Merge feature/remote-id-hashing-and-resilient-list

Two related changes, both addressing leakage and brittleness around
the public bytes the destination store sees.

First, every remote storage path that previously embedded a human
snapshot ID (e.g. metadata/heraklion_berlin.sneak.fs.photos.2026.
catalog_2026-06-24T07:00:15Z/...) now uses the hashed remote key:

  RemoteSnapshotKey(id) = hex(SHA256(SHA256("vaultik|" + id)))

Applied at:

  * uploadSnapshotArtifacts (snapshot create write path)
  * the manifest.json.zst snapshot_id field — manifest is
    unencrypted, so the human ID would otherwise be readable to
    anyone with bucket-list permission
  * cleanupIncompleteSnapshots metadata-existence probe
  * snapshot restore / verify (downloadSnapshotDB,
    loadVerificationData)
  * downloadManifestByKey, deleteRemoteSnapshotByKey
  * CleanupLocalSnapshots reconciliation
  * the locally-driven removal paths (RemoveSnapshot,
    RemoveAllSnapshots, confirmAndExecutePurge)

The local index database keeps human IDs everywhere — the hash is a
boundary translation, not a rename. A directory listing of the
backup destination now looks like
"metadata/<64-hex>/{db.zst.age,manifest.json.zst}" with no host,
snapshot-name, or timestamp information visible.

Second, snapshot list no longer fails just because remote storage is
unreachable, and only consults the remote when the local machine can
plausibly decrypt:

  * Listing is always driven by the local index database — that's
    what holds the human IDs, timestamps, and per-snapshot stats
    that the table actually shows.
  * If no age secret key is configured, we skip remote listing
    entirely (the box is treated as a write-only backup machine —
    there's no value showing it remote-only keys it could never
    restore).
  * If a key IS configured, we try the remote listing; failures
    (volume unmounted, permission denied, network error) downgrade
    to a warning instead of aborting the command.
  * When the remote listing succeeds, we cross-reference by hashing
    each local human ID and diffing against the returned key set.
    Local-only snapshots get the existing "stale local record"
    cleanup hint; remote-only keys are surfaced as a single
    "NOTE: N remote snapshot(s) found in backup destination store
    but not in local database" line.

FileStorer construction also no longer does an eager mkdir — the
basePath is recorded and the directory is created lazily on first
write. A missing or unmounted destination during `snapshot list`
should NOT block the command, and now it doesn't.

RemoveAllSnapshots is rewritten to drive deletion from the local
index instead of from a remote listing, hashing each local ID to
find the corresponding remote key. Orphan remote keys (no matching
local snapshot) are handled separately and only deleted when
--remote is set. Existing tests are updated to hash storage paths
through the new RemoteSnapshotKey helper.

The hash format is a hard pre-1.0 break: existing remote snapshots
written under the human-ID path scheme are no longer readable; they
need to be either re-uploaded under the new scheme or manually
renamed. There is no fallback path; matching the project policy of
"no migrations pre-1.0."

README.md

@@ -78,7 +78,7 @@ vaultik snapshot verify <snapshot-id>
 VAULTIK_AGE_SECRET_KEY='AGE-SECRET-KEY-...' vaultik snapshot verify --deep <snapshot-id>
 
 # restore (requires the private key)
-VAULTIK_AGE_SECRET_KEY='AGE-SECRET-KEY-...' vaultik restore <snapshot-id> /tmp/restored
+VAULTIK_AGE_SECRET_KEY='AGE-SECRET-KEY-...' vaultik snapshot restore <snapshot-id> /tmp/restored
 
 # daily cron job: back up, keep a 4-week rolling window of snapshots
 # 0 3 * * * vaultik snapshot create --cron --prune --keep-newer-than 4w
@@ -95,17 +95,17 @@ vaultik [--config <path>] config init
 vaultik [--config <path>] config edit
 vaultik [--config <path>] config get <key>
 vaultik [--config <path>] config set <key> <value>
-vaultik [--config <path>] snapshot create [snapshot-names...] [--cron] [--prune] [--keep-newer-than <duration>] [--skip-errors]
+vaultik [--config <path>] snapshot create [snapshot-names...] [--cron] [--prune] [--keep-newer-than <duration>]
 vaultik [--config <path>] snapshot list [--json]
 vaultik [--config <path>] snapshot verify <snapshot-id> [--deep] [--json]
 vaultik [--config <path>] snapshot purge [--keep-latest | --older-than <duration>] [--snapshot <name>...] [--force]
 vaultik [--config <path>] snapshot remove <snapshot-id|--all> [--dry-run] [--force] [--remote] [--json]
-vaultik [--config <path>] snapshot prune
 vaultik [--config <path>] snapshot cleanup
-vaultik [--config <path>] restore <snapshot-id> <target-dir> [paths...] [--verify]
+vaultik [--config <path>] snapshot restore <snapshot-id> <target-dir> [paths...] [--verify]
 vaultik [--config <path>] prune [--force] [--json]
 vaultik [--config <path>] info
 vaultik [--config <path>] remote info [--json]
+vaultik [--config <path>] remote nuke --force
 vaultik [--config <path>] store info
 vaultik [--config <path>] database purge [--force]
 vaultik completion <bash|zsh|fish|powershell>
@@ -117,11 +117,12 @@ vaultik version
 * `--config <path>`: Path to config file (default: `$VAULTIK_CONFIG`, then platform config dir, then `/etc/vaultik/config.yml`)
 * `--verbose`, `-v`: Enable verbose output
 * `--debug`: Enable debug output
-* `--quiet`, `-q`: Suppress non-error output
+* `--quiet`, `-q`: Suppress non-error output (also suppresses startup banner)
+* `--skip-errors`: Continue past per-file errors instead of aborting (applies to `snapshot create` and `restore`)
 
 ### environment variables
 
-* `VAULTIK_AGE_SECRET_KEY`: Age private key for decryption (required for `restore` and `verify --deep`)
+* `VAULTIK_AGE_SECRET_KEY`: Age private key for decryption (required for `snapshot restore` and `snapshot verify --deep`)
 * `VAULTIK_CONFIG`: Path to config file (overridden by `--config`)
 * `VAULTIK_INDEX_PATH`: Override local SQLite index path
 
@@ -155,11 +156,13 @@ existing file. Created with mode `0600` since it will contain credentials.
 **`config edit`**: Open the config file in `$EDITOR` (falls back to `vi`).
 
 **`config get`**: Print a config value addressed by dotted YAML path
-(e.g. `vaultik config get s3.bucket`). Non-scalar values print as YAML.
+(e.g. `vaultik config get storage_url`). Non-scalar values print as YAML.
 
 **`config set`**: Set a scalar config value by dotted YAML path
-(e.g. `vaultik config set compression_level 9`). Comments and formatting
-in the file are preserved; intermediate maps are created as needed.
+(e.g. `vaultik config set compression_level 9`,
+`vaultik config set storage_url "file:///mnt/backups"`). Comments and
+formatting in the file are preserved; intermediate maps are created as
+needed.
 
 **`snapshot create`**: Perform incremental backup of configured snapshots.
 * Optional snapshot names argument to create specific snapshots (default: all)
@@ -173,9 +176,12 @@ in the file are preserved; intermediate maps are created as needed.
   snapshot per name; use `--keep-newer-than` for a rolling window.
 * `--keep-newer-than <duration>`: With `--prune`, keep snapshots newer than
   this duration instead of only the latest (e.g. `4w`, `30d`, `6mo`, `1y`)
-* `--skip-errors`: Skip file read errors (log them loudly but continue)
 
-**`snapshot list`**: List all snapshots with their timestamps and sizes.
+**`snapshot list`**: Show every snapshot known to the destination
+store with timestamps and three sizes per snapshot (compressed
+remote size; total uncompressed chunk size; size of chunks newly
+referenced by that snapshot). The uncompressed and "new chunk"
+columns show `<remote only>` for snapshots not in the local index.
 * `--json`: Output in JSON format
 
 **`snapshot verify`**: Verify snapshot integrity.
@@ -193,28 +199,31 @@ latest globally).
 * `--force`: Skip confirmation prompt
 
 **`snapshot remove`**: Remove a specific snapshot from the local database.
+Automatically cleans up local rows (files, chunks, blobs) that the removed
+snapshot was the last referrer for — you don't need a separate prune step
+after removal.
 * `--remote`: Also remove snapshot metadata from remote storage
 * `--all`: Remove all snapshots (requires `--force`)
 * `--dry-run`: Show what would be deleted without deleting
 * `--force`: Skip confirmation prompt
 * `--json`: Output result as JSON
 
-**`snapshot prune`**: Clean orphaned data from the local database (files,
-chunks, blobs not referenced by any snapshot).
-
 **`snapshot cleanup`**: Remove stale local snapshot records that have no
 corresponding metadata in remote storage. These are typically left behind
 by incomplete or interrupted backups. Does not touch remote storage.
 
-**`restore`**: Restore files from a backup snapshot.
+**`snapshot restore`**: Restore files from a backup snapshot.
 * Requires `VAULTIK_AGE_SECRET_KEY` environment variable
 * Optional path arguments to restore specific files/directories (default: all)
 * Preserves file permissions, timestamps, ownership (ownership requires root),
   symlinks, and empty directories
 * `--verify`: After restoring, verify every file's chunk hashes match
 
-**`prune`**: Remove unreferenced blobs from remote storage.
-* Scans all snapshot manifests for referenced blobs, deletes any blob not referenced
+**`prune`**: Tidy up everything that isn't needed. Removes orphaned local
+database rows (files, chunks, blobs no longer referenced by any completed
+snapshot) AND deletes unreferenced blobs from remote storage. `snapshot
+create --prune`, `snapshot remove`, and `snapshot purge` run the same
+cleanup automatically; this is the manual entry point for the same work.
 * `--force`: Skip confirmation prompt
 * `--json`: Output stats as JSON
 
@@ -225,6 +234,11 @@ recipients, and local database statistics.
 metadata sizes, blob counts, and orphaned blob detection.
 * `--json`: Output as JSON
 
+**`remote nuke`**: Delete every snapshot's metadata and every blob from the
+backup destination store, leaving the bucket prefix empty. Destructive and
+irreversible.
+* `--force`: Required to confirm destruction.
+
 **`store info`**: Display storage backend type and statistics.
 
 **`database purge`**: Delete the local SQLite state database entirely. Remote
@@ -379,13 +393,71 @@ Key fields:
 
 ## roadmap
 
-Items for future releases:
+Items still to do before / shortly after 1.0. Loosely ordered by
+priority.
 
-* Error-condition tests (network failures, disk full, corrupted/missing blobs)
-* Parallel blob downloads during restore
-* Bandwidth limiting (`--bwlimit`)
-* Security audit of encryption implementation
-* Man pages and richer `--help` examples
+### correctness and operability
+
+* **Security audit of the encryption implementation.** Pre-1.0
+  blocker if we're advertising "secure" at the top of this README.
+  age + zstd + content-defined chunking is mostly off-the-shelf
+  pieces, but the seams (key handling, recipient parsing, manifest
+  trust boundary, restore-time identity validation) need an outside
+  read.
+* **Error-condition tests.** Today's coverage is the happy path
+  plus a few specific regressions. Need fault-injection coverage:
+  network failures mid-blob, disk-full during restore, corrupted /
+  truncated / missing blobs, partial uploads, kill -9 between
+  manifest and db.zst.age writes.
+* **Verify restored content end-to-end in CI.** The current
+  integration test does this for a small synthetic snapshot but
+  not at scale. A nightly job against a multi-GB representative
+  snapshot would catch silent regressions in the chunker, packer,
+  or restore planner.
+
+### performance
+
+* **Parallel blob downloads during restore.** Single-stream right
+  now. With a fast S3 endpoint and a multi-core machine restore is
+  bound by per-blob fetch + decrypt + decompress; running N of
+  those in parallel against the disk cache would close most of the
+  remaining gap. Needs to interact correctly with the locality
+  planner and sweeper.
+* **Bandwidth limiting (`--bwlimit`).** Both upload and download.
+  Useful for backing up over a shared link. Tricky to make work
+  correctly with the parallel-download story.
+* **Restart of interrupted restore.** Today restore is restartable
+  in the sense that re-running it overwrites partial output; it
+  doesn't resume from where it stopped or skip already-present
+  files. A `--resume` mode that checks targets before fetching
+  blobs would matter for very large restores.
+
+### usability
+
+* **Man pages and richer `--help` examples.** Cobra generates
+  basic help; man pages would be a separate target.
+* **`--bwlimit` style human-readable size flags** across the
+  command surface where they're currently raw integers.
+* **`vaultik snapshot diff <a> <b>`** — show which files changed
+  between two snapshots without restoring either.
+* **Status reporting hook for `--cron`.** When a backup fails
+  silently in cron, the user has no idea. A configurable
+  webhook / email / `notify-send` hook on completion (success and
+  failure) would close the loop.
+
+### infrastructure
+
+* **Cross-machine restore documentation.** The "restore from
+  another host" workflow works but isn't documented as a
+  first-class operation in this README. Worth a dedicated section
+  once it's settled.
+* **Schema migrations.** Currently nonexistent — pre-1.0 schema
+  changes are handled by `vaultik database purge` plus a full
+  re-scan. Post-1.0 we'll need a migration story to keep existing
+  index databases usable across upgrades.
+* **Storage backend coverage tests.** S3, file://, and rclone://
+  all share the Storer interface but the rclone path is the least
+  exercised in CI.
 
 ---
 

go.mod

@@ -18,7 +18,6 @@ require (
 	github.com/johannesboyne/gofakes3 v0.0.0-20250603205740-ed9094be7668
 	github.com/klauspost/compress v1.18.1
 	github.com/rclone/rclone v1.72.1
-	github.com/schollz/progressbar/v3 v3.19.0
 	github.com/spf13/afero v1.15.0
 	github.com/spf13/cobra v1.10.1
 	github.com/stretchr/testify v1.11.1
@@ -186,7 +185,6 @@ require (
 	github.com/mattn/go-colorable v0.1.14 // indirect
 	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/mattn/go-runewidth v0.0.19 // indirect
-	github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db // indirect
 	github.com/mitchellh/go-homedir v1.1.0 // indirect
 	github.com/mitchellh/mapstructure v1.5.0 // indirect
 	github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect
@@ -217,7 +215,6 @@ require (
 	github.com/relvacode/iso8601 v1.7.0 // indirect
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
 	github.com/rfjakob/eme v1.1.2 // indirect
-	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/ryanuber/go-glob v1.0.0 // indirect
 	github.com/ryszard/goskiplist v0.0.0-20150312221310-2dfbae5fcf46 // indirect
 	github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06 // indirect

go.sum

@@ -202,8 +202,6 @@ github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UF
 github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
 github.com/cevatbarisyilmaz/ara v0.0.4 h1:SGH10hXpBJhhTlObuZzTuFn1rrdmjQImITXnZVPSodc=
 github.com/cevatbarisyilmaz/ara v0.0.4/go.mod h1:BfFOxnUd6Mj6xmcvRxHN3Sr21Z1T3U2MYkYOmoQe4Ts=
-github.com/chengxilo/virtualterm v1.0.4 h1:Z6IpERbRVlfB8WkOmtbHiDbBANU7cimRIof7mk9/PwM=
-github.com/chengxilo/virtualterm v1.0.4/go.mod h1:DyxxBZz/x1iqJjFxTFcr6/x+jSpqN0iwWCOK1q10rlY=
 github.com/chilts/sid v0.0.0-20190607042430-660e94789ec9 h1:z0uK8UQqjMVYzvk4tiiu3obv2B44+XBsvgEJREQfnO8=
 github.com/chilts/sid v0.0.0-20190607042430-660e94789ec9/go.mod h1:Jl2neWsQaDanWORdqZ4emBl50J4/aRBBS4FyyG9/PFo=
 github.com/chzyer/logex v1.1.10/go.mod h1:+Ywpsq7O8HXn0nuIou7OrIPyXbp3wmkHB+jjWRnGsAI=
@@ -599,8 +597,6 @@ github.com/miekg/dns v1.1.41 h1:WMszZWJG0XmzbK9FEmzH2TVcqYzFesusSIB41b8KHxY=
 github.com/miekg/dns v1.1.41/go.mod h1:p6aan82bvRIyn+zDIv9xYNUpwa73JcSh9BKwknJysuI=
 github.com/mitchellh/cli v1.0.0/go.mod h1:hNIlj7HEI86fIcpObd7a0FcrxTWetlwJDGcceTlRvqc=
 github.com/mitchellh/cli v1.1.0/go.mod h1:xcISNoH86gajksDmfB23e/pu+B+GeFRMYmoHXxx3xhI=
-github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db h1:62I3jR2EmQ4l5rM/4FEfDWcRD+abF5XlKShorW5LRoQ=
-github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db/go.mod h1:l0dey0ia/Uv7NcFFVbCLtqEBQbrT4OCwCSKTEv6enCw=
 github.com/mitchellh/go-homedir v1.1.0 h1:lukF9ziXFxDFPkA1vsr5zpc1XuPDn/wFntq5mG+4E0Y=
 github.com/mitchellh/go-homedir v1.1.0/go.mod h1:SfyaCUpYCn1Vlf4IUYiD9fPX4A5wJrkLzIz1N1q0pr0=
 github.com/mitchellh/go-wordwrap v1.0.0/go.mod h1:ZXFpozHsX6DPmq2I0TCekCxypsnAUbP2oI0UX1GXzOo=
@@ -705,8 +701,6 @@ github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
 github.com/rfjakob/eme v1.1.2 h1:SxziR8msSOElPayZNFfQw4Tjx/Sbaeeh3eRvrHVMUs4=
 github.com/rfjakob/eme v1.1.2/go.mod h1:cVvpasglm/G3ngEfcfT/Wt0GwhkuO32pf/poW6Nyk1k=
-github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
-github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
 github.com/rogpeppe/go-internal v1.3.0/go.mod h1:M8bDsm7K2OlrFYOpmOWEs/qY81heoFRclV5y23lUDJ4=
 github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
 github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
@@ -721,8 +715,6 @@ github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06 h1:OkMGxebDj
 github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06/go.mod h1:+ePHsJ1keEjQtpvf9HHw0f4ZeJ0TLRsxhunSI2hYJSs=
 github.com/samber/lo v1.52.0 h1:Rvi+3BFHES3A8meP33VPAxiBZX/Aws5RxrschYGjomw=
 github.com/samber/lo v1.52.0/go.mod h1:4+MXEGsJzbKGaUEQFKBq2xtfuznW9oz/WrgyzMzRoM0=
-github.com/schollz/progressbar/v3 v3.19.0 h1:Ea18xuIRQXLAUidVDox3AbwfUhD0/1IvohyTutOIFoc=
-github.com/schollz/progressbar/v3 v3.19.0/go.mod h1:IsO3lpbaGuzh8zIMzgY3+J8l4C8GjO0Y9S69eFvNsec=
 github.com/sean-/seed v0.0.0-20170313163322-e2103e2c3529 h1:nn5Wsu0esKSJiIVhscUtVbo7ada43DJhG55ua/hjS5I=
 github.com/sean-/seed v0.0.0-20170313163322-e2103e2c3529/go.mod h1:DxrIzT+xaE7yg65j358z/aeFdxmN0P9QXhEzd20vsDc=
 github.com/sergi/go-diff v1.0.0/go.mod h1:0CfEIISq7TuYL3j771MWULgwwjU+GofnZX9QAmXWZgo=

internal/cli/app.go

@@ -4,7 +4,6 @@ import (
 	"context"
 	"errors"
 	"fmt"
-	"io"
 	"os"
 	"os/signal"
 	"path/filepath"
@@ -35,29 +34,36 @@ type AppOptions struct {
 	Invokes    []fx.Option
 }
 
