

Byte 1 picks the score’s scale, and with it the mood the interface reports as sky conditions:

0  major pentatonic   0 2 4 7 9         "clear"
1  minor pentatonic   0 3 5 7 10        "overcast"
2  ritusen            0 2 5 7 9         "dawn"
3  hirajoshi          0 2 3 7 8         "night"
4  ionian             0 2 4 5 7 9 11    "azure"
5  dorian             0 2 3 5 7 9 10    "mist"
6  lydian             0 2 4 6 7 9 11    "aurora"
7  mixolydian         0 2 4 5 7 9 10    "amber"
8  aeolian            0 2 3 5 7 8 10    "dusk"

Two thirds of blocks sing in the seven-note church modes; a third keep the pentatonic floor. Phrygian and locrian are deliberately absent: the one fights a drone, the other has no home to cadence to.

The eight pieces

Each piece is its own generator with its own small palette, not one generator with knobs. The band takes about a third of blocks; the other six pieces split the rest evenly, about one block in ten each; the solo piano is gated by fixed bytes.

band            short phrases from a curated table, arranged
                into a bar form over an idiomatic chord
                progression; the low voice takes one of five jobs
solo piano      the band's material, both hands of one felt piano
drone and line  one sustained voice over a held floor: phrases of
                three to five long notes with a breath between
tintinnabuli    a stepwise line around the upper tonic, and under
                every note the nearest tone of the tonic triad:
                two voices, one chord, nothing else
canon           a theme that re-enters on a second voice a bar or
                two later, at the octave or the fifth, sometimes
                on a third voice after that
chorale         four voices in block chords, one chord every two
                bars, each voice stepping to the nearest chord
                tone in its own lane
process         a short cell in two struck voices and a rule: the
                second voice slips one step every few bars, or
                the cell grows and shrinks a note at a time; the
                cell never divides the form, so it continues
                rather than loops
broken chords   a figure per bar over the chord's ladder, often
                the same figure as the bar before, half the time
                over a pedal point

The band and the solo piano keep the sketch’s tempo. The others set their own: drone and line 34 to 50 BPM, chorale 40 to 50, tintinnabuli 40 to 60, broken chords 40 to 70, canon 40 to 72, process 55 to 90.

Why no wrong notes

The guarantee has two tiers. On pentatonic blocks it comes from the scale itself: pentatonic scales contain no dissonant intervals, so any subset of their notes, in any order, at any time, is consonant. This is the oldest trick in generative music (and the reason a toddler on the black keys of a piano sounds wise rather than alarming). On diatonic blocks the scale offers no such shelter, so the rules take over. Every piece is written in scale degrees, so no note can leave the key. Chord tones come from the harmony’s own spelling, which can only ever name scale tones. Phrases are anchored on the chord that is sounding, and leaps are leashed. And before any line leaves the compiler it is swept for tritone leaps, the one interval this station never jumps: every voice of a canon, every lane of a chorale, and every pair of notes in a process cell, in any octave, since a cell meets itself in every alignment. Different guarantees, same promise.

Why it sings

Notes that cannot clash still cannot sing on their own. So the band does not place notes one at a time: it speaks in phrases. The hash picks short cells from a table of contours that already read as an opening, a turn, a rise, a fall or a return, each paired with a rhythm that suits it, and arranges them into a bar form. A quarter of blocks develop a few cells through the form, repeated and varied; the rest draw a fresh cell for every bar. Each phrase is anchored on a tone of the chord underneath, no leap is wider than a fifth, and the last bar is a cadence: most blocks come home to the tonic, usually by step, and a few rest on another chord tone or stop one step short of home, which is the difference between ending and merely stopping. One block in ten is allowed a single wild note outside the vocabulary, and the next note has to recover from it. Before anyone is cast to play it, the block draws how dense its material is (spacious, flowing or quick), the composer writes to that, and the performance then casts a player who can actually speak that fast. The low voice takes one of five jobs: silent, a foundation on the chord’s own bass, an ostinato on the root and a chord tone, a slow counterline, or the lead itself, with the treble line stepping back to answer it. Every one of these decisions still comes from the hash alone.

Track titles

The last bytes name the piece, drawing on two lists of 128 science and sky words in keeping with the network’s day job. Byte 29 picks one of six small grammars. Most blocks get a sentence, “The Solemn and Tender Observatory Listens”, or a phrase, “Coasting by the Solemn Harbor”, because a sentence is far easier to carry out of the room than a word pair. The two adjectives in a pair always come from opposite poles of the same class, so a title contrasts (“The Veiled and Clear Lagoon Widens”) rather than repeating itself.

The short forms are deliberately the rare ones. “The Ion Burns” is drawn from a pool of only 4,096, so names like it recur perhaps fifty times across the chain: the standards a station needs, the songs everyone has heard twice. The long forms draw on pools of four million, which is where the distinctiveness lives: about 55% of all blocks end up with a title no other block on the chain has, and a block that does share its name usually shares it with exactly one other. An opus number (“Iron Comet, Op. 517”) still turns up on one block in sixteen, the honest cataloguing convention for a composer several million works deep.

