Every bounce that leaves the room gets its own session file saved under the bounce's name, every track gets consolidated from bar 1, stems and the approved mix get printed as 24-bit files, and the finished folder lands on three drives before the week ends. That is the whole system. It earns its keep because a catalog does not stay asleep: the official Eurovision compilation albums for Turin 2022, Malmö 2024 and Basel 2025 each carry work of mine, and material that keeps re-entering the market is material whose archive has to answer on short notice.
Why does every delivered version get its own session file?
Because a recall means reopening the exact state that produced one specific file, and a single working session with a month of edits stacked on top cannot produce last month's bounce anymore. The rule is mechanical. After any bounce that goes to an artist, a label or a mastering engineer, the bounce goes out as song-v07.wav and the session saves as song-v07, no exceptions and no judgment calls about which versions matter. Session files are small next to audio, tens of megabytes even in a heavy production, so twenty saved versions cost less disk space than one printed chorus stem. The word final never appears in a name. Version numbers sort; final_final_NEW2 lies. Pro Tools has shipped a Save Copy In command for this exact move for decades, collecting a session and its media into one portable folder, and every serious DAW has an equivalent.
What happens to the audio before the folder goes to the shelf?
Consolidation, then commits. Every track gets rendered from bar 1, beat 1 to the end as one continuous file, so the next program to open the folder needs no edit list to place anything. Virtual instruments get committed to audio, because a synth patch is the first thing a five-year-old recall fails to load; the MIDI stays in the session next to the print and costs nothing. Prints are 24-bit WAV at the session's sample rate, and anything that touches picture or broadcast goes out as Broadcast Wave files, whose header carries a sample-accurate timestamp so a file dropped into a fresh session lands where it belongs. The same discipline covers a session leaving for another engineer, which is why this checklist shares most of its rows with the folder I send to a mix engineer.
Which prints answer a recall even if the session never opens?
Stems, plus three full mixes. My standard print set is drums, bass, music, vocals and effects as stereo stems, then the approved mix, an instrumental and an a cappella. Most of what arrives years later, a remaster, a broadcast performance version, a request to duck the lead vocal under an edit, comes out of that set without launching anything old; where the line runs between working from stems and recalling the full session is its own decision, one I mapped in when stem mastering helps. The approved bounce is the ground truth of the whole archive. Whatever a new master does, it gets checked against the original approved file by ear and by meter before it ships, because the version the audience already knows outranks anyone's memory of it.
Where do the three copies live?
By the 3-2-1 rule: three copies, on two kinds of media, with one of them offsite. In this room that means the working SSD, a spinning archive drive in a drawer, and a cloud copy that syncs overnight. The arithmetic is friendlier than people expect. A produced and mixed single, fully consolidated with committed instruments and printed stems, lands between 5 and 15 GB, so a busy year of output fits on one 4 TB drive that costs about a hundred dollars. Two habits keep the shelf honest: the archive drive gets powered up and spot-checked once a year, since spinning drives fail quietly in drawers, and the archive happens the week a master ships, never at some imagined year-end cleanup that no December has time for.
What still breaks a recall, and what covers it?
Plugins, almost every time. Companies fold, license servers retire, an OS update strands a whole format generation. Two covers. First, a plain text file in every archive folder: DAW and version, sample rate and bit depth, the plugin list with version numbers, the vocal tuning reference, and any hardware in the chain with its settings photographed. Reading that file five years later turns a scavenger hunt into a shopping list. Second, the printed stems above, which make the worst case survivable: if the session truly will not open, the stems still rebuild the record.
What does all of this cost in session time?
About ten minutes per delivered version and under an hour at archive time. The Save As takes seconds. Consolidation and instrument commits run unattended while the coffee brews. The text file is five minutes of typing while the last bounce renders. The other side of the ledger is what settles it: a recall without an archive means rebuilding a balance by ear from the released master, which costs days and still misses, and the client hears the difference against the version they have lived with.
If an old song of yours came back tomorrow
Pick the release of yours with the most streams, open its folder, and time how long it takes to find the approved mix, the stems and the session that made them. More than five minutes, and that song is this afternoon's project. Credits are at /credits, background at /about.