-// setupGlobals records the startup time and prints the startup banner.
-// In --cron mode the banner is suppressed (LogOptions.Cron == true).
+// setupGlobals records the startup time and, when an output-suppression
+// flag is active, marks the UI writer quiet so that Begin/Complete/
+// Info/Notice/Detail/Progress are silenced. Warning and Error are NOT
+// silenced — per the documented convention that --quiet suppresses
+// non-error output only. The startup banner is printed by CLIEntry
+// before cobra parses arguments, gated by the same arg-level check.
 func setupGlobals(lc fx.Lifecycle, g *globals.Globals, v *vaultik.Vaultik, opts log.LogOptions) {
 	lc.Append(fx.Hook{
 		OnStart: func(ctx context.Context) error {
 			g.StartTime = time.Now().UTC()
 			if opts.Cron || opts.Quiet {
-				// Replace UI writer with a discarding one so all
-				// user-facing output is suppressed.
-				v.UI = ui.NewWithColor(io.Discard, false)
-			} else {
-				v.UI.Banner("%s %s by %s (commit %s, built on %s) starting up at %s.",
-					g.Appname, g.Version, globals.Author,
-					g.ShortCommit(), g.CommitDate,
-					g.StartTime.Format(time.RFC3339))
-				v.UI.Banner("%s", globals.Homepage)
-				v.UI.Banner("")
+				v.UI.SetQuiet(true)
 			}
 			return nil
 		},
 	})
 }
 
+// writeStartupBanner prints the two-line application banner followed by a
+// blank line. Used both from the fx hook (for subcommand invocations) and
+// from the root cobra Run handler (for `vaultik` with no subcommand).
+func writeStartupBanner(w *ui.Writer, startTime time.Time, shortCommit string) {
+	w.Banner("%s %s by %s (commit %s, built on %s) starting up at %s.",
+		globals.Appname, globals.Version, globals.Author,
+		shortCommit, globals.CommitDate,
+		startTime.Format(time.RFC3339))
+	w.Banner("%s", globals.Homepage)
+	w.Banner("")
+}
+
 // NewApp creates a new fx application with common modules.
 // It sets up the base modules (config, database, logging, globals) and
 // combines them with any additional modules specified in the options.

internal/cli/config.go

@@ -285,7 +285,7 @@ func newConfigEditCommand() *cobra.Command {
 func newConfigGetCommand() *cobra.Command {
 	return &cobra.Command{
 		Use:   "get <key>",
-		Short: "Print a config value by dotted path (e.g. s3.bucket)",
+		Short: "Print a config value by dotted path (e.g. storage_url, compression_level)",
 		Args:  cobra.ExactArgs(1),
 		RunE: func(cmd *cobra.Command, args []string) error {
 			path, err := ResolveConfigPath()
@@ -328,9 +328,10 @@ the file back, preserving comments and formatting. Intermediate maps
 are created as needed.
 
 Examples:
+  vaultik config set storage_url "file:///mnt/backups"
+  vaultik config set storage_url "s3://bucket/prefix?endpoint=host&region=us-east-1"
   vaultik config set compression_level 9
-  vaultik config set s3.bucket mybucket
-  vaultik config set storage_url "file:///mnt/backups"`,
+  vaultik config set s3.bucket mybucket  # legacy S3 fields still supported`,
 		Args: cobra.ExactArgs(2),
 		RunE: func(cmd *cobra.Command, args []string) error {
 			path, err := ResolveConfigPath()

internal/cli/entry.go

@@ -2,14 +2,67 @@ package cli
 
 import (
 	"os"
+	"strings"
+	"time"
+
+	"sneak.berlin/go/vaultik/internal/globals"
+	"sneak.berlin/go/vaultik/internal/ui"
 )
 
 // CLIEntry is the main entry point for the CLI application.
-// It creates the root command, executes it, and exits with status 1
-// if an error occurs. This function should be called from main().
+// It prints the startup banner (unless a quiet flag is present in os.Args),
+// executes the root cobra command, and routes any returned error through
+// the ui.Writer so the user sees a properly formatted "🛑 ERROR:" line.
 func CLIEntry() {
+	if !bannerSuppressedInArgs(os.Args[1:]) {
+		short := globals.Commit
+		if len(short) > 12 {
+			short = short[:12]
+		}
+		writeStartupBanner(ui.New(os.Stdout), time.Now().UTC(), short)
+	}
+
 	rootCmd := NewRootCommand()
+	rootCmd.SilenceErrors = true
+
 	if err := rootCmd.Execute(); err != nil {
+		ReportError("%s", err.Error())
 		os.Exit(1)
 	}
 }
+
+// ReportError emits a user-facing error to stderr in the standard
+// 🛑 ERROR: format. Use it from goroutine error paths (where returning
+// an error to cobra isn't an option) and anywhere else a CLI command
+// must surface a failure outside the normal RunE return path.
+func ReportError(format string, args ...any) {
+	ui.New(os.Stderr).Error(format, args...)
+}
+
+// bannerSuppressedInArgs reports whether any of args is a flag that
+// should suppress the startup banner (--quiet/-q/--cron). Stops at the
+// "--" argument terminator. Recognizes both long forms and short -q,
+// including combined short flags like "-qv".
+func bannerSuppressedInArgs(args []string) bool {
+	for _, a := range args {
+		if a == "--" {
+			return false
+		}
+		switch a {
+		case "--quiet", "-q", "--cron":
+			return true
+		}
+		if strings.HasPrefix(a, "--quiet=") || strings.HasPrefix(a, "--cron=") {
+			return true
+		}
+		// Combined short flags like -qv or -vq.
+		if len(a) > 1 && a[0] == '-' && a[1] != '-' {
+			for _, c := range a[1:] {
+				if c == 'q' {
+					return true
+				}
+			}
+		}
+	}
+	return false
+}

internal/cli/entry_test.go

@@ -18,7 +18,7 @@ func TestCLIEntry(t *testing.T) {
 	}
 
 	// Verify all subcommands are registered
-	expectedCommands := []string{"config", "snapshot", "store", "restore", "prune", "info", "version", "remote", "database"}
+	expectedCommands := []string{"config", "snapshot", "store", "prune", "info", "version", "remote", "database"}
 	for _, expected := range expectedCommands {
 		found := false
 		for _, cmd := range cmd.Commands() {
@@ -38,7 +38,7 @@ func TestCLIEntry(t *testing.T) {
 		t.Errorf("Failed to find snapshot command: %v", err)
 	} else {
 		// Check snapshot subcommands
-		expectedSubCommands := []string{"create", "list", "purge", "verify", "cleanup"}
+		expectedSubCommands := []string{"create", "list", "purge", "verify", "cleanup", "restore"}
 		for _, expected := range expectedSubCommands {
 			found := false
 			for _, subcmd := range snapshotCmd.Commands() {

internal/cli/info.go

@@ -47,6 +47,7 @@ func NewInfoCommand() *cobra.Command {
 									if err := v.ShowInfo(); err != nil {
 										if err != context.Canceled {
 											log.Error("Failed to show info", "error", err)
+											ReportError("Failed to show info: %v", err)
 											os.Exit(1)
 										}
 									}

internal/cli/prune.go

@@ -16,14 +16,19 @@ func NewPruneCommand() *cobra.Command {
 
 	cmd := &cobra.Command{
 		Use:   "prune",
-		Short: "Remove unreferenced blobs",
-		Long: `Removes blobs that are not referenced by any snapshot.
+		Short: "Tidy local database and remote storage",
+		Long: `Removes orphaned data from both the local index database and
+unreferenced blobs from the backup destination store.
 
-This command scans all snapshots and their manifests to build a list of
-referenced blobs, then removes any blobs in storage that are not in this list.
+Local cleanup drops incomplete snapshots and any files, chunks, or
+blobs no longer referenced by a completed snapshot. Remote cleanup
+scans every snapshot manifest in the destination store, builds the
+set of still-referenced blob hashes, and deletes any blob not in that
+set.
 
-Use this command after deleting snapshots with 'vaultik purge' to reclaim
-storage space.`,
+Snapshot create --prune and snapshot remove run the same cleanup
+automatically; this command is the manual entry point for the same
+work (e.g. after a crashed backup or to reclaim storage).`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, args []string) error {
 			// Use unified config resolution
@@ -49,10 +54,11 @@ storage space.`,
 								// Start the prune operation in a goroutine
 								go func() {
 									// Run the prune operation
-									if err := v.PruneBlobs(opts); err != nil {
+									if err := v.Prune(opts); err != nil {
 										if err != context.Canceled {
 											if !opts.JSON {
 												log.Error("Prune operation failed", "error", err)
+												ReportError("Prune failed: %v", err)
 											}
 											os.Exit(1)
 										}

internal/cli/remote.go

@@ -2,6 +2,7 @@ package cli
 
 import (
 	"context"
+	"fmt"
 	"os"
 
 	"github.com/spf13/cobra"
@@ -20,6 +21,73 @@ func NewRemoteCommand() *cobra.Command {
 
 	// Add subcommands
 	cmd.AddCommand(newRemoteInfoCommand())
+	cmd.AddCommand(newRemoteNukeCommand())
+
+	return cmd
+}
+
+// newRemoteNukeCommand creates the 'remote nuke' subcommand.
+func newRemoteNukeCommand() *cobra.Command {
+	var force bool
+
+	cmd := &cobra.Command{
+		Use:   "nuke",
+		Short: "Delete ALL snapshot metadata and blobs from the backup destination store",
+		Long: `Removes every snapshot's metadata and every blob from remote
+storage. After this command completes successfully the bucket prefix is
+empty and the next backup starts from scratch.
+
+This is destructive and irreversible. Requires --force.`,
+		Args: cobra.NoArgs,
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if !force {
+				return fmt.Errorf("remote nuke requires --force (this deletes ALL remote snapshots and blobs)")
+			}
+
+			configPath, err := ResolveConfigPath()
+			if err != nil {
+				return err
+			}
+
+			rootFlags := GetRootFlags()
+			return RunWithApp(cmd.Context(), AppOptions{
+				ConfigPath: configPath,
+				LogOptions: log.LogOptions{
+					Verbose: rootFlags.Verbose,
+					Debug:   rootFlags.Debug,
+					Quiet:   rootFlags.Quiet,
+				},
+				Modules: []fx.Option{},
+				Invokes: []fx.Option{
+					fx.Invoke(func(v *vaultik.Vaultik, lc fx.Lifecycle) {
+						lc.Append(fx.Hook{
+							OnStart: func(ctx context.Context) error {
+								go func() {
+									if err := v.NukeRemote(true); err != nil {
+										if err != context.Canceled {
+											log.Error("Remote nuke failed", "error", err)
+											ReportError("Remote nuke failed: %v", err)
+											os.Exit(1)
+										}
+									}
+									if err := v.Shutdowner.Shutdown(); err != nil {
+										log.Error("Failed to shutdown", "error", err)
+									}
+								}()
+								return nil
+							},
+							OnStop: func(ctx context.Context) error {
+								v.Cancel()
+								return nil
+							},
+						})
+					}),
+				},
+			})
+		},
+	}
+
+	cmd.Flags().BoolVar(&force, "force", false, "Required: confirm destruction of ALL remote data")
 
 	return cmd
 }
@@ -62,6 +130,7 @@ func newRemoteInfoCommand() *cobra.Command {
 										if err != context.Canceled {
 											if !jsonOutput {
 												log.Error("Failed to get remote info", "error", err)
+												ReportError("Failed to get remote info: %v", err)
 											}
 											os.Exit(1)
 										}

internal/cli/root.go

@@ -17,6 +17,7 @@ type RootFlags struct {
 	Verbose    bool
 	Debug      bool
 	Quiet      bool
+	SkipErrors bool
 }
 
 var rootFlags RootFlags
@@ -32,6 +33,13 @@ func NewRootCommand() *cobra.Command {
 public keys and uploads to S3-compatible storage. No private keys are needed
 on the source system.`,
 		SilenceUsage: true,
+		// Bare 'vaultik' (no subcommand): print help. The banner is
+		// printed once at process startup by CLIEntry, before cobra
+		// parses arguments, so it appears even when cobra rejects
+		// args (e.g. "requires at least 2 arg(s)") and on --help.
+		Run: func(cmd *cobra.Command, args []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add global flags
@@ -39,11 +47,11 @@ on the source system.`,
 	cmd.PersistentFlags().BoolVarP(&rootFlags.Verbose, "verbose", "v", false, "Enable verbose output")
 	cmd.PersistentFlags().BoolVar(&rootFlags.Debug, "debug", false, "Enable debug output")
 	cmd.PersistentFlags().BoolVarP(&rootFlags.Quiet, "quiet", "q", false, "Suppress non-error output")
+	cmd.PersistentFlags().BoolVar(&rootFlags.SkipErrors, "skip-errors", false, "Continue past per-file errors instead of aborting (applies to snapshot create and restore)")
 
 	// Add subcommands
 	cmd.AddCommand(
 		NewConfigCommand(),
-		NewRestoreCommand(),
 		NewPruneCommand(),
 		NewStoreCommand(),
 		NewSnapshotCommand(),

internal/cli/snapshot.go

@@ -25,8 +25,8 @@ func NewSnapshotCommand() *cobra.Command {
 	cmd.AddCommand(newSnapshotPurgeCommand())
 	cmd.AddCommand(newSnapshotVerifyCommand())
 	cmd.AddCommand(newSnapshotRemoveCommand())
-	cmd.AddCommand(newSnapshotPruneCommand())
 	cmd.AddCommand(newSnapshotCleanupCommand())
+	cmd.AddCommand(newSnapshotRestoreCommand())
 
 	return cmd
 }
@@ -49,6 +49,8 @@ specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
 		RunE: func(cmd *cobra.Command, args []string) error {
 			// Pass snapshot names from args
 			opts.Snapshots = args
+			// --skip-errors is a global flag on the root command.
+			opts.SkipErrors = rootFlags.SkipErrors
 			// Use unified config resolution
 			configPath, err := ResolveConfigPath()
 			if err != nil {
@@ -76,6 +78,7 @@ specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
 									if err := v.CreateSnapshot(opts); err != nil {
 										if err != context.Canceled {
 											log.Error("Snapshot creation failed", "error", err)
+											ReportError("Snapshot creation failed: %v", err)
 											os.Exit(1)
 										}
 									}
@@ -103,7 +106,6 @@ specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
 	cmd.Flags().BoolVar(&opts.Cron, "cron", false, "Run in cron mode (silent unless error)")
 	cmd.Flags().BoolVar(&opts.Prune, "prune", false, "After backup, drop older snapshots of the same name and remove orphaned blobs")
 	cmd.Flags().StringVar(&opts.KeepNewerThan, "keep-newer-than", "", "With --prune: keep snapshots newer than this duration (e.g. 4w, 30d, 6mo) instead of only the latest")
-	cmd.Flags().BoolVar(&opts.SkipErrors, "skip-errors", false, "Skip file read errors (log them loudly but continue)")
 
 	return cmd
 }
@@ -142,6 +144,7 @@ func newSnapshotListCommand() *cobra.Command {
 									if err := v.ListSnapshots(jsonOutput); err != nil {
 										if err != context.Canceled {
 											log.Error("Failed to list snapshots", "error", err)
+											ReportError("Failed to list snapshots: %v", err)
 											os.Exit(1)
 										}
 									}
@@ -212,6 +215,7 @@ restrict the operation to specific snapshot names.`,
 									if err := v.PurgeSnapshotsWithOptions(opts); err != nil {
 										if err != context.Canceled {
 											log.Error("Failed to purge snapshots", "error", err)
+											ReportError("Failed to purge snapshots: %v", err)
 											os.Exit(1)
 										}
 									}
@@ -285,6 +289,7 @@ func newSnapshotVerifyCommand() *cobra.Command {
 										if err != context.Canceled {
 											if !opts.JSON {
 												log.Error("Verification failed", "error", err)
+												ReportError("Verification failed: %v", err)
 											}
 											os.Exit(1)
 										}
@@ -378,6 +383,7 @@ Use --all --force to remove all snapshots.`,
 										if err != context.Canceled {
 											if !opts.JSON {
 												log.Error("Failed to remove snapshot", "error", err)
+												ReportError("Failed to remove snapshot: %v", err)
 											}
 											os.Exit(1)
 										}
@@ -408,63 +414,6 @@ Use --all --force to remove all snapshots.`,
 	return cmd
 }
 
