What We Keep After Delivery, and for How Long

The question that arrives six months late

The files get asked about after the drive has gone back to the shelf. A distributor wants a textless master, a festival wants a new DCP, a composer wants the stems for a live performance. Post production archive retention is the answer to "who still has this," and it is easier to settle before delivery than after.

The hot window after delivery

Keep the project live for a fixed stretch after delivery, then archive it. One widely read guide suggests about 60 days hot, with the archive made around day 90, and a contract window of 30 or 60 days plus a further 30 for late notes. Whatever numbers you choose, write them down. Late notes are the reason the window exists; a window nobody agreed to is just an argument waiting for a date.

The client's copy is a separate matter: how a delivery drive is formatted says nothing about what the studio keeps.

What goes in the archive, and what does not

The same guide lists the keepers: the final master or the DCP, a consolidated edit with handles (it mentions 12 or 24 frames), dialogue and M&E stems, cutdowns and social assets, and the creation files: editing projects and bins, audio and color project files, an EDL. The DCP is generally under 500GB, and a 4K master plus DCP generally stays under 1TB.

Dailies are the usual exclusion. A feature can shoot 20 to 30TB, which the guide says can multiply archive costs by 20 to 30 times, so most projects skip them. Make that call with the client, in writing. It's the one decision you can't undo.

Not everything is equal, either. The Academy's digital preservation forum makes the point that not all production data is archival and that retention priority should follow the importance of each dataset. The same is true of music: what happens to a score after the festival run ends depends on who holds the session files.

Tape, disk or SSD

The Academy paper gives the comparison plainly. Manufacturers estimate dependable LTO storage at 15 to 30 years, ideally at 16 to 25°C. Hard drives are limited by their moving parts to roughly 3 to 5 years of service life. Unpowered SSDs leak charge, and heavily used consumer drives may lose data after one to two years in storage.

The paper's migration cycles follow from that: tape every 3 to 5 years, hard drives every 2 to 3, SSDs every year. Tape is not a set-and-forget answer. Through generation 7, drives read back two generations; generation 8 and 9 drives read and write only one back. A shelf of old cartridges needs a drive that can still read them.

Copies and verification

One copy is not an archive. The Frame.io guide describes a common plan of three copies across two formats: two LTO backups stored in separate places, plus one online. It also warns against treating consumer sync services as an archive, since you don't choose which media they move to colder storage. The Academy paper adds geographic separation and regular verification, and notes that a migration is a good moment to check integrity. The habit is the same one behind redundancy on set: assume a device will fail and plan for it.

Who pays to keep a project alive

Storage is cheap. Attention isn't. The Frame.io guide puts LTO as low as $5 per terabyte compressed, with a drive under $5,000 and tapes under $100, and notes that for one or two projects a year a local vendor may cost less than owning a drive. Cold cloud storage ran about $3.50 per terabyte per month in the guide, before retrieval fees, and retrieval can take hours.

So the cost that matters is the third-year one: migrating, verifying, answering the request. We'd put a retention period in the delivery agreement, say what is kept, and name a fee for extending it. A client who wants the project alive for ten years can pay for ten years.

The working note

Short hot window, a named list of what is kept, two media in two places, and terms in writing. Then put the first migration date in the calendar. The same discipline covers the picture, the score and the mix: sessions and stems age like everything else.