The performance

The score is data; the station performs it. A small synthesizer of its own performs each block once, on the station’s server, through a delay and a reverb that the hash also mixes: the delay’s timing and depth vary per block (some blocks audibly echo), and the room itself ranges from tight to cavernous. That performance is kept, and every listener hears the same file, sample for sample, with the sky and the dial read from a second file made in the same pass. A new block is performed as soon as it is staked; one nobody has asked for before takes about half a minute the first time. Which parts are sampled and which are synthesised is not an accident. Anything that carries a tune is nearly always a real recorded instrument: thirteen of the fifteen lead seats, and two of the four bells, are public-domain notes from Versilian Studios, the Versilian Community Sample Library and the open sfz libraries. The two remaining lead seats are synthesised on purpose, an electric piano and a plucked string, and each block draws its own settings for them, so no two blocks play quite the same instrument. The bass sits on ten seats: six of them are four real electric basses at their softest recorded layer, four are plain oscillators. We tried sampling the bass with a bowed double bass, then a tuba, then a double bass played pizzicato before those, and every one of them had too much attack. A plucked or bowed string opens with a transient sharp enough to pull your ear onto the beat, and the bottom of this music is supposed to sit still.

The instruments

The hash casts the band, deciding both who plays and what they play it on, and it casts material first: the piece writes, the score says how short its shortest note is and how high the line sits, and the block’s stream picks a lead from the seats that can actually play it. There are fifteen lead seats. Thirteen play real recordings. The felt piano holds four of those seats. Two play it straight, because the second chair auditioned a sax, a trumpet, a synth and a kalimba and kept none of them. The tape piano plays it off a worn tape, deep in a cavern, and the mood piano follows each note with a reversed and a half-speed ghost of itself. The rest are a harp, a glockenspiel, a violin section, a cello section, an erhu, a marimba, a solo cello, a vibraphone and a cithara. Two are synthesised: an electric piano built from a hammer knock, a tine bark and two pickups beating against each other, and a plucked nylon string, and each block draws its own instrument of each family (how bright, how long it rings, where the string is picked). Each piece has its own palette of seats: a drone-and-line block sings on the tape piano, the mood piano, an erhu or a cello; a canon pairs instruments from different families so the voices can be told apart; a chorale is violins over cellos, or three vibraphones; a process piece is two struck voices. In those pieces the felt piano is where a line goes when nobody in its palette can play it. The bass is one of ten seats: six are four real electric basses at their softest recorded layer, each with a sine an octave below it for weight and a filter close to the fundamental so nothing down there can turn sharp; the other four are a sine, a triangle, a filtered saw and a filtered square. The bells are one of four characters: two are a real glockenspiel, bright and far; the third is a synthetic gong whose partials are deliberately inharmonic, which nothing samples; the fourth is an FM bell whose clang the block tunes itself. About one block in eight also tolls: a deep tubular bell, struck softly two octaves below its home, doubles the bass root once at every chord change, a buoy marking the harmony, sparse enough that each toll rings out completely before the next. The drone is built twelve ways. Nine are oscillators: warm saws, airy glass, church-still organ, a dark growl, a quiet choir, a slowly breathing pulse, a hollow vowel-like formant, two tidal layers swelling against each other, or no held tone at all but a slow piano arpeggio walking the chords. Two are recorded synthesizer stems, a shifting pad and an open fifth. The twelfth is a granular floor: the piano and cello notes already on disk, read in overlapping grains until they stop being notes and become a floor. Every oscillator drone breathes: the oscillators swell against each other at rates that never line up, and their detuning drifts, so the beating itself keeps moving rather than settling into something the ear can stop hearing. About one droned block in five is performed outdoors instead, with no held tone at all: one of eighty-two field recordings carries the floor. There is rain and wind and forest, but also fire, frogs, a cave, a subway, and a city street at night. The hash picks which one, how loud it sits, and how far away it sounds, that last one being nothing more than how far the filter is closed. Instruments also live in different registers. The glockenspiel sounds two octaves above the written line, the cello section and the solo cello an octave below, so a change of lead moves the whole tune’s centre of gravity. Those offsets are whole octaves, always: anything else would transpose the lead into a different key from the bass. And because a block led by a glockenspiel carries far less energy than one led by a piano, a broadcast leveller trims the output so every block meets the volume you set once. It changes no note, only playback gain. The roster itself varies. In a band, some hashes sit the bass out, some silence the bells, some drop the drone entirely and play naked in the room, and a third of them hire an arpeggist who spells out each bar’s chord. A block can arrive as a bare duo of drone and lead, or as a full house. About one block in eleven dismisses the band altogether and performs as solo felt piano: the melody in the right hand, the bass line in the left, nothing else, in a room turned large and soft for the occasion. On top of that the block’s transaction count only ever adds players to a band: three or more transactions add an octave echo shadowing the lead, four bring in soft brush-noise percussion, six or more swell a sustained chord pad. The other pieces keep their own instrumentation whatever the traffic: a drone-and-line block is one voice over a floor however busy the block was. Between the roster and the layers a band runs from duo to octet, all of it still a pure function of public data, so the determinism promise is untouched.