-// newSnapshotPruneCommand creates the 'snapshot prune' subcommand
-func newSnapshotPruneCommand() *cobra.Command {
-	cmd := &cobra.Command{
-		Use:   "prune",
-		Short: "Remove orphaned data from local database",
-		Long: `Removes orphaned files, chunks, and blobs from the local database.
-
-This cleans up data that is no longer referenced by any snapshot, which can
-accumulate from incomplete backups or deleted snapshots.`,
-		Args: cobra.NoArgs,
-		RunE: func(cmd *cobra.Command, args []string) error {
-			// Use unified config resolution
-			configPath, err := ResolveConfigPath()
-			if err != nil {
-				return err
-			}
-
-			rootFlags := GetRootFlags()
-			return RunWithApp(cmd.Context(), AppOptions{
-				ConfigPath: configPath,
-				LogOptions: log.LogOptions{
-					Verbose: rootFlags.Verbose,
-					Debug:   rootFlags.Debug,
-					Quiet:   rootFlags.Quiet,
-				},
-				Modules: []fx.Option{},
-				Invokes: []fx.Option{
-					fx.Invoke(func(v *vaultik.Vaultik, lc fx.Lifecycle) {
-						lc.Append(fx.Hook{
-							OnStart: func(ctx context.Context) error {
-								go func() {
-									if _, err := v.PruneDatabase(); err != nil {
-										if err != context.Canceled {
-											log.Error("Failed to prune database", "error", err)
-											os.Exit(1)
-										}
-									}
-									if err := v.Shutdowner.Shutdown(); err != nil {
-										log.Error("Failed to shutdown", "error", err)
-									}
-								}()
-								return nil
-							},
-							OnStop: func(ctx context.Context) error {
-								v.Cancel()
-								return nil
-							},
-						})
-					}),
-				},
-			})
-		},
-	}
-
-	return cmd
-}
-
 // newSnapshotCleanupCommand creates the 'snapshot cleanup' subcommand
 func newSnapshotCleanupCommand() *cobra.Command {
 	cmd := &cobra.Command{
@@ -499,6 +448,7 @@ This command does not delete anything from remote storage.`,
 									if err := v.CleanupLocalSnapshots(); err != nil {
 										if err != context.Canceled {
 											log.Error("Cleanup failed", "error", err)
+											ReportError("Cleanup failed: %v", err)
 											os.Exit(1)
 										}
 									}

internal/cli/snapshot_restore.go

@@ -29,13 +29,13 @@ type RestoreApp struct {
 	Shutdowner fx.Shutdowner
 }
 
-// NewRestoreCommand creates the restore command
-func NewRestoreCommand() *cobra.Command {
+// newSnapshotRestoreCommand creates the 'snapshot restore' subcommand
+func newSnapshotRestoreCommand() *cobra.Command {
 	opts := &RestoreOptions{}
 
 	cmd := &cobra.Command{
 		Use:   "restore <snapshot-id> <target-dir> [paths...]",
-		Short: "Restore files from backup",
+		Short: "Restore files from a snapshot",
 		Long: `Download and decrypt files from a backup snapshot.
 
 This command will restore files from the specified snapshot to the target directory.
@@ -46,16 +46,16 @@ Requires the VAULTIK_AGE_SECRET_KEY environment variable to be set with the age
 
 Examples:
   # Restore entire snapshot
-  vaultik restore myhost_docs_2025-01-01T12:00:00Z /restore
+  vaultik snapshot restore myhost_docs_2025-01-01T12:00:00Z /restore
 
   # Restore specific file
-  vaultik restore myhost_docs_2025-01-01T12:00:00Z /restore /home/user/important.txt
+  vaultik snapshot restore myhost_docs_2025-01-01T12:00:00Z /restore /home/user/important.txt
 
   # Restore specific directory
-  vaultik restore myhost_docs_2025-01-01T12:00:00Z /restore /home/user/documents/
+  vaultik snapshot restore myhost_docs_2025-01-01T12:00:00Z /restore /home/user/documents/
 
   # Restore and verify all files
-  vaultik restore --verify myhost_docs_2025-01-01T12:00:00Z /restore`,
+  vaultik snapshot restore --verify myhost_docs_2025-01-01T12:00:00Z /restore`,
 		Args: cobra.MinimumNArgs(2),
 		RunE: func(cmd *cobra.Command, args []string) error {
 			return runRestore(cmd, args, opts)
@@ -127,10 +127,12 @@ func buildRestoreInvokes(snapshotID string, opts *RestoreOptions) []fx.Option {
 							TargetDir:  opts.TargetDir,
 							Paths:      opts.Paths,
 							Verify:     opts.Verify,
+							SkipErrors: GetRootFlags().SkipErrors,
 						}
 						if err := app.Vaultik.Restore(restoreOpts); err != nil {
 							if err != context.Canceled {
 								log.Error("Restore operation failed", "error", err)
+								ReportError("Restore failed: %v", err)
 								os.Exit(1)
 							}
 						}

internal/database/blobs.go

@@ -130,6 +130,51 @@ func (r *BlobRepository) GetByID(ctx context.Context, id string) (*Blob, error)
 	return &blob, nil
 }
 
+// GetAll returns every blob row keyed by blob ID. Useful at restore
+// start to translate the per-chunk blob_id references in chunkToBlobMap
+// into blob hashes without doing one GetByID query per chunk.
+func (r *BlobRepository) GetAll(ctx context.Context) (map[string]*Blob, error) {
+	query := `
+		SELECT id, blob_hash, created_ts, finished_ts, uncompressed_size, compressed_size, uploaded_ts
+		FROM blobs
+	`
+
+	rows, err := r.db.conn.QueryContext(ctx, query)
+	if err != nil {
+		return nil, fmt.Errorf("querying blobs: %w", err)
+	}
+	defer CloseRows(rows)
+
+	out := make(map[string]*Blob)
+	for rows.Next() {
+		var blob Blob
+		var createdTSUnix int64
+		var finishedTSUnix, uploadedTSUnix sql.NullInt64
+		if err := rows.Scan(
+			&blob.ID,
+			&blob.Hash,
+			&createdTSUnix,
+			&finishedTSUnix,
+			&blob.UncompressedSize,
+			&blob.CompressedSize,
+			&uploadedTSUnix,
+		); err != nil {
+			return nil, fmt.Errorf("scanning blob: %w", err)
+		}
+		blob.CreatedTS = time.Unix(createdTSUnix, 0).UTC()
+		if finishedTSUnix.Valid {
+			ts := time.Unix(finishedTSUnix.Int64, 0).UTC()
+			blob.FinishedTS = &ts
+		}
+		if uploadedTSUnix.Valid {
+			ts := time.Unix(uploadedTSUnix.Int64, 0).UTC()
+			blob.UploadedTS = &ts
+		}
+		out[blob.ID.String()] = &blob
+	}
+	return out, rows.Err()
+}
+
 // UpdateFinished updates a blob when it's finalized
 func (r *BlobRepository) UpdateFinished(ctx context.Context, tx *sql.Tx, id string, hash string, uncompressedSize, compressedSize int64) error {
 	query := `

internal/database/snapshots.go

@@ -331,6 +331,43 @@ func (r *SnapshotRepository) AddFilesByIDBatch(ctx context.Context, tx *sql.Tx,
 	return nil
 }
 
+// PopulateReferencedBlobs ensures snapshot_blobs contains an entry for
+// every blob that holds a chunk referenced by any file in the snapshot.
+// This is necessary because the AddBlob hook only runs when a blob is
+// newly uploaded during a snapshot — fully-deduplicated snapshots (where
+// every chunk already exists in storage from a prior run) would otherwise
+// have an empty snapshot_blobs set and be impossible to restore.
+//
+// Returns the number of rows inserted (i.e. blobs that were previously
+// referenced indirectly via file_chunks but not yet recorded in
+// snapshot_blobs for this snapshot).
+func (r *SnapshotRepository) PopulateReferencedBlobs(ctx context.Context, tx *sql.Tx, snapshotID string) (int64, error) {
+	query := `
+		INSERT OR IGNORE INTO snapshot_blobs (snapshot_id, blob_id, blob_hash)
+		SELECT DISTINCT ?, blobs.id, blobs.blob_hash
+		FROM blobs
+		JOIN blob_chunks ON blob_chunks.blob_id = blobs.id
+		JOIN file_chunks ON file_chunks.chunk_hash = blob_chunks.chunk_hash
+		JOIN snapshot_files ON snapshot_files.file_id = file_chunks.file_id
+		WHERE snapshot_files.snapshot_id = ?
+		  AND blobs.blob_hash IS NOT NULL
+	`
+
+	var result sql.Result
+	var err error
+	if tx != nil {
+		result, err = tx.ExecContext(ctx, query, snapshotID, snapshotID)
+	} else {
+		result, err = r.db.ExecWithLog(ctx, query, snapshotID, snapshotID)
+	}
+	if err != nil {
+		return 0, fmt.Errorf("populating referenced blobs: %w", err)
+	}
+
+	n, _ := result.RowsAffected()
+	return n, nil
+}
+
 // AddBlob adds a blob to a snapshot
 func (r *SnapshotRepository) AddBlob(ctx context.Context, tx *sql.Tx, snapshotID string, blobID types.BlobID, blobHash types.BlobHash) error {
 	query := `
@@ -397,6 +434,65 @@ func (r *SnapshotRepository) GetSnapshotTotalCompressedSize(ctx context.Context,
 	return totalSize, nil
 }
 
+// GetSnapshotUncompressedChunkSize returns the sum of plaintext sizes of all unique
+// chunks referenced by a snapshot (via snapshot_files → file_chunks → chunks).
+func (r *SnapshotRepository) GetSnapshotUncompressedChunkSize(ctx context.Context, snapshotID string) (int64, error) {
+	query := `
+		SELECT COALESCE(SUM(c.size), 0)
+		FROM (
+			SELECT DISTINCT fc.chunk_hash
+			FROM snapshot_files sf
+			JOIN file_chunks fc ON sf.file_id = fc.file_id
+			WHERE sf.snapshot_id = ?
+		) sc
+		JOIN chunks c ON sc.chunk_hash = c.chunk_hash
+	`
+
+	var totalSize int64
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&totalSize)
+	if err != nil {
+		return 0, fmt.Errorf("querying uncompressed chunk size: %w", err)
+	}
+
+	return totalSize, nil
+}
+
+// GetSnapshotNewChunkSize returns the sum of plaintext sizes of chunks that are
+// referenced by this snapshot but not by any earlier completed snapshot known to
+// the local database. The result is the marginal uncompressed data this snapshot
+// added to the dedup pool — i.e., the delta from prior snapshots.
+func (r *SnapshotRepository) GetSnapshotNewChunkSize(ctx context.Context, snapshotID string) (int64, error) {
+	query := `
+		WITH this_snap_chunks AS (
+			SELECT DISTINCT fc.chunk_hash
+			FROM snapshot_files sf
+			JOIN file_chunks fc ON sf.file_id = fc.file_id
+			WHERE sf.snapshot_id = ?
+		),
+		prior_chunks AS (
+			SELECT DISTINCT fc.chunk_hash
+			FROM snapshots s
+			JOIN snapshot_files sf ON sf.snapshot_id = s.id
+			JOIN file_chunks fc ON fc.file_id = sf.file_id
+			WHERE s.completed_at IS NOT NULL
+			  AND s.id != ?
+			  AND s.started_at < (SELECT started_at FROM snapshots WHERE id = ?)
+		)
+		SELECT COALESCE(SUM(c.size), 0)
+		FROM chunks c
+		JOIN this_snap_chunks t ON c.chunk_hash = t.chunk_hash
+		WHERE c.chunk_hash NOT IN (SELECT chunk_hash FROM prior_chunks)
+	`
+
+	var totalSize int64
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID, snapshotID, snapshotID).Scan(&totalSize)
+	if err != nil {
+		return 0, fmt.Errorf("querying new chunk size: %w", err)
+	}
+
+	return totalSize, nil
+}
+
 // GetIncompleteSnapshots returns all snapshots that haven't been completed
 func (r *SnapshotRepository) GetIncompleteSnapshots(ctx context.Context) ([]*Snapshot, error) {
 	query := `

internal/log/log.go

@@ -46,8 +46,12 @@ func Initialize(cfg Config) {
 	var level slog.Level
 
 	if cfg.Cron || cfg.Quiet {
-		// In quiet/cron mode, only show errors
-		level = slog.LevelError
+		// In cron/quiet mode keep warnings and errors visible — the
+		// whole point of --cron is to stay silent only on total
+		// success, so that anything cron emails to root is genuinely
+		// "something went wrong, look at it." A backup with stuck
+		// permission errors or skipped files should NOT be silent.
+		level = slog.LevelWarn
 	} else if cfg.Debug || strings.Contains(os.Getenv("GODEBUG"), "vaultik") {
 		level = slog.LevelDebug
 	} else if cfg.Verbose {

internal/snapshot/remotekey.go

@@ -0,0 +1,40 @@
+package snapshot
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+)
+
+// remoteKeyPrefix is mixed into the snapshot ID hash so the resulting
+// hex digest is domain-separated from any other "double SHA256 of a
+// string" identifier the user might also use. Keeping this stable is a
+// hard compatibility requirement: changing it invalidates every
+// existing snapshot's remote storage path.
+const remoteKeyPrefix = "vaultik|"
+
+// RemoteSnapshotKey returns the storage-side identifier for a snapshot
+// given its human snapshot ID. It is hex(SHA256(SHA256(prefix + id))).
+// The two SHA256 rounds match Bitcoin's "hash256" convention so the
+// output looks like a 64-character hex blob with no exploitable
+// structure visible to a remote observer.
+//
+// We use this in three places:
+//
+//   - the "metadata/<remote-key>/..." subdirectory on the storage
+//     backend so a directory listing of the bucket / file:// dest
+//     doesn't reveal hostnames, configured snapshot names, or backup
+//     timestamps;
+//   - the `snapshot_id` field of the unencrypted manifest.json.zst
+//     for the same reason;
+//   - any code path that needs to translate a known local snapshot ID
+//     into the path it would occupy on remote storage.
+//
+// The human ID stays the user-visible handle everywhere else — local
+// database joins, CLI arguments, summary lines, log fields — because
+// it's never written to the public bytes once this function gates
+// every storage-path construction.
+func RemoteSnapshotKey(snapshotID string) string {
+	first := sha256.Sum256([]byte(remoteKeyPrefix + snapshotID))
+	second := sha256.Sum256(first[:])
+	return hex.EncodeToString(second[:])
+}

internal/snapshot/scanner.go

@@ -224,13 +224,13 @@ func (s *Scanner) Scan(ctx context.Context, path string, snapshotID string) (*Sc
 
 	// Phase 2: Process files and create chunks
 	if len(filesToProcess) > 0 {
-		s.ui.Begin("Processing %s snapshot source files (chunking, compressing, encrypting, uploading).", s.ui.Count(len(filesToProcess)))
+		s.ui.Begin("Backing up %s snapshot source files (chunking, compressing, encrypting, uploading).", s.ui.Count(len(filesToProcess)))
 		log.Info("Phase 2/3: Creating snapshot (chunking, compressing, encrypting, and uploading blobs)")
 		if err := s.processPhase(ctx, filesToProcess, result); err != nil {
 			return nil, fmt.Errorf("process phase failed: %w", err)
 		}
 	} else {
-		s.ui.Info("Snapshot file processing skipped: no changed files (creating metadata-only snapshot).")
+		s.ui.Info("Snapshot file backup skipped: no changed files (creating metadata-only snapshot).")
 		log.Info("Phase 2/3: Skipping (no files need processing, metadata-only snapshot)")
 	}
 
@@ -278,7 +278,7 @@ func (s *Scanner) summarizeScanPhase(result *ScanResult, filesToProcess []*FileT
 		"files_skipped", result.FilesSkipped,
 		"bytes_skipped", humanize.Bytes(uint64(result.BytesSkipped)))
 
