The archive decision arrives at the worst possible moment - right after delivery, when the drives are full, the client is happy, and every instinct says delete. The working answer: keep the master, the project file with conform data, the paperwork, and enough source media to re-conform - on three copies, two media types, one off-site - and put the retention rules in the contract before the wrap. What follows is the wrap checklist: the keep-list, the 3-2-1 storage plan, the LTO and cloud options with verified numbers, and the cost trap at the bottom.
What to keep (and what you can drop)
The keep-list splits into four buckets. The masters: the graded, final deliverables - plus the pre-grade picture lock and the clean, un-graded version if one exists, because the grade is the hardest thing to recreate. The rebuild kit: the project file(s), the conform data (EDL/XML/AAF - the same turnover file a colorist handoff uses), the LUTs and looks, and enough consolidated source media with handles to re-conform the cut; media you cannot re-link is a master you cannot re-cut. The paperwork: contracts, music licenses and cue sheets, stock licenses (music and footage), talent and property releases. The record: review notes and approval states from the rounds - the delivery-decision trail (why the audit trail matters). Frame.io's archiving classic (Charles Haine, 2023) adds two details worth stealing: keep audio stems - dialogue plus an M&E stem, the difference between a cheap and an expensive foreign dub later - and treat dailies as the budget decision, since a feature's 20-30 TB of dailies can cost far more to store than everything else combined.
The 3-2-1 rule
The storage plan is the old standby: three copies of the data, on two different media, with one copy off-site. The wording survives in the CISA/NSA/FBI joint #StopRansomware advisory (AA23-165A, June 2023): "three copies of data (one copy of production data and two backup copies) on two different media, such as disk and tape, with one copy kept off-site." Credit for the shorthand itself goes to photographer Peter Krogh, who coined it in The DAM Book. For a finished project the rule translates to: working copy on disk, tape or second disk in the studio, and one copy off-site - cloud counts as the off-site copy if the on-site copies are physical.
Tape: LTO in 2026
LTO tape is still the archival baseline because it is cheap per terabyte, offline by default (ransomware cannot encrypt a shelf), and built to verify itself: IBM's tape documentation describes the drive performing "a read after write for verification. Incorrectly written data... is automatically rewritten by the drive in a new location." The current generation numbers, from the LTO consortium: LTO-9 holds 18 TB native / up to 45 TB compressed; LTO-10 ships two cartridge formats, 30 TB and 40 TB native, rising to 75 TB and 100 TB with compression - and notably drops backward compatibility with LTO-9 drives. Read-back generations are narrower than folklore says: per the consortium's FAQ, a drive reads its own generation and two prior up through LTO-7 - an LTO-8 drive reads 8/7/7-Type-M, and an LTO-9 drive reads only LTO-9 and LTO-8 media. Plan the restore path for the decade, not the purchase.
Cloud: the storage-class ladder
Cloud archive is a ladder of cheapness-with-patience, and the retrieval math is the part that bites. AWS's S3 storage classes (verified from Amazon's documentation and price list, November 2026): Glacier Instant Retrieval at $0.004/GB-month - millisecond access, 90-day minimum billing; Glacier Flexible Retrieval at $0.0036/GB-month - restores in 1-5 minutes (expedited) or 3-5 hours (standard), bulk retrievals free, 90-day minimum; Glacier Deep Archive at roughly a dollar per terabyte per month - restores in 12 hours (standard) or 48 (bulk), 180-day minimum. Storage vendors pitch the same tiers simpler: Backblaze B2 lists $6.95/TB/month with egress up to three times your average storage free, and Wasabi lists $7.99/TB/month with no egress fees and a 90-day minimum duration on pay-as-you-go. The honest way to pick: masters and rebuild kits on the instant-access tiers, dailies and intermediates on deep tiers you will rarely open.
Verify, then label
Two habits make an archive trustworthy years later. First, checksums: hash every archived file and store the manifest with it - SHA-256 is the safe archival choice (NIST deprecated SHA-1, disallowing it for signatures back in 2013 with full phase-out by 2030), and the on-set chain-of-custody format is the media hash list if the project already has one. A checksum manifest is what turns "the archive exists" into "the archive is intact." Second, labels and a readme: a plain-text index in the archive root saying what each folder is, which project file opens which cut, what the conform files correspond to, and the date. The archive's future reader is you, tired, in year three - write the readme for them. Naming discipline comes free from the file-naming convention.
How long, and who pays
Retention is a contract clause, not a courtesy - the wrap conversation belongs in the contract clauses (archive period, storage fees after handoff, and the deletion date), and the client's own offboarding step decides when media leaves your machines (the offboarding checklist has the handoff). Practitioner ranges run from one year (commercial spots, after final approval and payment) to indefinite for documentary and archival work where footage reuse is likely. The cost framing is the honest close: post-delivery archive is storage you hold for a client who has stopped paying monthly - price it into the job, or hand the archive to the client with the delivery package and a checksum manifest.
FAQ
What files should I keep after delivering a video?
The masters (graded finals, picture lock), the rebuild kit (project files, EDL/XML/AAF conform data, LUTs, consolidated source with handles), the paperwork (contracts, music and stock licenses, releases, cue sheets), and the round record (review notes and approval states). Audio stems - especially an M&E - belong with the masters.
Is LTO tape still worth it in 2026?
For terabyte-scale archives, yes: lowest cost per TB, offline-by-default security, and drive-level read-after-write verification. The caveat is generations - an LTO-9 drive reads only LTO-9/8 media and LTO-10 breaks compatibility further, so plan the migration path when you buy.
How much does cloud archiving cost?
Verified November 2026: AWS Glacier Instant Retrieval $0.004/GB-month with millisecond access; Glacier Flexible $0.0036 with 3-5 hour standard restores; Deep Archive around $1/TB/month with 12-48 hour restores. Backblaze B2 is $6.95/TB-month; Wasabi $7.99/TB-month with no egress fees. Masters on fast tiers, dailies on deep ones.
How long should I keep a client's project archived?
Whatever the contract says - one year past final delivery is a common commercial default, indefinite for work likely to be revised or re-licensed. Put the period, the storage fee after handoff, and the deletion date in the contract clauses so the wrap is paperwork, not a negotiation.
Should I archive the dailies?
Only with a budget decision attached: raw camera media is the bulk of the archive and the part least likely to be needed. Keep it when reshoots, re-edits, or archival licensing are plausible; otherwise archive selects and consolidated media with handles, and write the decision down in the readme.
How do I prove the archive has not rotted?
Checksums and periodic verification: hash every file at archive time (SHA-256 is the safe default), store the manifest alongside the data, and re-verify against the manifest on a schedule. LTO drives verify at write; the manifest is what verifies year three.