The harmony

Chord roots on random scale degrees rated worst on the station, so the harmony draws from a short list of idioms per mode instead: the progressions the ear already knows the way home from. Ionian has I-V-vi-IV and the plagal IV-I, aeolian i-VI-III-VII, mixolydian I-bVII-IV, and so on; the pentatonic modes get the root motions they can actually harmonise. The hash decides whether the harmony travels and how fast: held on one chord, a slow two-chord vamp, a chord per bar, or a quick change every half bar. Then it colours the chords. Most stay plain triads; some become sevenths, suspensions, added ninths or open fifths; three in ten are inverted so the floor takes the third or the fifth; and one moving band progression in four sits on a pedal point, the ambient device above all others, where the floor holds the tonic while the chords move over it. A chorale moves one chord every two bars, broken chords sit on a pedal half the time, and drone and line, tintinnabuli, canon and process hold the tonic throughout. Every chord is spelled on the score itself (its root, what the floor holds, its tones), and those tones can only ever be scale tones, so harmony and line agree by construction. Being drawn from the hash, a block’s harmony is as permanent as its tune.

The hash also sets the time feel: whether the offbeats swing, how many milliseconds each note strays from the grid, how far behind the beat the lead sits, whether the tempo breathes over a bar or two, and where each player stands on the stage. And it picks one of three tunings. Equal temperament is the station default. A quarter of the diatonic blocks tune the drone and the bass justly under a tempered lead, which locks the bottom of the mix without touching the melody. Pentatonic blocks and drone-and-line pieces, which have no semitone to go sour or hold one chord throughout, are tuned in just intonation seven times in ten and otherwise take the just floor.

When a new block is staked, the station segues: at the next bar line the outgoing tune fades out, ringing tails and all, while the incoming one fades in from its own first bar. If no block arrives, the current work simply plays again; the chain is occasionally between ideas.

The long programme

“Plays again” does not mean “repeats itself.” The chain is sometimes minutes between blocks, and a short form looped verbatim through a ten-minute gap would be a broken record. So each pass through the form is a variation. In a band or a solo piano block the harmony, the bass and the drone hold their ground while the melody journeys: played as written, resting for a pass, lifted an octave, sequenced up a scale step, or mirrored around its opening note. The hash picks which journeys the block takes and in what order; every cycle opens as written and closes with a reprise, four or six passes long, and once a cycle completes it comes around again a shade quieter, the tune settling into its own landscape. The other pieces are told as written, with a resting pass where the piece can bear one: a canon’s leader drops out and the followers carry on, a chorale and a tintinnabuli rest their voices and leave the floor. A process piece does not loop at all: the score carries the cell and the rule, and the player renders every later pass from them, so the two voices keep drifting through every alignment for as long as the block is on air. Every variation is a pure function of the hash and the pass number, transformed in scale-degree space with the counterpoint rules re-applied, so a varied melody still leans on the chords, still never leaps a tritone, and still cadences home. The determinism promise extends through time, not just across listeners: the fourth pass of a block sounds the same for everyone, forever.

The weather

One thing in the sound does not come from the hash: the brightness. The staking difficulty the network had when the block was staked drives the filter on the drone: a block staked while the network worked harder sounds brighter, and keeps that weather for good. The song comes from the hash; the weather comes from the chain. Neither comes from us.

The sky

Everything you see is compiled from the hash the same way the music is. Each block paints the sky its own three colors and sets its weather in motion: how crowded the mote field is, how fast it drifts, how thick the fog sits, how often a star falls. One block in four also turns that drift into a flight toward the listener, and another carries the field away into its own off-center vanishing point; the block’s sideways direction still banks both journeys. One or two signature effects from a palette of twelve ride the music (rings radiating on bass strikes, comets, orbiting satellites, aurora curtains, embers, ripples spreading from wherever a bell lands), chosen and tinted by the block. The dial the spectrum plays on is the block’s own instrument too: bars, dots, a wave, petals, needles or an eclipsed star’s corona, turning at the block’s pace in the block’s direction. And the hash decides what hangs inside the dial. About half the blocks keep a quiet void. A third hold a soft glow that breathes with the music and drifts slowly around the block’s own palette, never settling on one color. And roughly one block in eight collapses into a black hole: an event horizon, a thin photon ring, and an accretion ring that runs hot at an angle the block chose, forever. Two listeners on opposite sides of the world see the same sky.

Same hash, same tune, forever

Because the mapping is frozen and pure, every block’s page is a permanent musical record. Block 1,000,000 played the tune it plays the moment it was staked, years before anyone listened; it will still play it when everyone has stopped. There is something quietly pleasing about an archive of millions of compositions in which nothing can ever be remastered.

What this is not

There is no AI here and no machine learning. It is modular arithmetic and a small audio engine written to the Web Audio specification, a browser standard from 2013, the same year the chain started composing. The blockchain does not know it is making music, which puts it in the company of most great composers’ harshest critics. And nothing about this service is useful. That is not an apology.