-	msg := fmt.Sprintf("Enumerated %s snapshot source files (%s total), %s to process (%s)",
+	msg := fmt.Sprintf("Enumerated %s snapshot source files (%s total), %s to back up (%s)",
 		s.ui.Count(result.FilesScanned),
 		s.ui.Size(totalSizeToProcess+result.BytesSkipped),
 		s.ui.Count(len(filesToProcess)),
@@ -1067,7 +1067,7 @@ func (s *Scanner) printProcessingProgress(filesProcessed, totalFiles int, bytesP
 	}
 
 	if eta > 0 {
-		s.ui.Progress("Snapshot file processing: %s/%s files (%s), %s/%s, %s, %.0f files/sec, processing elapsed: %s, processing ETA: %s (est remain %s).",
+		s.ui.Progress("Snapshot backup: %s/%s files (%s), %s/%s, %s, %.0f files/sec, backup elapsed: %s, backup ETA: %s (est remain %s).",
 			s.ui.Count(filesProcessed),
 			s.ui.Count(totalFiles),
 			s.ui.Percent(pct),
@@ -1079,7 +1079,7 @@ func (s *Scanner) printProcessingProgress(filesProcessed, totalFiles int, bytesP
 			s.ui.Time(time.Now().Add(eta)),
 			s.ui.Duration(eta))
 	} else {
-		s.ui.Progress("Snapshot file processing: %s/%s files (%s), %s/%s, %s, %.0f files/sec, processing elapsed: %s.",
+		s.ui.Progress("Snapshot backup: %s/%s files (%s), %s/%s, %s, %.0f files/sec, backup elapsed: %s.",
 			s.ui.Count(filesProcessed),
 			s.ui.Count(totalFiles),
 			s.ui.Percent(pct),
@@ -1177,16 +1177,17 @@ func (s *Scanner) uploadBlobIfNeeded(ctx context.Context, blobPath string, blobW
 	finishedBlob := blobWithReader.FinishedBlob
 
 	// Check if blob already exists (deduplication after restart)
+	destination := s.storage.Info().Location
 	if _, err := s.storage.Stat(ctx, blobPath); err == nil {
 		log.Info("Blob already exists in storage, skipping upload",
 			"hash", finishedBlob.Hash, "size", humanize.Bytes(uint64(finishedBlob.Compressed)))
-		s.ui.Info("Blob %s (%s) already exists in backup destination store. Skipping upload.",
-			s.ui.Hex(finishedBlob.Hash), s.ui.Size(finishedBlob.Compressed))
+		s.ui.Info("Blob %s (%s) already exists at %s. Skipping upload.",
+			s.ui.Hex(finishedBlob.Hash), s.ui.Size(finishedBlob.Compressed), s.ui.Path(destination))
 		return true, nil
 	}
 
-	s.ui.Begin("Uploading blob %s (%s) to backup destination store.",
-		s.ui.Hex(finishedBlob.Hash), s.ui.Size(finishedBlob.Compressed))
+	s.ui.Begin("Uploading blob %s (%s) to %s.",
+		s.ui.Hex(finishedBlob.Hash), s.ui.Size(finishedBlob.Compressed), s.ui.Path(destination))
 
 	progressCallback := s.makeUploadProgressCallback(ctx, finishedBlob, startTime)
 

internal/snapshot/snapshot.go

@@ -180,10 +180,20 @@ func (sm *SnapshotManager) UpdateSnapshotStatsExtended(ctx context.Context, snap
 	})
 }
 
-// CompleteSnapshot marks a snapshot as completed and exports its metadata
+// CompleteSnapshot marks a snapshot as completed and ensures snapshot_blobs
+// is populated with every blob holding any chunk referenced by the
+// snapshot's files (including deduplicated blobs uploaded by prior
+// snapshots). Without this, fully-deduplicated snapshots are unrestorable.
 func (sm *SnapshotManager) CompleteSnapshot(ctx context.Context, snapshotID string) error {
-	// Mark the snapshot as completed
 	err := sm.repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		added, err := sm.repos.Snapshots.PopulateReferencedBlobs(ctx, tx, snapshotID)
+		if err != nil {
+			return err
+		}
+		if added > 0 {
+			log.Info("Populated snapshot_blobs with dedup-referenced blobs",
+				"snapshot_id", snapshotID, "added", added)
+		}
 		return sm.repos.Snapshots.MarkComplete(ctx, tx, snapshotID)
 	})
 
@@ -304,10 +314,17 @@ func (sm *SnapshotManager) prepareExportDB(ctx context.Context, dbPath, snapshot
 	return finalData, tempDBPath, nil
 }
 
-// uploadSnapshotArtifacts uploads the database backup and blob manifest to S3
+// uploadSnapshotArtifacts uploads the database backup and blob manifest
+// to remote storage at metadata/<remote-key>/, where remote-key is the
+// double-SHA256 derivation of the snapshot ID (see RemoteSnapshotKey).
+// We never write the human-readable snapshot ID into any unencrypted
+// part of remote storage so a listing of the destination bucket leaks
+// no host, configuration, or scheduling information.
 func (sm *SnapshotManager) uploadSnapshotArtifacts(ctx context.Context, snapshotID string, dbData, manifestData []byte) error {
+	remoteKey := RemoteSnapshotKey(snapshotID)
+
 	// Upload database backup (compressed and encrypted)
-	dbKey := fmt.Sprintf("metadata/%s/db.zst.age", snapshotID)
+	dbKey := fmt.Sprintf("metadata/%s/db.zst.age", remoteKey)
 
 	dbUploadStart := time.Now()
 	if err := sm.storage.Put(ctx, dbKey, bytes.NewReader(dbData)); err != nil {
@@ -322,7 +339,7 @@ func (sm *SnapshotManager) uploadSnapshotArtifacts(ctx context.Context, snapshot
 		"speed", humanize.SI(dbUploadSpeed, "bps"))
 
 	// Upload blob manifest (compressed only, not encrypted)
-	manifestKey := fmt.Sprintf("metadata/%s/manifest.json.zst", snapshotID)
+	manifestKey := fmt.Sprintf("metadata/%s/manifest.json.zst", remoteKey)
 	manifestUploadStart := time.Now()
 	if err := sm.storage.Put(ctx, manifestKey, bytes.NewReader(manifestData)); err != nil {
 		return fmt.Errorf("uploading blob manifest: %w", err)
@@ -597,9 +614,11 @@ func (sm *SnapshotManager) generateBlobManifest(ctx context.Context, dbPath stri
 		}
 	}
 
-	// Create manifest
+	// Create manifest. SnapshotID in the unencrypted manifest is the
+	// double-SHA256 remote key, not the human ID, so the public bytes
+	// don't reveal hostname/snapshot-name/timestamp metadata.
 	manifest := &Manifest{
-		SnapshotID:          snapshotID,
+		SnapshotID:          RemoteSnapshotKey(snapshotID),
 		Timestamp:           time.Now().UTC().Format(time.RFC3339),
 		BlobCount:           len(blobs),
 		TotalCompressedSize: totalCompressedSize,
@@ -670,8 +689,9 @@ func (sm *SnapshotManager) CleanupIncompleteSnapshots(ctx context.Context, hostn
 
 	// Check each incomplete snapshot for metadata in storage
 	for _, snapshot := range incompleteSnapshots {
-		// Check if metadata exists in storage
-		metadataKey := fmt.Sprintf("metadata/%s/db.zst", snapshot.ID)
+		// Check if metadata exists in storage (paths use the hashed
+		// remote key so we don't leak host info to the listing).
+		metadataKey := fmt.Sprintf("metadata/%s/db.zst", RemoteSnapshotKey(snapshot.ID.String()))
 		_, err := sm.storage.Stat(ctx, metadataKey)
 
 		if err != nil {

internal/storage/file.go

@@ -19,15 +19,20 @@ type FileStorer struct {
 }
 
 // NewFileStorer creates a new filesystem storage backend.
-// The basePath directory will be created if it doesn't exist.
-// Uses the real OS filesystem by default; call SetFilesystem to override for testing.
+//
+// Construction is intentionally cheap and does not touch the filesystem.
+// The basePath is recorded; the directory is created lazily on first
+// write. Reads (Get/Stat/List) tolerate a missing basePath — a missing
+// or unmounted destination during `snapshot list` should NOT block the
+// command, it should degrade to "no remote snapshots reachable" with a
+// warning. Write operations (Put/PutWithProgress) call MkdirAll for the
+// per-blob parent directory, which also covers basePath on first use.
+//
+// Uses the real OS filesystem by default; call SetFilesystem to
+// override for testing.
 func NewFileStorer(basePath string) (*FileStorer, error) {
-	fs := afero.NewOsFs()
-	if err := fs.MkdirAll(basePath, 0755); err != nil {
-		return nil, fmt.Errorf("file:// storage: cannot create or access %s: %w (check that the volume is mounted and writable)", basePath, err)
-	}
 	return &FileStorer{
-		fs:       fs,
+		fs:       afero.NewOsFs(),
 		basePath: basePath,
 	}, nil
 }

internal/ui/ui.go

@@ -49,9 +49,15 @@ const Marker = "》"
 // It also counts warnings and errors emitted so the caller can summarize at
 // the end of an operation ("Finished successfully." vs "Finished with
 // warnings.").
+//
+// When Quiet is set, Begin/Complete/Info/Notice/Detail/Progress/Banner
+// are silently dropped, but Warning and Error always emit. This honors
+// the convention that --quiet "Suppresses non-error output" — warnings
+// and errors are by definition not suppressible.
 type Writer struct {
 	out      io.Writer
 	color    bool
+	quiet    bool
 	warnings int
 	errors   int
 }
@@ -70,6 +76,13 @@ func NewWithColor(out io.Writer, color bool) *Writer {
 	return &Writer{out: out, color: color}
 }
 
+// SetQuiet toggles the writer's quiet mode. In quiet mode all message
+// classes are silenced except Warning and Error.
+func (w *Writer) SetQuiet(quiet bool) { w.quiet = quiet }
+
+// Quiet reports whether the writer is in quiet mode.
+func (w *Writer) Quiet() bool { return w.quiet }
+
 // Out returns the underlying writer.
 func (w *Writer) Out() io.Writer { return w.out }
 
@@ -100,21 +113,33 @@ func (w *Writer) paint(color, s string) string {
 
 // Begin prints an operation-start line, left-aligned with a white marker.
 func (w *Writer) Begin(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	w.emit(ansiWhite, Marker, "", format, args)
 }
 
 // Complete prints an operation-completion line in green, left-aligned.
 func (w *Writer) Complete(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	w.emit(ansiGreen, Marker, ansiGreen, format, args)
 }
 
 // Info prints a neutral status line, left-aligned with a white marker.
 func (w *Writer) Info(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	w.emit(ansiWhite, Marker, "", format, args)
 }
 
 // Notice prints an attention-worthy informational line, marker in cyan.
 func (w *Writer) Notice(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	w.emit(ansiCyan, Marker, "", format, args)
 }
 
@@ -139,6 +164,9 @@ func (w *Writer) Error(format string, args ...any) {
 // Distinct from Progress (semantically a "heartbeat") in usage but
 // visually identical.
 func (w *Writer) Detail(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	w.emit(ansiWhite, "  "+Marker, "", format, args)
 }
 
@@ -150,12 +178,18 @@ func (w *Writer) ErrorCount() int { return w.errors }
 
 // Progress prints an indented heartbeat / per-item update, marker in white.
 func (w *Writer) Progress(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	w.emit(ansiWhite, "  "+Marker, "", format, args)
 }
 
 // Banner prints a line with no marker, left-aligned. Bold when color
 // is enabled. Used for the application startup banner only.
 func (w *Writer) Banner(format string, args ...any) {
+	if w.quiet {
+		return
+	}
 	body := fmt.Sprintf(format, args...)
 	if w.color {
 		body = ansiBold + body + ansiReset
@@ -203,9 +237,26 @@ func (w *Writer) Size(bytes int64) string {
 	return w.paint(ansiMagenta, humanize.Bytes(uint64(bytes)))
 }
 
-// Speed colorizes a byte-per-second value as "<size>/sec".
+// Speed colorizes a network transfer rate. Input is bytes/sec; output is
+// bits/sec with an appropriate SI unit (bit/s, Kbit/s, Mbit/s, Gbit/s) —
+// network transfer rates are conventionally expressed in bits.
 func (w *Writer) Speed(bytesPerSec float64) string {
-	return w.paint(ansiMagenta, humanize.Bytes(uint64(bytesPerSec))+"/sec")
+	if bytesPerSec <= 0 {
+		return w.paint(ansiMagenta, "N/A")
+	}
+	bitsPerSec := bytesPerSec * 8
+	var s string
+	switch {
+	case bitsPerSec >= 1e9:
+		s = fmt.Sprintf("%.1f Gbit/sec", bitsPerSec/1e9)
+	case bitsPerSec >= 1e6:
+		s = fmt.Sprintf("%.0f Mbit/sec", bitsPerSec/1e6)
+	case bitsPerSec >= 1e3:
+		s = fmt.Sprintf("%.0f Kbit/sec", bitsPerSec/1e3)
+	default:
+		s = fmt.Sprintf("%.0f bit/sec", bitsPerSec)
+	}
+	return w.paint(ansiMagenta, s)
 }
 
 // Duration colorizes a time.Duration rounded to the nearest second.

internal/ui/ui_test.go

@@ -100,6 +100,17 @@ func TestValueFormattersPlain(t *testing.T) {
 		t.Errorf("Percent: got %q", got)
 	}
 
+	// Speed: input is bytes/sec, output is bits/sec.
+	if got := w.Speed(0); got != "N/A" {
+		t.Errorf("Speed(0): got %q, want N/A", got)
+	}
+	if got := w.Speed(125_000_000); got != "1.0 Gbit/sec" { // 1 Gbit/s = 125 MB/s
+		t.Errorf("Speed(125e6): got %q", got)
+	}
+	if got := w.Speed(125_000); got != "1 Mbit/sec" {
+		t.Errorf("Speed(125e3): got %q", got)
+	}
+
 	// Time format: today → HH:MM:SS, other day → YYYY-MM-DD HH:MM:SS.
 	today := time.Date(time.Now().Year(), time.Now().Month(), time.Now().Day(), 14, 30, 45, 0, time.Local)
 	if got := w.Time(today); got != "14:30:45" {

internal/vaultik/blob_fetch.go

@@ -6,9 +6,11 @@ import (
 	"encoding/hex"
 	"fmt"
 	"io"
+	"time"
 
 	"filippo.io/age"
 	"sneak.berlin/go/vaultik/internal/blobgen"
+	"sneak.berlin/go/vaultik/internal/log"
 )
 
 // hashVerifyReader wraps a blobgen.Reader and verifies the double-SHA-256 hash
@@ -75,19 +77,34 @@ func (v *Vaultik) FetchAndDecryptBlob(ctx context.Context, blobHash string, expe
 }
 
 // FetchBlob downloads a blob and returns a reader for the encrypted data.
+// Times the Storage.Get and Storage.Stat round-trips separately at
+// debug level so we can see whether the size-only Stat (which is an
+// extra request on every fetch) is hurting throughput.
 func (v *Vaultik) FetchBlob(ctx context.Context, blobHash string, expectedSize int64) (io.ReadCloser, int64, error) {
 	blobPath := fmt.Sprintf("blobs/%s/%s/%s", blobHash[:2], blobHash[2:4], blobHash)
 
+	t0 := time.Now()
 	rc, err := v.Storage.Get(ctx, blobPath)
+	getDur := time.Since(t0)
 	if err != nil {
 		return nil, 0, fmt.Errorf("downloading blob %s: %w", blobHash[:16], err)
 	}
 
+	t0 = time.Now()
 	info, err := v.Storage.Stat(ctx, blobPath)
+	statDur := time.Since(t0)
 	if err != nil {
 		_ = rc.Close()
 		return nil, 0, fmt.Errorf("stat blob %s: %w", blobHash[:16], err)
 	}
 
+	log.Debug("FetchBlob round-trips",
+		"hash", blobHash[:16],
+		"ms_storage_get", getDur.Milliseconds(),
+		"ms_storage_stat", statDur.Milliseconds(),
+		"expected_size", expectedSize,
+		"stat_size", info.Size,
+	)
+
 	return rc, info.Size, nil
 }

internal/vaultik/blobcache.go

@@ -2,6 +2,7 @@ package vaultik
 
 import (
 	"fmt"
+	"io"
 	"os"
 	"path/filepath"
 	"sync"
@@ -15,9 +16,22 @@ type blobDiskCacheEntry struct {
 	next *blobDiskCacheEntry
 }
 
-// blobDiskCache is an LRU cache that stores blobs on disk instead of in memory.
-// Blobs are written to a temp directory keyed by their hash. When total size
-// exceeds maxBytes, the least-recently-used entries are evicted (deleted from disk).
+// blobDiskCache stores blobs on disk keyed by hash. It exposes ReadAt
+// for slice reads (the restore path uses this so chunk extraction
+// never reads a whole blob into memory) plus Get/Put for whole-blob
+// access.
+//
+// Eviction policy is caller-controlled. The cache keeps an LRU list
+// internally and will fall back to LRU eviction if curBytes exceeds
+// maxBytes. Restore passes math.MaxInt64 as maxBytes and drives
+// eviction itself via Delete() through restoreSweeper, which deletes
+// each blob the moment every file that references its chunks has been
+// written. LRU never fires under that configuration; it is kept as a
+// safety net for callers that don't manage eviction themselves.
+//
+// Get/ReadAt/peak-Len counters are debugging instrumentation used by
+// tests to assert that the restore code path uses ReadAt rather than
+// Get and to bound peak disk-cache occupancy.
 type blobDiskCache struct {
 	mu       sync.Mutex
 	dir      string
@@ -26,6 +40,11 @@ type blobDiskCache struct {
 	items    map[string]*blobDiskCacheEntry
 	head     *blobDiskCacheEntry // most recent
 	tail     *blobDiskCacheEntry // least recent
+
+	// Instrumentation. Mutated under mu; readable via the methods below.
+	getCalls    int
+	readAtCalls int
+	peakLen     int
 }
 
