Archiving Policy

Archiving moves files nobody has touched for a set time out of a customer's live storage into cheap storage, and records where each one went. Two explicit acts: mark the folders, then run a job. Nothing is removed until the copy is proven and catalogued, and not even then unless the job's removal switch is on.

1. Mark the folders #

Switch archiving on for the connection. Storage → the target → ⋮ → Archiving… → on. Until this is on, no folder in the target can be marked, so nobody tags a folder on a connection where archiving was never intended.
Browse the target and use Archiving rule on a folder: Archive after N days, months or years — judged by last modified, or by last opened where the source reports it — or Never archive. Quick picks: 6 months, 1, 3, 6, 7 years.
Let inheritance do the rest. A rule covers everything beneath it; the nearest rule wins. Mark Finance for 3 years and every subfolder follows; mark Finance/Contracts as Never and that subtree is protected. Badges show the rule in force on each folder — solid where it is set here, faded where it is inherited, with the ancestor named on hover.

Switching archiving off for a connection pauses everything but keeps the rules, so turning it back on restores exactly what was marked.

Marking folders from Windows #

The people who know which folders are finished are usually the customer's own staff. On a Windows 11 PC with the connection mounted as a Caryvane drive, they can set a folder's rule from Explorer:

Allow it for the connection. Storage → the target → ⋮ → Archiving… → tick Allow marking folders from Windows → Save. It is off by default, and only takes effect while archiving is on.
On the PC, right-click a folder on the drive → Archiving…. A small Caryvane window shows the rule the folder follows (its own, or the folder it inherits from) and offers: archive files older than N days, months or years; never archive this folder; or remove the folder's own rule. Archiving asks for confirmation.

What to expect:

Only where it is allowed. On a drive whose connection has archiving off, Explorer shows nothing about archiving at all. With archiving on but Windows marking not allowed, the window shows the rule read-only. The menu appears within a few minutes of the switch being changed.
One folder at a time. The menu hides when several items are selected, so nobody believes they marked four folders when only one changed.
By drive letter or network path. It works the same on R:\Projects and on \\Caryvane\<drive>\Projects.
Who did it is recorded. A mark set from Windows names the Windows user and the PC in the audit trail, and on the folder's badge in the console.
Marking still moves nothing. The Archiving job below does the moving, and removal from the source still needs its own switch.
Windows 10 does not have the newer Explorer menu, so the entry appears on Windows 11 only.

2. Create the Archiving job #

Jobs → New Job → Archiving (or the Archive old files template). It always runs on the Cloud Agent: the marked connection is the source, any storage target the archive, an optional prefix, weekly on Sunday at 03:00 by default. Leave the containers list empty to sweep every container that has a rule.

The Options step leads with the one destructive switch, Remove archived files from the source, which is off until you turn it on. Run the job that way first: everything is copied and catalogued, nothing is removed, and you can inspect the catalogue and the archive before the first real removal.

What a run does #

For every file under a marked folder (subtrees nothing can select are not even listed), the sweep judges age against the nearest rule using the file's own listing metadata — no file is opened to decide — and for each eligible file, in this order:

Copy it to the archive under a deterministic key: <prefix>/src<target id>/<container>/<path>, so a retried run lands on the same key rather than duplicating.
Verify the copy: matching content hash when both sides report the same kind; else matching size; else presence. A mismatch fails the file and the original is untouched.
Catalogue it — the console records source, destination, size, date, the rule and proof method. If the catalogue write fails, the file counts as failed and the original stays.
Remove the original — only with the switch on, only with hash or size proof (a copy proven merely to exist is never enough), and only after the catalogue has it. The catalogue entry is then updated to source removed.

The run log names every archived, kept and failed file with the reason. A file with no modified date cannot be judged and is left alone.

Getting a file back #

Archived files are retrieved from the catalogue in the console — search by original path or name, download — not from the original folder, which no longer has them once removal is on. Tell the customer where their old files went before the first removal run. The archive is ordinary storage: the file is also there under the key above for anything that can read the bucket.

Which sources #

Storage connections the Cloud Agent can browse with dates: SharePoint first (its listings carry modified dates and SHA-256 hashes), then S3-style, Azure, Egnyte, Dropbox, OneDrive and Google Drive targets. Local folders and SMB shares — an agent-run sweep — are the next phase. Marking folders from Windows Explorer is described above.