A game soundtrack can be perfectly saved and still be gone.
The files are intact. The melody is recognisable. The recording begins and ends in exactly the expected places.
Yet the thing once called music in play has disappeared.
Its entrances belonged to a machine, a state, a timing rule.
And sometimes a service that no longer answers.
Metaphysics, until the hardware is allowed to be literal.
The Atari 2600 could generate two tones at once.
Space Invaders made four descending bass notes work inside that arrangement.
Then made them accelerate as the enemies came down.
The score wasn’t four notes.
It was four notes borrowing urgency from a variable somewhere else in the game.
Keep counting voices. Keep assigning jobs. Keep deciding which sound has to leave so another can arrive.
The NES sound hardware offers five channels.
Two pulse generators, a triangle, a noise channel, and one for sample playback.
The Commodore 64’s SID offers three voices. The Game Boy offers four. The Genesis reaches ten, by running two different synthesis methods side by side.
Those aren’t trivia from a museum label.
They’re arrangement decisions made before anybody hears a note.
On the SID, a three-voice ceiling makes arpeggiation more than a flourish.
One voice cycles rapidly through a chord because it can’t sustain all of it.
Each voice carries its own envelope.
All three then pass through a single shared filter.
A composition can survive as code.
The code can’t be separated from the chip’s permissions.
Historically that split composition into two jobs.
A composer could write melody and harmony away from the target machine.
A sound programmer then translated it into register writes the chip could actually execute.
Some people did both.
Rob Hubbard learned machine code so he could program the SID himself.
Which is why his work sounds like an argument with the hardware.
Not a delivery to it.
A credit line can describe a tune.
It doesn’t always describe the path by which the tune became sound.
The chip is not underneath the music. It is part of the performance.
Recorded audio changed the bargain without dissolving it.
Once optical media could hold an orchestral recording or a band track, the channel ceiling stopped being the negotiation.
A recording preserves a fixed performance extremely well.
It keeps the particular breath, the sample, the mix, and the mistake.
It also made the score portable in a commercial sense.
A soundtrack could become a product.
A track list, a jewel case, and none of the game attached.
Commerce is always relieved to find a responsive system can be sold twice.
Provided it first agrees to become an album.
But an exact recording can preserve the sound and mislay the reason.
Dig Dug stops its music when the player stops moving.
The silence isn’t an absence of audio.
It’s the audio.
Frogger runs a set of short gameplay tracks.
It switches between them according to what the player has done.
The audio data is one part of either design.
The other part is the rule.
It decides when silence, this cue, or that cue is the right thing to hear.
Longer and less predictable games made that gap impossible to ignore.
The solutions all took the same shape.
A cue needs a state. A transition needs a clock. A layer needs a condition.
Remove the conditions and the saved audio stays authentic in every audible detail.
It also becomes strangely indifferent.
It plays.
It just no longer has an opinion about when.
Preservation begins, properly enough, by refusing indifference.
Dumping produces a byte-for-byte copy of a cartridge, disc or ROM chip.
Checksum verification compares that copy against a known-good reference.
A worn contact shouldn’t become a convincing but corrupted survivor.
That work is indispensable.
Without a verified image there may be nothing to run, inspect or hear.
It carries a cost that’s easy to miss, because the cost looks like success.
A complete dump feels so total that it invites a quiet equivalence.
Retaining the medium starts to sound like retaining what the medium once caused.
The dependency stack doesn’t stop at the image.
A binary may need original hardware, or accurate emulation of it.
It may need its original sound behaviour. It may need an authentication server. It may need code and data that were never on the disc.
A copy can contain every byte it was sold with.
And still lack the conditions that turned those bytes into a score.
Which makes game music an unusually strict test of what preservation means.
A film print is a fixed sequence, and an archive can hold it.
An adaptive score is a set of conditional entrances.
Holding the entrances is not the same as holding the conditions.
Preserve the output and you have a recording of one playthrough’s luck.
Preserve the procedure and you need the machine, the state, the timing rule.
And the patience to keep all three running.
The archive’s case is easy to state and hard to argue with.
Stable objects, verified copies and named dependencies are not bureaucratic decoration.
They’re the difference between a collection and a rumour.
A perfect recreation of every contingent state is not a workable collection policy.
Neither, regrettably, is asking the original machine to stay young.
The adaptive score’s case is just as strong, and it isn’t sentimental.
Its identity isn’t exhausted by its notes.
It’s Dig Dug’s silence after movement stops. The transition that waits for the end of a bar. The layer that arrives because danger did.
Preserve only the output and a responsive composition becomes evidence of itself.
Both survivals are real.
They don’t overlap as much as the word “soundtrack” suggests.
A fixed recording carries a performance across decades intact.
A runnable binary recovers a system no recording could describe.
A full reconstruction wants source, tools, hardware behaviour, state logic, services.
No surviving collection holds all of it.
What counts as the music once its play context is gone?