 // newBlobDiskCache creates a new disk-based blob cache with the given max size.
@@ -115,12 +134,77 @@ func (c *blobDiskCache) Put(key string, data []byte) error {
 		c.evictLRU()
 	}
 
+	if n := len(c.items); n > c.peakLen {
+		c.peakLen = n
+	}
+
 	return nil
 }
 
+// PutFromReader streams r into the cache file for key, returning the
+// total number of bytes written. Unlike Put, the data never has to
+// reside fully in memory at any point — io.Copy uses an internal
+// 32 KiB buffer. Used by restore to land a freshly decrypted blob on
+// disk without buffering its entire plaintext (which may be tens of GB)
+// in RAM.
+func (c *blobDiskCache) PutFromReader(key string, r io.Reader) (int64, error) {
+	c.mu.Lock()
+	// Remove any prior entry first; we'll re-link after the file is
+	// written successfully.
+	if e, ok := c.items[key]; ok {
+		c.unlink(e)
+		c.curBytes -= e.size
+		_ = os.Remove(c.path(key))
+		delete(c.items, key)
+	}
+	c.mu.Unlock()
+
+	f, err := os.OpenFile(c.path(key), os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0o600)
+	if err != nil {
+		return 0, fmt.Errorf("creating cache file: %w", err)
+	}
+	written, copyErr := io.Copy(f, r)
+	closeErr := f.Close()
+	if copyErr != nil {
+		_ = os.Remove(c.path(key))
+		return written, fmt.Errorf("streaming to cache file: %w", copyErr)
+	}
+	if closeErr != nil {
+		_ = os.Remove(c.path(key))
+		return written, fmt.Errorf("closing cache file: %w", closeErr)
+	}
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	// If the entry would exceed maxBytes outright, drop it on the
+	// floor — but the restore path passes math.MaxInt64 as maxBytes
+	// so this branch is effectively unreachable there.
+	if written > c.maxBytes {
+		_ = os.Remove(c.path(key))
+		return written, nil
+	}
+
+	e := &blobDiskCacheEntry{key: key, size: written}
+	c.pushFront(e)
+	c.items[key] = e
+	c.curBytes += written
+
+	for c.curBytes > c.maxBytes && c.tail != nil {
+		c.evictLRU()
+	}
+
+	if n := len(c.items); n > c.peakLen {
+		c.peakLen = n
+	}
+
+	return written, nil
+}
+
 // Get reads a cached blob from disk. Returns data and true on hit.
 func (c *blobDiskCache) Get(key string) ([]byte, bool) {
 	c.mu.Lock()
+	c.getCalls++
 	e, ok := c.items[key]
 	if !ok {
 		c.mu.Unlock()
@@ -147,6 +231,7 @@ func (c *blobDiskCache) Get(key string) ([]byte, bool) {
 // ReadAt reads a slice of a cached blob without loading the entire blob into memory.
 func (c *blobDiskCache) ReadAt(key string, offset, length int64) ([]byte, error) {
 	c.mu.Lock()
+	c.readAtCalls++
 	e, ok := c.items[key]
 	if !ok {
 		c.mu.Unlock()
@@ -181,6 +266,34 @@ func (c *blobDiskCache) Has(key string) bool {
 	return ok
 }
 
+// Delete removes a blob from the cache and its disk file. No-op if absent.
+// Used by restore's sweep logic to free blobs whose chunks have all been
+// restored (so they will never be needed again during this restore).
+func (c *blobDiskCache) Delete(key string) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	e, ok := c.items[key]
+	if !ok {
+		return
+	}
+	c.unlink(e)
+	delete(c.items, key)
+	c.curBytes -= e.size
+	_ = os.Remove(c.path(key))
+}
+
+// Keys returns a snapshot of all cached keys. Safe for iteration without
+// holding the cache lock; the cache may change concurrently.
+func (c *blobDiskCache) Keys() []string {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	keys := make([]string, 0, len(c.items))
+	for k := range c.items {
+		keys = append(keys, k)
+	}
+	return keys
+}
+
 // Size returns current total cached bytes.
 func (c *blobDiskCache) Size() int64 {
 	c.mu.Lock()
@@ -195,6 +308,28 @@ func (c *blobDiskCache) Len() int {
 	return len(c.items)
 }
 
+// GetCalls returns the number of times Get has been called.
+func (c *blobDiskCache) GetCalls() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.getCalls
+}
+
+// ReadAtCalls returns the number of times ReadAt has been called.
+func (c *blobDiskCache) ReadAtCalls() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.readAtCalls
+}
+
+// PeakLen returns the maximum number of cached entries ever held at
+// once during this cache's lifetime.
+func (c *blobDiskCache) PeakLen() int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.peakLen
+}
+
 // Close removes the cache directory and all cached blobs.
 func (c *blobDiskCache) Close() error {
 	c.mu.Lock()

internal/vaultik/helpers.go

@@ -10,11 +10,17 @@ import (
 	"sneak.berlin/go/vaultik/internal/types"
 )
 
-// SnapshotInfo contains information about a snapshot
+// SnapshotInfo contains information about a snapshot.
+// UncompressedSize and NewChunkSize are populated only when the snapshot
+// is present in the local database; LocallyTracked indicates whether
+// those values are meaningful.
 type SnapshotInfo struct {
-	ID             types.SnapshotID `json:"id"`
-	Timestamp      time.Time        `json:"timestamp"`
-	CompressedSize int64            `json:"compressed_size"`
+	ID               types.SnapshotID `json:"id"`
+	Timestamp        time.Time        `json:"timestamp"`
+	CompressedSize   int64            `json:"compressed_size"`
+	UncompressedSize int64            `json:"uncompressed_size,omitempty"`
+	NewChunkSize     int64            `json:"new_chunk_size,omitempty"`
+	LocallyTracked   bool             `json:"locally_tracked"`
 }
 
 // formatBytes formats bytes in a human-readable format

internal/vaultik/info.go

@@ -22,14 +22,29 @@ func (v *Vaultik) ShowInfo() error {
 	v.printfStdout("Go Version:      %s\n", runtime.Version())
 	v.printlnStdout()
 
-	// Storage Configuration
+	// Storage Configuration. The backend is selected by storage_url
+	// (s3://, file://, rclone://); the legacy s3.* fields are only
+	// printed when they're actually populated, since the URL scheme
+	// is the primary configuration.
 	v.printfStdout("=== Storage Configuration ===\n")
-	v.printfStdout("S3 Bucket:       %s\n", v.Config.S3.Bucket)
+	storageInfo := v.Storage.Info()
+	v.printfStdout("Type:            %s\n", storageInfo.Type)
+	v.printfStdout("Location:        %s\n", storageInfo.Location)
+	if v.Config.StorageURL != "" {
+		v.printfStdout("Storage URL:     %s\n", v.Config.StorageURL)
+	}
+	if v.Config.S3.Bucket != "" {
+		v.printfStdout("S3 Bucket:       %s\n", v.Config.S3.Bucket)
+	}
 	if v.Config.S3.Prefix != "" {
 		v.printfStdout("S3 Prefix:       %s\n", v.Config.S3.Prefix)
 	}
-	v.printfStdout("S3 Endpoint:     %s\n", v.Config.S3.Endpoint)
-	v.printfStdout("S3 Region:       %s\n", v.Config.S3.Region)
+	if v.Config.S3.Endpoint != "" {
+		v.printfStdout("S3 Endpoint:     %s\n", v.Config.S3.Endpoint)
+	}
+	if v.Config.S3.Region != "" {
+		v.printfStdout("S3 Region:       %s\n", v.Config.S3.Region)
+	}
 	v.printlnStdout()
 
 	// Backup Settings
@@ -337,7 +352,7 @@ func (v *Vaultik) printRemoteInfoTable(result *RemoteInfoResult) {
 		humanize.Comma(int64(result.OrphanedBlobCount)), humanize.Bytes(uint64(result.OrphanedBlobSize)))
 
 	if result.OrphanedBlobCount > 0 {
-		v.printfStdout("\nRun 'vaultik prune --remote' to remove orphaned blobs.\n")
+		v.printfStdout("\nRun 'vaultik prune' to remove orphaned blobs.\n")
 	}
 }
 

internal/vaultik/integration_test.go

@@ -20,6 +20,7 @@ import (
 	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/storage"
 	"sneak.berlin/go/vaultik/internal/types"
+	"sneak.berlin/go/vaultik/internal/ui"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
@@ -520,6 +521,7 @@ func TestBackupAndRestore(t *testing.T) {
 		Fs:      fs,
 		Stdout:  io.Discard,
 		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
 	}
 	vaultikApp.SetContext(ctx)
 
@@ -649,10 +651,12 @@ func TestEndToEndFileStorage(t *testing.T) {
 	require.NoError(t, sm.ExportSnapshotMetadata(ctx, dbPath, snapshotID))
 
 	// Verify the backup actually landed on disk under blobs/ and metadata/.
+	// The metadata subdirectory uses the hashed remote key, not the human
+	// snapshot ID, so the on-disk structure doesn't leak hostname/name/time.
 	blobInfo, err := os.Stat(filepath.Join(storeDir, "blobs"))
 	require.NoError(t, err)
 	require.True(t, blobInfo.IsDir())
-	metaInfo, err := os.Stat(filepath.Join(storeDir, "metadata", snapshotID))
+	metaInfo, err := os.Stat(filepath.Join(storeDir, "metadata", snapshot.RemoteSnapshotKey(snapshotID)))
 	require.NoError(t, err)
 	require.True(t, metaInfo.IsDir())
 
@@ -666,6 +670,7 @@ func TestEndToEndFileStorage(t *testing.T) {
 		Fs:      fs,
 		Stdout:  io.Discard,
 		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
 	}
 	restoreVaultik.SetContext(ctx)
 
@@ -703,6 +708,124 @@ func TestEndToEndFileStorage(t *testing.T) {
 	assert.Equal(t, "small.txt", target, "symlink target should be preserved")
 }
 
+// TestDedupOnlySnapshotRestores backs up the same directory twice without
+// touching it between runs, then restores the SECOND (fully-deduplicated)
+// snapshot. The second snapshot uploads no new blobs — every chunk is
+// already in storage from the first run. This test guards against the
+// regression where snapshot_blobs was populated only for blobs uploaded
+// during the snapshot, leaving fully-deduplicated snapshots unrestorable
+// with "chunk X not found in any blob" errors.
+func TestDedupOnlySnapshotRestores(t *testing.T) {
+	log.Initialize(log.Config{})
+
+	fs := afero.NewOsFs()
+	tempDir, err := os.MkdirTemp("", "vaultik-dedup-")
+	require.NoError(t, err)
+	defer func() { _ = os.RemoveAll(tempDir) }()
+
+	dataDir := filepath.Join(tempDir, "source")
+	storeDir := filepath.Join(tempDir, "remote")
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	chunkSize := int64(64 * 1024)
+	maxBlobSize := int64(512 * 1024)
+
+	testFiles := map[string][]byte{
+		filepath.Join(dataDir, "a.bin"): bytesPattern("a-", int(chunkSize*3)),
+		filepath.Join(dataDir, "b.bin"): bytesPattern("b-", int(chunkSize*2)),
+	}
+	for path, content := range testFiles {
+		require.NoError(t, fs.MkdirAll(filepath.Dir(path), 0o755))
+		require.NoError(t, afero.WriteFile(fs, path, content, 0o644))
+	}
+
+	storer, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	agePublicKey := "age1ezrjmfpwsc95svdg0y54mums3zevgzu0x0ecq2f7tp8a05gl0sjq9q9wjg"
+	ageSecretKey := "AGE-SECRET-KEY-19CR5YSFW59HM4TLD6GXVEDMZFTVVF7PPHKUT68TXSFPK7APHXA2QS2NJA5"
+
+	cfg := &config.Config{
+		AgeRecipients:    []string{agePublicKey},
+		AgeSecretKey:     ageSecretKey,
+		CompressionLevel: 3,
+		Hostname:         "test-host",
+	}
+
+	ctx := context.Background()
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+	repos := database.NewRepositories(db)
+
+	makeScanner := func() *snapshot.Scanner {
+		return snapshot.NewScanner(snapshot.ScannerConfig{
+			FS:               fs,
+			Storage:          storer,
+			ChunkSize:        chunkSize,
+			MaxBlobSize:      maxBlobSize,
+			CompressionLevel: cfg.CompressionLevel,
+			AgeRecipients:    cfg.AgeRecipients,
+			Repositories:     repos,
+		})
+	}
+	sm := snapshot.NewSnapshotManager(snapshot.SnapshotManagerParams{
+		Repos: repos, Storage: storer, Config: cfg,
+	})
+	sm.SetFilesystem(fs)
+
+	// First snapshot — uploads all blobs.
+	id1, err := sm.CreateSnapshotWithName(ctx, cfg.Hostname, "dedup", "v", "g")
+	require.NoError(t, err)
+	r1, err := makeScanner().Scan(ctx, dataDir, id1)
+	require.NoError(t, err)
+	require.Greater(t, r1.BlobsCreated, 0, "first snapshot should upload at least one blob")
+	require.NoError(t, sm.CompleteSnapshot(ctx, id1))
+	require.NoError(t, sm.ExportSnapshotMetadata(ctx, dbPath, id1))
+
+	// Second snapshot — same data, every chunk dedups. Sleep past the
+	// second-precision timestamp so the snapshot IDs differ.
+	time.Sleep(1100 * time.Millisecond)
+	id2, err := sm.CreateSnapshotWithName(ctx, cfg.Hostname, "dedup", "v", "g")
+	require.NoError(t, err)
+	r2, err := makeScanner().Scan(ctx, dataDir, id2)
+	require.NoError(t, err)
+	require.Equal(t, 0, r2.BlobsCreated, "second snapshot should upload zero new blobs (fully dedup'd)")
+	require.NoError(t, sm.CompleteSnapshot(ctx, id2))
+	require.NoError(t, sm.ExportSnapshotMetadata(ctx, dbPath, id2))
+
+	// snapshot_blobs for id2 must be populated despite no uploads.
+	blobHashes, err := repos.Snapshots.GetBlobHashes(ctx, id2)
+	require.NoError(t, err)
+	require.NotEmpty(t, blobHashes, "snapshot_blobs for fully-dedup'd snapshot must reference blobs uploaded by prior snapshot")
+
+	require.NoError(t, db.Close())
+
+	restoreVaultik := &vaultik.Vaultik{
+		Config:  cfg,
+		Storage: storer,
+		Fs:      fs,
+		Stdout:  io.Discard,
+		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
+	}
+	restoreVaultik.SetContext(ctx)
+
+	require.NoError(t, restoreVaultik.Restore(&vaultik.RestoreOptions{
+		SnapshotID: id2,
+		TargetDir:  restoreDir,
+		Verify:     true,
+	}))
+
+	for origPath, expected := range testFiles {
+		restoredPath := filepath.Join(restoreDir, origPath)
+		got, err := afero.ReadFile(fs, restoredPath)
+		require.NoError(t, err, "restored file missing: %s", restoredPath)
+		require.Equalf(t, expected, got, "byte-equality failed for %s", origPath)
+	}
+}
+
 // bytesPattern returns a deterministic byte slice of length n with a tag prefix,
 // useful for forcing chunker behavior with reproducible content.
 func bytesPattern(tag string, n int) []byte {

internal/vaultik/prune.go

@@ -15,6 +15,31 @@ type PruneOptions struct {
 	JSON  bool
 }
 
+// NukeRemote deletes every snapshot's metadata and every blob from remote
+// storage. After this returns successfully the bucket prefix is empty and
+// the next backup starts from scratch.
+//
+// Refuses to run unless force is true. The caller is responsible for
+// confirming with the user.
+func (v *Vaultik) NukeRemote(force bool) error {
+	if !force {
+		return fmt.Errorf("nuke requires --force (this deletes ALL remote snapshots and blobs)")
+	}
+
+	v.UI.Begin("Removing all snapshot metadata from backup destination store.")
+	if _, err := v.RemoveAllSnapshots(&RemoveOptions{Force: true, Remote: true}); err != nil {
+		return fmt.Errorf("removing all snapshots: %w", err)
+	}
+
+	v.UI.Begin("Removing all blobs from backup destination store.")
+	if err := v.PruneBlobs(&PruneOptions{Force: true}); err != nil {
+		return fmt.Errorf("pruning blobs: %w", err)
+	}
+
+	v.UI.Complete("Backup destination store is now empty.")
+	return nil
+}
+
 // PruneBlobsResult contains the result of a blob prune operation
 type PruneBlobsResult struct {
 	BlobsFound   int   `json:"blobs_found"`
@@ -23,6 +48,19 @@ type PruneBlobsResult struct {
 	BytesFreed   int64 `json:"bytes_freed"`
 }
 
+// Prune removes orphaned data from the local index database AND
+// unreferenced blobs from the backup destination store. This is the
+// single user-facing prune entry point — the split between local and
+// remote cleanup is an implementation detail. Calling code should
+// prefer this method over PruneDatabase or PruneBlobs individually
+// unless it specifically wants one half.
+func (v *Vaultik) Prune(opts *PruneOptions) error {
+	if _, err := v.PruneDatabase(); err != nil {
+		return fmt.Errorf("pruning local database: %w", err)
+	}
+	return v.PruneBlobs(opts)
+}
+
 // PruneBlobs removes unreferenced blobs from storage
 func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 	log.Info("Starting prune operation")
@@ -85,20 +123,22 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 // collectReferencedBlobs downloads all manifests and returns the set of referenced blob hashes
 func (v *Vaultik) collectReferencedBlobs() (map[string]bool, error) {
 	log.Info("Listing remote snapshots")
-	snapshotIDs, err := v.listUniqueSnapshotIDs()
+	// IDs returned by listUniqueSnapshotIDs are remote keys (hashed
+	// subdirectories under metadata/), not human snapshot IDs.
+	remoteKeys, err := v.listUniqueSnapshotIDs()
 	if err != nil {
-		return nil, fmt.Errorf("listing snapshot IDs: %w", err)
+		return nil, fmt.Errorf("listing snapshot keys: %w", err)
 	}
-	log.Info("Found manifests in remote storage", "count", len(snapshotIDs))
+	log.Info("Found manifests in remote storage", "count", len(remoteKeys))
 
 	allBlobsReferenced := make(map[string]bool)
 	manifestCount := 0
 
-	for _, snapshotID := range snapshotIDs {
-		log.Debug("Processing manifest", "snapshot_id", snapshotID)
-		manifest, err := v.downloadManifest(snapshotID)
+	for _, remoteKey := range remoteKeys {
+		log.Debug("Processing manifest", "remote_key", remoteKey)
+		manifest, err := v.downloadManifestByKey(remoteKey)
 		if err != nil {
-			log.Error("Failed to download manifest", "snapshot_id", snapshotID, "error", err)
+			log.Error("Failed to download manifest", "remote_key", remoteKey, "error", err)
 			continue
 		}
 		for _, blob := range manifest.Blobs {

internal/vaultik/remove_snapshot_test.go

@@ -132,7 +132,9 @@ func (s *testStorer) Info() storage.StorageInfo {
 	}
 }
 
-// addManifest creates a compressed manifest in storage
+// addManifest creates a compressed manifest in storage at the same
+// hashed path the production code uses. snapshotID is the human ID;
+// the storage path uses RemoteSnapshotKey(id).
 func addManifest(t *testing.T, store *testStorer, snapshotID string, blobHashes []string) {
 	t.Helper()
 
@@ -144,8 +146,9 @@ func addManifest(t *testing.T, store *testStorer, snapshotID string, blobHashes
 		}
 	}
 
+	remoteKey := snapshot.RemoteSnapshotKey(snapshotID)
 	manifest := &snapshot.Manifest{
-		SnapshotID: snapshotID,
+		SnapshotID: remoteKey,
 		BlobCount:  len(blobs),
 		Blobs:      blobs,
 	}
@@ -153,11 +156,19 @@ func addManifest(t *testing.T, store *testStorer, snapshotID string, blobHashes
 	data, err := snapshot.EncodeManifest(manifest, 3)
 	require.NoError(t, err)
 
-	key := "metadata/" + snapshotID + "/manifest.json.zst"
+	key := "metadata/" + remoteKey + "/manifest.json.zst"
 	err = store.Put(context.Background(), key, bytes.NewReader(data))
 	require.NoError(t, err)
 }
 
+// remoteKeyPath returns the storage-relative path to a snapshot's
+// metadata directory or manifest under the hashed remote-key scheme.
+// Tests use this in hasKey/asserts to avoid scattering RemoteSnapshotKey
+// calls throughout.
+func remoteKeyPath(snapshotID, suffix string) string {
+	return "metadata/" + snapshot.RemoteSnapshotKey(snapshotID) + "/" + suffix
+}
+
 // addBlob adds a fake blob to storage
 func addBlob(t *testing.T, store *testStorer, hash string) {
 	t.Helper()
@@ -198,7 +209,7 @@ func TestRemoveSnapshot_LocalOnly(t *testing.T) {
 	// Blobs should NOT be deleted (that's what prune is for)
 	assert.True(t, store.hasKey("blobs/aa/aa/"+blobA))
 	// Remote metadata should NOT be deleted (no --remote flag)
-	assert.True(t, store.hasKey("metadata/snapshot-001/manifest.json.zst"))
+	assert.True(t, store.hasKey(remoteKeyPath("snapshot-001", "manifest.json.zst")))
 
 	// Verify output
 	assert.Contains(t, tv.Stdout.String(), "Removed snapshot 'snapshot-001' from local database")
@@ -225,7 +236,7 @@ func TestRemoveSnapshot_WithRemote(t *testing.T) {
 	// Blobs should NOT be deleted
 	assert.True(t, store.hasKey("blobs/aa/aa/"+blobA))
 	// Remote metadata SHOULD be deleted
-	assert.False(t, store.hasKey("metadata/snapshot-001/manifest.json.zst"))
+	assert.False(t, store.hasKey(remoteKeyPath("snapshot-001", "manifest.json.zst")))
 
 	// Verify output mentions prune
 	assert.Contains(t, tv.Stdout.String(), "Removed snapshot 'snapshot-001' from local database")
@@ -255,7 +266,7 @@ func TestRemoveSnapshot_DryRun(t *testing.T) {
 	// Nothing should be deleted
 	assert.Equal(t, initialCount, store.keyCount())
 	assert.True(t, store.hasKey("blobs/aa/aa/"+blobA))
-	assert.True(t, store.hasKey("metadata/snapshot-001/manifest.json.zst"))
+	assert.True(t, store.hasKey(remoteKeyPath("snapshot-001", "manifest.json.zst")))
 
 	// Verify dry run message
 	assert.Contains(t, tv.Stdout.String(), "[Dry run - no changes made]")
@@ -299,8 +310,8 @@ func TestRemoveAllSnapshots_WithForce(t *testing.T) {
 	// Blobs should NOT be deleted
 	assert.True(t, store.hasKey("blobs/aa/aa/"+blobA))
 	// Remote metadata SHOULD be deleted
-	assert.False(t, store.hasKey("metadata/snapshot-001/manifest.json.zst"))
-	assert.False(t, store.hasKey("metadata/snapshot-002/manifest.json.zst"))
+	assert.False(t, store.hasKey(remoteKeyPath("snapshot-001", "manifest.json.zst")))
+	assert.False(t, store.hasKey(remoteKeyPath("snapshot-002", "manifest.json.zst")))
 
 	// Verify output
 	assert.Contains(t, tv.Stdout.String(), "Removed 2 snapshot(s)")
@@ -318,7 +329,10 @@ func TestRemoveAllSnapshots_DryRun(t *testing.T) {
 
 	tv := vaultik.NewForTesting(store)
 
-	opts := &vaultik.RemoveOptions{All: true, Force: true, DryRun: true}
+	// --remote is required to enumerate orphan remote keys; without
+	// it, RemoveAll only acts on local snapshots, and NewForTesting
+	// has no local DB.
+	opts := &vaultik.RemoveOptions{All: true, Force: true, DryRun: true, Remote: true}
 	result, err := tv.RemoveAllSnapshots(opts)
 
 	require.NoError(t, err)

internal/vaultik/restore.go

@@ -7,34 +7,28 @@ import (
 	"encoding/hex"
 	"fmt"
 	"io"
+	"math"
 	"os"
 	"path/filepath"
 	"time"
 
 	"filippo.io/age"
 	"github.com/dustin/go-humanize"
-	"github.com/schollz/progressbar/v3"
 	"github.com/spf13/afero"
-	"golang.org/x/term"
 	"sneak.berlin/go/vaultik/internal/blobgen"
 	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/types"
 )
 
-const (
-	// progressBarWidth is the character width of the progress bar display.
-	progressBarWidth = 40
-	// progressBarThrottle is the minimum interval between progress bar redraws.
-	progressBarThrottle = 100 * time.Millisecond
-)
-
 // RestoreOptions contains options for the restore operation
 type RestoreOptions struct {
 	SnapshotID string
 	TargetDir  string
 	Paths      []string // Optional paths to restore (empty = all)
 	Verify     bool     // Verify restored files by checking chunk hashes
+	SkipErrors bool     // Continue past file-restore errors instead of aborting
 }
 
 // RestoreResult contains statistics from a restore operation
@@ -92,10 +86,12 @@ func (v *Vaultik) Restore(opts *RestoreOptions) error {
 
 	if len(files) == 0 {
 		log.Warn("No files found to restore")
+		v.UI.Warning("No files found to restore.")
 		return nil
 	}
 
 	log.Info("Found files to restore", "count", len(files))
+	v.UI.Info("Found %s files to restore.", v.UI.Count(len(files)))
 
 	// Step 3: Create target directory
 	if err := v.Fs.MkdirAll(opts.TargetDir, 0755); err != nil {
@@ -124,16 +120,16 @@ func (v *Vaultik) Restore(opts *RestoreOptions) error {
 		"duration", result.Duration,
 	)
 
-	v.printfStdout("Restored %d files (%s) in %s\n",
-		result.FilesRestored,
-		humanize.Bytes(uint64(result.BytesRestored)),
-		result.Duration.Round(time.Second),
+	v.UI.Complete("Restored %s files (%s) in %s.",
+		v.UI.Count(result.FilesRestored),
+		v.UI.Size(result.BytesRestored),
+		v.UI.Duration(result.Duration),
 	)
 
 	if result.FilesFailed > 0 {
-		_, _ = fmt.Fprintf(v.Stdout, "\nWARNING: %d file(s) failed to restore:\n", result.FilesFailed)
+		v.UI.Warning("%d file(s) failed to restore:", result.FilesFailed)
 		for _, path := range result.FailedFiles {
-			_, _ = fmt.Fprintf(v.Stdout, "  - %s\n", path)
+			v.UI.Detail("%s", v.UI.Path(path))
 		}
 	}
 
@@ -164,7 +160,12 @@ func (v *Vaultik) prepareRestoreIdentity() (age.Identity, error) {
 	return identity, nil
 }
 
-// restoreAllFiles iterates over files and restores each one, tracking progress and failures
+// restoreAllFiles processes files in blob-locality order: drain every
+// file whose blob set is on disk, download the missing blobs for the
+// pending file with the smallest uncached count, repeat. This keeps
+// peak cache occupancy near 1 even on snapshots whose path order
+// interleaves blobs, and lets the sweeper free each blob the moment
+// its file set is exhausted.
 func (v *Vaultik) restoreAllFiles(
 	files []*database.File,
 	repos *database.Repositories,
@@ -173,56 +174,199 @@ func (v *Vaultik) restoreAllFiles(
 	chunkToBlobMap map[string]*database.BlobChunk,
 ) (*RestoreResult, error) {
 	result := &RestoreResult{}
-	blobCache, err := newBlobDiskCache(4 * v.Config.BlobSizeLimit.Int64())
+
+	// The restore-side blob cache is unbounded — restores may read any
+	// blob many times across deduplicated files and we want to avoid
+	// re-downloading until we can prove a blob is no longer needed.
+	// Cleanup is driven by the sweeper below, not by LRU.
+	blobCache, err := newBlobDiskCache(math.MaxInt64)
 	if err != nil {
 		return nil, fmt.Errorf("creating blob cache: %w", err)
 	}
-	defer func() { _ = blobCache.Close() }()
+	if v.restoreCacheObserver != nil {
+		v.restoreCacheObserver(blobCache)
+	}
+	defer func() {
+		if v.restoreCacheObserver != nil {
+			v.restoreCacheObserver(blobCache)
+		}
+		_ = blobCache.Close()
+	}()
 
-	// Calculate total bytes for progress bar
+	// Per-restore sweep state: every blob_size_limit/100 bytes written,
+	// scan the cache and delete any blob whose remaining file references
+	// are all already restored.
+	sweeper := newRestoreSweeper(v.ctx, repos, blobCache, v.Config.BlobSizeLimit.Int64()/100)
+
+	// Pre-fetch every blob row once so chunk extraction can map a
+	// blob_id to its hash without a DB round-trip per chunk.
+	blobsByID, err := repos.Blobs.GetAll(v.ctx)
+	if err != nil {
+		return nil, fmt.Errorf("fetching blob index: %w", err)
+	}
+	blobIDToHash := make(map[string]string, len(blobsByID))
+	blobByHash := make(map[string]*database.Blob, len(blobsByID))
+	for id, blob := range blobsByID {
+		hash := blob.Hash.String()
+		blobIDToHash[id] = hash
+		blobByHash[hash] = blob
+	}
+
+	plan, err := newRestorePlan(v.ctx, repos, files, chunkToBlobMap, blobIDToHash)
+	if err != nil {
+		return nil, fmt.Errorf("building restore plan: %w", err)
+	}
+
+	// Index files by ID so the loop can look them up by the IDs the
+	// plan hands back.
+	filesByID := make(map[types.FileID]*database.File, len(files))
+	for _, f := range files {
+		filesByID[f.ID] = f
+	}
+
+	// Calculate total bytes expected for percentage / ETA arithmetic.
 	var totalBytesExpected int64
 	for _, file := range files {
 		totalBytesExpected += file.Size
 	}
 
-	// Create progress bar if output is a terminal
-	bar := v.newProgressBar("Restoring", totalBytesExpected)
+	v.UI.Begin("Restoring %s files (%s) to %s.",
+		v.UI.Count(len(files)),
+		v.UI.Size(totalBytesExpected),
+		v.UI.Path(opts.TargetDir))
 
-	for i, file := range files {
+	session := &restoreSession{
+		v:              v,
+		ctx:            v.ctx,
+		repos:          repos,
+		opts:           opts,
+		identity:       identity,
+		chunkToBlobMap: chunkToBlobMap,
+		blobByHash:     blobByHash,
+		blobIDToHash:   blobIDToHash,
+		blobCache:      blobCache,
+		sweeper:        sweeper,
+		result:         result,
+	}
+
+	// Periodic progress output, matching the snapshot create cadence.
+	startTime := time.Now()
+	lastStatusTime := startTime
+	const statusInterval = 15 * time.Second
+
+	processed := 0
+	for plan.hasPending() {
 		if v.ctx.Err() != nil {
 			return nil, v.ctx.Err()
 		}
 
-		if err := v.restoreFile(v.ctx, repos, file, opts.TargetDir, identity, chunkToBlobMap, blobCache, result); err != nil {
-			log.Error("Failed to restore file", "path", file.Path, "error", err)
-			result.FilesFailed++
-			result.FailedFiles = append(result.FailedFiles, file.Path.String())
-			// Update progress bar even on failure
-			if bar != nil {
-				_ = bar.Add64(file.Size)
+		fileID, ready := plan.popReady()
+		if !ready {
+			// No file is fully cache-served. First free any blobs
+			// whose file sets are exhausted — without this, the
+			// blob whose last file we just finished would still be
+			// cached when we Put the next one, briefly pushing
+			// peak occupancy from 1 to 2.
+			sweeper.sweep()
+
+			// Pick the pending file with the smallest uncached
+			// blob set and download its blobs. After each blob
+			// lands, the plan moves any pending file whose set
+			// just emptied onto the ready queue.
+			next := plan.pickNextDownload()
+			if next.IsZero() {
+				break
+			}
+			for _, hash := range plan.blobsNeeded(next) {
+				blob, ok := blobByHash[hash]
+				if !ok {
+					return nil, fmt.Errorf("blob hash %s missing from blob index", hash[:16])
+				}
+				if err := session.downloadBlobToCache(hash, blob.CompressedSize); err != nil {
+					return nil, fmt.Errorf("downloading blob %s: %w", hash[:16], err)
+				}
+				result.BlobsDownloaded++
+				result.BytesDownloaded += blob.CompressedSize
+				plan.markBlobCached(hash)
 			}
 			continue
 		}
 
-		// Update progress bar
-		if bar != nil {
-			_ = bar.Add64(file.Size)
+		file := filesByID[fileID]
+		if err := session.restoreFile(file); err != nil {
+			log.Error("Failed to restore file", "path", file.Path, "error", err)
+			if !opts.SkipErrors {
+				return nil, fmt.Errorf("restoring %s: %w (pass --skip-errors to continue past restore failures)", file.Path, err)
+			}
+			v.UI.Error("Failed to restore %s: %v. Skipping (--skip-errors).", v.UI.Path(file.Path.String()), err)
+			result.FilesFailed++
+			result.FailedFiles = append(result.FailedFiles, file.Path.String())
+			plan.finishFile(fileID)
+			continue
 		}
 
-		// Progress logging (for non-terminal or structured logs)
-		if (i+1)%100 == 0 || i+1 == len(files) {
+		// Record the file as restored so the sweeper can free blobs
+		// once all referencing files are done, and drop it from the
+		// plan's indexes so future picks ignore it.
+		sweeper.fileRestored(fileID.String())
+		plan.finishFile(fileID)
+		processed++
+
+		if time.Since(lastStatusTime) >= statusInterval {
+			v.printRestoreProgress(processed, len(files), result.BytesRestored, totalBytesExpected, startTime)
+			lastStatusTime = time.Now()
+		}
+
+		// Structured progress log for --verbose / JSON consumers.
+		if processed%100 == 0 || processed == len(files) {
 			log.Info("Restore progress",
-				"files", fmt.Sprintf("%d/%d", i+1, len(files)),
+				"files", fmt.Sprintf("%d/%d", processed, len(files)),
 				"bytes", humanize.Bytes(uint64(result.BytesRestored)),
 			)
 		}
 	}
 
-	if bar != nil {
-		_ = bar.Finish()
+	return result, nil
+}
+
+// printRestoreProgress emits a periodic restore-phase status line via
+// the UI writer, mirroring scanner.printProcessingProgress so the two
+// long-running commands have the same on-screen rhythm.
+func (v *Vaultik) printRestoreProgress(filesDone, totalFiles int, bytesDone, totalBytes int64, startTime time.Time) {
+	elapsed := time.Since(startTime)
+	pct := float64(bytesDone) / float64(totalBytes) * 100
+	byteRate := float64(bytesDone) / elapsed.Seconds()
+	fileRate := float64(filesDone) / elapsed.Seconds()
+
+	remainingBytes := totalBytes - bytesDone
+	var eta time.Duration
+	if byteRate > 0 && remainingBytes > 0 {
+		eta = time.Duration(float64(remainingBytes)/byteRate) * time.Second
 	}
 
-	return result, nil
+	if eta > 0 {
+		v.UI.Progress("Restore: %s/%s files (%s), %s/%s, %s, %.0f files/sec, restore elapsed: %s, restore ETA: %s (est remain %s).",
+			v.UI.Count(filesDone),
+			v.UI.Count(totalFiles),
+			v.UI.Percent(pct),
+			v.UI.Size(bytesDone),
+			v.UI.Size(totalBytes),
+			v.UI.Speed(byteRate),
+			fileRate,
+			v.UI.Duration(elapsed),
+			v.UI.Time(time.Now().Add(eta)),
+			v.UI.Duration(eta))
+		return
+	}
+	v.UI.Progress("Restore: %s/%s files (%s), %s/%s, %s, %.0f files/sec, restore elapsed: %s.",
+		v.UI.Count(filesDone),
+		v.UI.Count(totalFiles),
+		v.UI.Percent(pct),
+		v.UI.Size(bytesDone),
+		v.UI.Size(totalBytes),
+		v.UI.Speed(byteRate),
+		fileRate,
+		v.UI.Duration(elapsed))
 }
 
 // handleRestoreVerification runs post-restore verification if requested
@@ -237,24 +381,26 @@ func (v *Vaultik) handleRestoreVerification(
 	}
 
 	if result.FilesFailed > 0 {
-		v.printfStdout("\nVerification FAILED: %d files did not match expected checksums\n", result.FilesFailed)
+		v.UI.Error("Verification failed: %s files did not match expected checksums.",
+			v.UI.Count(result.FilesFailed))
 		for _, path := range result.FailedFiles {
-			v.printfStdout("  - %s\n", path)
+			v.UI.Detail("%s", v.UI.Path(path))
 		}
 		return fmt.Errorf("%d files failed verification", result.FilesFailed)
 	}
 
-	v.printfStdout("Verified %d files (%s)\n",
-		result.FilesVerified,
-		humanize.Bytes(uint64(result.BytesVerified)),
-	)
+	v.UI.Complete("Verified %s files (%s).",
+		v.UI.Count(result.FilesVerified),
+		v.UI.Size(result.BytesVerified))
 	return nil
 }
 
-// downloadSnapshotDB downloads and decrypts the snapshot metadata database
+// downloadSnapshotDB downloads and decrypts the snapshot metadata
+// database. The snapshotID is the human ID; we hash it to the remote
+// key for the storage path.
 func (v *Vaultik) downloadSnapshotDB(snapshotID string, identity age.Identity) (*database.DB, error) {
 	// Download encrypted database from storage
-	dbKey := fmt.Sprintf("metadata/%s/db.zst.age", snapshotID)
+	dbKey := fmt.Sprintf("metadata/%s/db.zst.age", snapshot.RemoteSnapshotKey(snapshotID))
 
 	reader, err := v.Storage.Get(v.ctx, dbKey)
 	if err != nil {
@@ -372,209 +518,211 @@ func (v *Vaultik) buildChunkToBlobMap(ctx context.Context, repos *database.Repos
 	return result, rows.Err()
 }
 
-// restoreFile restores a single file
-func (v *Vaultik) restoreFile(
-	ctx context.Context,
-	repos *database.Repositories,
-	file *database.File,
-	targetDir string,
-	identity age.Identity,
-	chunkToBlobMap map[string]*database.BlobChunk,
-	blobCache *blobDiskCache,
-	result *RestoreResult,
-) error {
-	// Calculate target path - use full original path under target directory
-	targetPath := filepath.Join(targetDir, file.Path.String())
-
-	// Create parent directories
-	parentDir := filepath.Dir(targetPath)
-	if err := v.Fs.MkdirAll(parentDir, 0755); err != nil {
-		return fmt.Errorf("creating parent directory: %w", err)
-	}
-
-	// Handle symlinks
-	if file.IsSymlink() {
-		return v.restoreSymlink(file, targetPath, result)
-	}
-
-	// Handle directories
-	if file.Mode&uint32(os.ModeDir) != 0 {
-		return v.restoreDirectory(file, targetPath, result)
-	}
-
-	// Handle regular files
-	return v.restoreRegularFile(ctx, repos, file, targetPath, identity, chunkToBlobMap, blobCache, result)
+// restoreSession holds every piece of per-restore state shared by the
+// restore-time methods. Each restore builds one of these from the
+// snapshot's metadata and then drives the file loop through methods on
+// it. Keeping this state on the struct rather than threading it
+// through every function signature keeps the inner-loop call sites
+// readable: restoreFile(file) instead of a ten-argument helper.
+type restoreSession struct {
+	v              *Vaultik
+	ctx            context.Context
+	repos          *database.Repositories
+	opts           *RestoreOptions
+	identity       age.Identity
+	chunkToBlobMap map[string]*database.BlobChunk
+	blobByHash     map[string]*database.Blob
+	blobIDToHash   map[string]string
+	blobCache      *blobDiskCache
+	sweeper        *restoreSweeper
+	result         *RestoreResult
 }
 
-// restoreSymlink restores a symbolic link
-func (v *Vaultik) restoreSymlink(file *database.File, targetPath string, result *RestoreResult) error {
-	// Remove existing file if it exists
-	_ = v.Fs.Remove(targetPath)
+// restoreFile dispatches to the right per-kind restorer.
+func (s *restoreSession) restoreFile(file *database.File) error {
+	targetPath := filepath.Join(s.opts.TargetDir, file.Path.String())
+	parentDir := filepath.Dir(targetPath)
+	if err := s.v.Fs.MkdirAll(parentDir, 0755); err != nil {
+		return fmt.Errorf("creating parent directory: %w", err)
+	}
+	if file.IsSymlink() {
+		return s.restoreSymlink(file, targetPath)
+	}
+	if file.Mode&uint32(os.ModeDir) != 0 {
+		return s.restoreDirectory(file, targetPath)
+	}
+	return s.restoreRegularFile(file, targetPath)
+}
 
-	// Create symlink
-	// Note: afero.MemMapFs doesn't support symlinks, so we use os for real filesystems
-	if osFs, ok := v.Fs.(*afero.OsFs); ok {
-		_ = osFs // silence unused variable warning
+// restoreSymlink restores a symbolic link.
+func (s *restoreSession) restoreSymlink(file *database.File, targetPath string) error {
+	_ = s.v.Fs.Remove(targetPath)
+	// afero.MemMapFs doesn't support symlinks, so route real-FS
+	// symlinks through os.
+	if _, ok := s.v.Fs.(*afero.OsFs); ok {
 		if err := os.Symlink(file.LinkTarget.String(), targetPath); err != nil {
 			return fmt.Errorf("creating symlink: %w", err)
 		}
 	} else {
 		log.Debug("Symlink creation not supported on this filesystem", "path", file.Path, "target", file.LinkTarget)
 	}
-
-	result.FilesRestored++
+	s.result.FilesRestored++
 	log.Debug("Restored symlink", "path", file.Path, "target", file.LinkTarget)
 	return nil
 }
 
-// restoreDirectory restores a directory with proper permissions
-func (v *Vaultik) restoreDirectory(file *database.File, targetPath string, result *RestoreResult) error {
-	// Create directory
-	if err := v.Fs.MkdirAll(targetPath, os.FileMode(file.Mode)); err != nil {
+// restoreDirectory restores a directory with its permissions, mtime,
+// and (on real filesystems, with sufficient privileges) ownership.
+func (s *restoreSession) restoreDirectory(file *database.File, targetPath string) error {
+	if err := s.v.Fs.MkdirAll(targetPath, os.FileMode(file.Mode)); err != nil {
 		return fmt.Errorf("creating directory: %w", err)
 	}
-
-	// Set permissions
-	if err := v.Fs.Chmod(targetPath, os.FileMode(file.Mode)); err != nil {
+	if err := s.v.Fs.Chmod(targetPath, os.FileMode(file.Mode)); err != nil {
 		log.Debug("Failed to set directory permissions", "path", targetPath, "error", err)
 	}
-
-	// Set ownership (requires root)
-	if osFs, ok := v.Fs.(*afero.OsFs); ok {
-		_ = osFs
+	if _, ok := s.v.Fs.(*afero.OsFs); ok {
 		if err := os.Chown(targetPath, int(file.UID), int(file.GID)); err != nil {
 			log.Debug("Failed to set directory ownership", "path", targetPath, "error", err)
 		}
 	}
-
-	// Set mtime
-	if err := v.Fs.Chtimes(targetPath, file.MTime, file.MTime); err != nil {
+	if err := s.v.Fs.Chtimes(targetPath, file.MTime, file.MTime); err != nil {
 		log.Debug("Failed to set directory mtime", "path", targetPath, "error", err)
 	}
-
-	result.FilesRestored++
+	s.result.FilesRestored++
 	return nil
 }
 
-// restoreRegularFile restores a regular file by reconstructing it from chunks
-func (v *Vaultik) restoreRegularFile(
-	ctx context.Context,
-	repos *database.Repositories,
-	file *database.File,
-	targetPath string,
-	identity age.Identity,
-	chunkToBlobMap map[string]*database.BlobChunk,
-	blobCache *blobDiskCache,
-	result *RestoreResult,
-) error {
-	// Get file chunks in order
-	fileChunks, err := repos.FileChunks.GetByFileID(ctx, file.ID)
+// restoreRegularFile reconstructs a regular file by reading chunks
+// directly out of cached blobs via ReadAt. The expectation when this
+// method runs is that every blob this file needs is already in the
+// disk cache — the planner guarantees that by only marking files
+// "ready" once their full blob set is on disk.
+func (s *restoreSession) restoreRegularFile(file *database.File, targetPath string) error {
+	fileStart := time.Now()
+
+	t0 := time.Now()
+	fileChunks, err := s.repos.FileChunks.GetByFileID(s.ctx, file.ID)
+	fileChunksQueryDur := time.Since(t0)
 	if err != nil {
 		return fmt.Errorf("getting file chunks: %w", err)
 	}
 
-	// Create output file
-	outFile, err := v.Fs.Create(targetPath)
+	t0 = time.Now()
+	outFile, err := s.v.Fs.Create(targetPath)
+	createDur := time.Since(t0)
 	if err != nil {
 		return fmt.Errorf("creating output file: %w", err)
 	}
 	defer func() { _ = outFile.Close() }()
 
-	// Write chunks in order
-	var bytesWritten int64
+	var (
+		readAtDur    time.Duration
+		writeDur     time.Duration
+		sweeperDur   time.Duration
+		bytesWritten int64
+	)
+
 	for _, fc := range fileChunks {
-		// Find which blob contains this chunk
 		chunkHashStr := fc.ChunkHash.String()
-		blobChunk, ok := chunkToBlobMap[chunkHashStr]
+		blobChunk, ok := s.chunkToBlobMap[chunkHashStr]
 		if !ok {
 			return fmt.Errorf("chunk %s not found in any blob", chunkHashStr[:16])
 		}
-
-		// Get the blob's hash from the database
-		blob, err := repos.Blobs.GetByID(ctx, blobChunk.BlobID.String())
-		if err != nil {
-			return fmt.Errorf("getting blob %s: %w", blobChunk.BlobID, err)
-		}
-
-		// Download and decrypt blob if not cached
-		blobHashStr := blob.Hash.String()
-		blobData, ok := blobCache.Get(blobHashStr)
+		blobHash, ok := s.blobIDToHash[blobChunk.BlobID.String()]
 		if !ok {
-			blobData, err = v.downloadBlob(ctx, blobHashStr, blob.CompressedSize, identity)
-			if err != nil {
-				return fmt.Errorf("downloading blob %s: %w", blobHashStr[:16], err)
-			}
-			if putErr := blobCache.Put(blobHashStr, blobData); putErr != nil {
-				log.Debug("Failed to cache blob on disk", "hash", blobHashStr[:16], "error", putErr)
-			}
-			result.BlobsDownloaded++
-			result.BytesDownloaded += blob.CompressedSize
+			return fmt.Errorf("blob id %s missing from hash index", blobChunk.BlobID)
 		}
 
-		// Extract chunk from blob
-		if blobChunk.Offset+blobChunk.Length > int64(len(blobData)) {
-			return fmt.Errorf("chunk %s extends beyond blob data (offset=%d, length=%d, blob_size=%d)",
-				fc.ChunkHash[:16], blobChunk.Offset, blobChunk.Length, len(blobData))
+		t0 = time.Now()
+		chunkData, err := s.blobCache.ReadAt(blobHash, blobChunk.Offset, blobChunk.Length)
+		readAtDur += time.Since(t0)
+		if err != nil {
+			return fmt.Errorf("reading chunk %s from cached blob %s: %w", fc.ChunkHash[:16], blobHash[:16], err)
 		}
-		chunkData := blobData[blobChunk.Offset : blobChunk.Offset+blobChunk.Length]
 
-		// Write chunk to output file
+		t0 = time.Now()
 		n, err := outFile.Write(chunkData)
+		writeDur += time.Since(t0)
 		if err != nil {
 			return fmt.Errorf("writing chunk: %w", err)
 		}
 		bytesWritten += int64(n)
+
+		t0 = time.Now()
+		s.sweeper.chunkRestored(int64(n))
+		sweeperDur += time.Since(t0)
 	}
 
-	// Close file before setting metadata
+	log.Debug("Restored regular file (timings)",
+		"path", file.Path,
+		"chunks", len(fileChunks),
+		"bytes_written", bytesWritten,
+		"ms_total", time.Since(fileStart).Milliseconds(),
+		"ms_file_chunks_query", fileChunksQueryDur.Milliseconds(),
+		"ms_create", createDur.Milliseconds(),
+		"ms_readat", readAtDur.Milliseconds(),
+		"ms_writes", writeDur.Milliseconds(),
+		"ms_sweeper", sweeperDur.Milliseconds(),
+	)
+
 	if err := outFile.Close(); err != nil {
 		return fmt.Errorf("closing output file: %w", err)
 	}
-
-	// Set permissions
-	if err := v.Fs.Chmod(targetPath, os.FileMode(file.Mode)); err != nil {
+	if err := s.v.Fs.Chmod(targetPath, os.FileMode(file.Mode)); err != nil {
 		log.Debug("Failed to set file permissions", "path", targetPath, "error", err)
 	}
-
-	// Set ownership (requires root)
-	if osFs, ok := v.Fs.(*afero.OsFs); ok {
-		_ = osFs
+	if _, ok := s.v.Fs.(*afero.OsFs); ok {
 		if err := os.Chown(targetPath, int(file.UID), int(file.GID)); err != nil {
 			log.Debug("Failed to set file ownership", "path", targetPath, "error", err)
 		}
 	}
-
-	// Set mtime
-	if err := v.Fs.Chtimes(targetPath, file.MTime, file.MTime); err != nil {
+	if err := s.v.Fs.Chtimes(targetPath, file.MTime, file.MTime); err != nil {
 		log.Debug("Failed to set file mtime", "path", targetPath, "error", err)
 	}
 
-	result.FilesRestored++
-	result.BytesRestored += bytesWritten
+	s.result.FilesRestored++
+	s.result.BytesRestored += bytesWritten
 
 	log.Debug("Restored file", "path", file.Path, "size", humanize.Bytes(uint64(bytesWritten)))
 	return nil
 }
 
-// downloadBlob downloads and decrypts a blob
-func (v *Vaultik) downloadBlob(ctx context.Context, blobHash string, expectedSize int64, identity age.Identity) ([]byte, error) {
-	rc, err := v.FetchAndDecryptBlob(ctx, blobHash, expectedSize, identity)
+// downloadBlobToCache streams a blob from remote storage straight into
+// the disk cache, decrypting and decompressing on the fly. The
+// plaintext never lives fully in memory — io.Copy through
+// blobDiskCache.PutFromReader uses a 32 KiB buffer regardless of blob
+// size, which is what makes multi-GB blobs tractable on machines with
+// less RAM than the blob.
+func (s *restoreSession) downloadBlobToCache(blobHash string, expectedSize int64) error {
+	start := time.Now()
+
+	t0 := time.Now()
+	rc, err := s.v.FetchAndDecryptBlob(s.ctx, blobHash, expectedSize, s.identity)
+	fetchSetupDur := time.Since(t0)
 	if err != nil {
-		return nil, err
+		return err
 	}
 
-	data, err := io.ReadAll(rc)
-	if err != nil {
-		_ = rc.Close()
-		return nil, fmt.Errorf("reading blob data: %w", err)
+	t0 = time.Now()
+	written, copyErr := s.blobCache.PutFromReader(blobHash, rc)
+	streamDur := time.Since(t0)
+	closeErr := rc.Close()
+	if copyErr != nil {
+		return copyErr
+	}
+	if closeErr != nil {
+		return closeErr
 	}
 
-	// Close triggers hash verification
-	if err := rc.Close(); err != nil {
-		return nil, err
-	}
-
-	return data, nil
+	log.Debug("Streamed blob into disk cache",
+		"hash", blobHash[:16],
+		"compressed_bytes", expectedSize,
+		"plaintext_bytes", written,
+		"ms_total", time.Since(start).Milliseconds(),
+		"ms_fetch_setup", fetchSetupDur.Milliseconds(),
+		"ms_stream_decrypt_decompress", streamDur.Milliseconds(),
+	)
+	return nil
 }
 
 // verifyRestoredFiles verifies that all restored files match their expected chunk hashes
@@ -606,16 +754,16 @@ func (v *Vaultik) verifyRestoredFiles(
 		"files", len(regularFiles),
 		"bytes", humanize.Bytes(uint64(totalBytes)),
 	)
-	v.printfStdout("\nVerifying %d files (%s)...\n",
-		len(regularFiles),
-		humanize.Bytes(uint64(totalBytes)),
-	)
+	v.UI.Begin("Verifying %s files (%s).",
+		v.UI.Count(len(regularFiles)),
+		v.UI.Size(totalBytes))
 
-	// Create progress bar if output is a terminal
-	bar := v.newProgressBar("Verifying", totalBytes)
+	startTime := time.Now()
+	lastStatusTime := startTime
+	const statusInterval = 15 * time.Second
 
-	// Verify each file
-	for _, file := range regularFiles {
+	var bytesProcessed int64
+	for i, file := range regularFiles {
 		if ctx.Err() != nil {
 			return ctx.Err()
 		}
@@ -630,17 +778,14 @@ func (v *Vaultik) verifyRestoredFiles(
 			result.FilesVerified++
 			result.BytesVerified += bytesVerified
 		}
+		bytesProcessed += file.Size
 
-		// Update progress bar
-		if bar != nil {
-			_ = bar.Add64(file.Size)
+		if time.Since(lastStatusTime) >= statusInterval {
+			v.printVerifyProgress(i+1, len(regularFiles), bytesProcessed, totalBytes, startTime)
+			lastStatusTime = time.Now()
 		}
 	}
 
-	if bar != nil {
-		_ = bar.Finish()
-	}
-
 	log.Info("Verification complete",
 		"files_verified", result.FilesVerified,
 		"bytes_verified", humanize.Bytes(uint64(result.BytesVerified)),
@@ -650,6 +795,46 @@ func (v *Vaultik) verifyRestoredFiles(
 	return nil
 }
 
+// printVerifyProgress emits a periodic verify-phase status line. Same
+// shape as the restore progress line so user-facing pacing is uniform
+// across the two phases.
+func (v *Vaultik) printVerifyProgress(filesDone, totalFiles int, bytesDone, totalBytes int64, startTime time.Time) {
+	elapsed := time.Since(startTime)
+	pct := float64(bytesDone) / float64(totalBytes) * 100
+	byteRate := float64(bytesDone) / elapsed.Seconds()
+	fileRate := float64(filesDone) / elapsed.Seconds()
+
+	remainingBytes := totalBytes - bytesDone
+	var eta time.Duration
+	if byteRate > 0 && remainingBytes > 0 {
+		eta = time.Duration(float64(remainingBytes)/byteRate) * time.Second
+	}
+
+	if eta > 0 {
+		v.UI.Progress("Verify: %s/%s files (%s), %s/%s, %s, %.0f files/sec, verify elapsed: %s, verify ETA: %s (est remain %s).",
+			v.UI.Count(filesDone),
+			v.UI.Count(totalFiles),
+			v.UI.Percent(pct),
+			v.UI.Size(bytesDone),
+			v.UI.Size(totalBytes),
+			v.UI.Speed(byteRate),
+			fileRate,
+			v.UI.Duration(elapsed),
+			v.UI.Time(time.Now().Add(eta)),
+			v.UI.Duration(eta))
+		return
+	}
+	v.UI.Progress("Verify: %s/%s files (%s), %s/%s, %s, %.0f files/sec, verify elapsed: %s.",
+		v.UI.Count(filesDone),
+		v.UI.Count(totalFiles),
+		v.UI.Percent(pct),
+		v.UI.Size(bytesDone),
+		v.UI.Size(totalBytes),
+		v.UI.Speed(byteRate),
+		fileRate,
+		v.UI.Duration(elapsed))
+}
+
 // verifyFile verifies a single restored file by checking its chunk hashes
 func (v *Vaultik) verifyFile(
 	ctx context.Context,
@@ -705,38 +890,3 @@ func (v *Vaultik) verifyFile(
 	log.Debug("File verified", "path", file.Path, "bytes", bytesVerified, "chunks", len(fileChunks))
 	return bytesVerified, nil
 }
-
-// newProgressBar creates a terminal-aware progress bar with standard options.
-// It returns nil if stdout is not a terminal.
-func (v *Vaultik) newProgressBar(description string, total int64) *progressbar.ProgressBar {
-	if !v.isTerminal() {
-		return nil
-	}
-	return progressbar.NewOptions64(
-		total,
-		progressbar.OptionSetDescription(description),
-		progressbar.OptionSetWriter(v.Stderr),
-		progressbar.OptionShowBytes(true),
-		progressbar.OptionShowCount(),
-		progressbar.OptionSetWidth(progressBarWidth),
-		progressbar.OptionThrottle(progressBarThrottle),
-		progressbar.OptionOnCompletion(func() {
-			v.printfStderr("\n")
-		}),
-		progressbar.OptionSetRenderBlankState(true),
-	)
-}
-
-// isTerminal returns true if stdout is a terminal.
-// It checks whether v.Stdout implements Fd() (i.e. is an *os.File),
-// and falls back to false for non-file writers (e.g. in tests).
-func (v *Vaultik) isTerminal() bool {
-	type fder interface {
-		Fd() uintptr
-	}
-	f, ok := v.Stdout.(fder)
-	if !ok {
-		return false
-	}
-	return term.IsTerminal(int(f.Fd()))
-}

internal/vaultik/restore_locality_test.go

@@ -0,0 +1,315 @@
+package vaultik
+
+import (
+	"bytes"
+	"context"
+	"crypto/rand"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"sort"
+	"sync"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/config"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/ui"
+)
+
+// TestRestoreLocalityAndReadAt asserts three properties of the restore
+// hot path that together produce acceptable throughput on real-world
+// snapshots. All three currently fail on main:
+//
+//  1. Peak blob cache occupancy ≤ 1.
+//     Restore order must respect blob locality: every file fully
+//     contained within the currently cached blob should be restored
+//     before any other blob is downloaded. The sweeper then frees
+//     each blob as soon as its file set is exhausted. Without smart
+//     ordering, path-order interleaves blobs and the cache holds
+//     every touched blob until the last file referencing it lands.
+//
+//  2. Each remote blob is fetched exactly once.
+//     Counted via wrapping the Storer.
+//
+//  3. blobDiskCache.Get is never called during restore.
+//     Chunk extraction from a cached blob must go through ReadAt,
+//     which reads only the chunk's bytes from disk. Get reads the
+//     entire blob (up to 50 GB in production) into memory just to
+//     slice out a few KB — currently the dominant cost in restore.
+//
+// The test deliberately constructs an adversarial scenario: three
+// blobs A/B/C of ~6 MB each, nine files distributed across them, and
+// path-ordered names that interleave the blobs (a1, b1, c1, a2, b2,
+// c2, …) so naive path-order processing would touch every blob before
+// finishing any of them.
+func TestRestoreLocalityAndReadAt(t *testing.T) {
+	log.Initialize(log.Config{})
+
+	fs := afero.NewOsFs()
+	tempDir, err := os.MkdirTemp("", "vaultik-locality-")
+	require.NoError(t, err)
+	defer func() { _ = os.RemoveAll(tempDir) }()
+
+	dataDir := filepath.Join(tempDir, "source")
+	storeDir := filepath.Join(tempDir, "remote")
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	require.NoError(t, fs.MkdirAll(dataDir, 0o755))
+
+	// Layout: 15 source files of exactly 1 MiB each. With
+	// chunkSize (avg) = 4 MiB the chunker's minSize is 1 MiB, so any
+	// file of 1 MiB becomes a single chunk. With a 5 MiB blob limit
+	// the packer fits exactly 5 chunks per blob, producing 3 blobs
+	// containing src-001..005, src-006..010, src-011..015.
+	//
+	// Then add 9 "copy" files — byte-for-byte clones of three of the
+	// sources (one from each blob group) — with interleaved names
+	// (cp-001-A, cp-002-B, cp-003-C, cp-004-A, …) so a naive
+	// path-ordered restore would touch all three blobs before
+	// finishing any of them.
+	const (
+		srcBytes   = 1024 * 1024
+		srcCount   = 15
+		blobsCount = 3
+		perBlob    = srcCount / blobsCount
+	)
+
+	type source struct {
+		path string
+		data []byte
+	}
+	sources := make([]*source, srcCount)
+	for i := 0; i < srcCount; i++ {
+		s := &source{
+			path: fmt.Sprintf("src-%03d.bin", i+1),
+			data: randomBytes(t, srcBytes),
+		}
+		sources[i] = s
+		require.NoError(t, afero.WriteFile(fs, filepath.Join(dataDir, s.path), s.data, 0o644))
+	}
+
+	// Pick one representative source per blob group (src-001 → blob
+	// 1, src-006 → blob 2, src-011 → blob 3) and create 3 copies of
+	// each with interleaved alphabetical names.
+	type copyFile struct {
+		path        string
+		data        []byte
+		sourceBlob  int // 0, 1, or 2
+		sourceIndex int // index into sources slice
+	}
+	groupReps := []int{0, perBlob, 2 * perBlob} // 0, 5, 10
+	letters := []byte{'A', 'B', 'C'}
+	var copies []copyFile
+	for i := 0; i < 3; i++ {
+		for j := 0; j < blobsCount; j++ {
+			seq := i*blobsCount + j + 1
+			name := fmt.Sprintf("cp-%03d-%c.bin", seq, letters[j])
+			path := filepath.Join(dataDir, name)
+			src := sources[groupReps[j]]
+			require.NoError(t, afero.WriteFile(fs, path, src.data, 0o644))
+			copies = append(copies, copyFile{path: path, data: src.data, sourceBlob: j, sourceIndex: groupReps[j]})
+		}
+	}
+
+	// chunkSize avg = 4 MiB makes minSize = 1 MiB, so a 1 MiB file
+	// becomes one chunk. maxBlobSize = 5 MiB packs exactly 5 chunks
+	// per blob, yielding 3 blobs from 15 source files.
+	chunkSize := int64(4 * 1024 * 1024)
+	maxBlobSize := int64(5 * 1024 * 1024)
+
+	storer, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	agePublicKey := "age1ezrjmfpwsc95svdg0y54mums3zevgzu0x0ecq2f7tp8a05gl0sjq9q9wjg"
+	ageSecretKey := "AGE-SECRET-KEY-19CR5YSFW59HM4TLD6GXVEDMZFTVVF7PPHKUT68TXSFPK7APHXA2QS2NJA5"
+
+	cfg := &config.Config{
+		AgeRecipients:    []string{agePublicKey},
+		AgeSecretKey:     ageSecretKey,
+		CompressionLevel: 3,
+		Hostname:         "test-host",
+		BlobSizeLimit:    config.Size(maxBlobSize),
+	}
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+
+	repos := database.NewRepositories(db)
+
+	sm := snapshot.NewSnapshotManager(snapshot.SnapshotManagerParams{
+		Repos:   repos,
+		Storage: storer,
+		Config:  cfg,
+	})
+	sm.SetFilesystem(fs)
+
+	scanner := snapshot.NewScanner(snapshot.ScannerConfig{
+		FS:               fs,
+		Storage:          storer,
+		ChunkSize:        chunkSize,
+		MaxBlobSize:      maxBlobSize,
+		CompressionLevel: cfg.CompressionLevel,
+		AgeRecipients:    cfg.AgeRecipients,
+		Repositories:     repos,
+	})
+
+	snapshotID, err := sm.CreateSnapshotWithName(ctx, cfg.Hostname, "locality", "test-version", "test-git")
+	require.NoError(t, err)
+
+	_, err = scanner.Scan(ctx, dataDir, snapshotID)
+	require.NoError(t, err)
+
+	require.NoError(t, sm.CompleteSnapshot(ctx, snapshotID))
+	require.NoError(t, sm.ExportSnapshotMetadata(ctx, dbPath, snapshotID))
+
+	blobsOnDisk := listBlobKeys(t, storeDir)
+	t.Logf("backup produced %d blobs", len(blobsOnDisk))
+	require.GreaterOrEqual(t, len(blobsOnDisk), 3, "expected at least 3 blobs from 3 filler groups")
+
+	require.NoError(t, db.Close())
+
+	// Wrap the storer so we can count downloads per blob key.
+	counter := newCountingStorer(storer)
+
+	// Capture the restore-side cache for instrumentation inspection.
+	// The observer fires twice (immediately after creation and
+	// immediately before close) so we read PeakLen and call counters
+	// from the same instance the production code used.
+	var cacheRef *blobDiskCache
+	v := &Vaultik{
+		Config:  cfg,
+		Storage: counter,
+		Fs:      fs,
+		Stdout:  io.Discard,
+		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
+		restoreCacheObserver: func(c *blobDiskCache) {
+			cacheRef = c
+		},
+	}
+	v.SetContext(ctx)
+
+	require.NoError(t, v.Restore(&RestoreOptions{
+		SnapshotID: snapshotID,
+		TargetDir:  restoreDir,
+	}))
+
+	require.NotNil(t, cacheRef, "restoreCacheObserver must fire during restore")
+
+	// Verify restored content matches.
+	for _, s := range sources {
+		restored := filepath.Join(restoreDir, dataDir, s.path)
+		got, err := afero.ReadFile(fs, restored)
+		require.NoErrorf(t, err, "source missing after restore: %s", s.path)
+		require.Truef(t, bytes.Equal(got, s.data), "byte mismatch for source %s", s.path)
+	}
+	for _, c := range copies {
+		restored := filepath.Join(restoreDir, c.path)
+		got, err := afero.ReadFile(fs, restored)
+		require.NoErrorf(t, err, "copy missing after restore: %s", c.path)
+		require.Truef(t, bytes.Equal(got, c.data), "byte mismatch for copy %s", c.path)
+	}
+
+	// (1) Each blob fetched exactly once.
+	for key, n := range counter.snapshot() {
+		if !filterBlobKey(key) {
+			continue
+		}
+		assert.Equalf(t, 1, n, "blob %s fetched %d times, want exactly 1", key, n)
+	}
+
+	// (2) Peak cache size ≤ 1. The sweeper plus locality-aware
+	// ordering should free each blob before the next one downloads.
+	assert.LessOrEqualf(t, cacheRef.PeakLen(), 1,
+		"peak cached blobs was %d; expected ≤ 1 with locality-ordered restore", cacheRef.PeakLen())
+
+	// (3) Cache.Get must never be called during restore — chunk
+	// extraction has to go through ReadAt so we never read the whole
+	// blob from disk to grab a few KB slice.
+	assert.Equalf(t, 0, cacheRef.GetCalls(),
+		"blobDiskCache.Get was called %d times during restore; restore must use ReadAt exclusively", cacheRef.GetCalls())
+
+	t.Logf("blob cache stats: peak_len=%d get_calls=%d readat_calls=%d",
+		cacheRef.PeakLen(), cacheRef.GetCalls(), cacheRef.ReadAtCalls())
+}
+
+// randomBytes returns n bytes of random data. Used to make sure the
+// chunker picks non-degenerate FastCDC boundaries.
+func randomBytes(t *testing.T, n int) []byte {
+	t.Helper()
+	b := make([]byte, n)
+	_, err := rand.Read(b)
+	require.NoError(t, err)
+	return b
+}
+
+// listBlobKeys walks the FileStorer blobs/ tree and returns the
+// relative keys for every blob file present.
+func listBlobKeys(t *testing.T, storeDir string) []string {
+	t.Helper()
+	var keys []string
+	root := filepath.Join(storeDir, "blobs")
+	err := filepath.Walk(root, func(p string, info os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+		if info.IsDir() {
+			return nil
+		}
+		rel, _ := filepath.Rel(storeDir, p)
+		keys = append(keys, rel)
+		return nil
+	})
+	require.NoError(t, err)
+	sort.Strings(keys)
+	return keys
+}
+
+// filterBlobKey returns true when key looks like a blob storage path
+// (rather than a snapshot metadata path).
+func filterBlobKey(key string) bool {
+	return len(key) > 6 && key[:6] == "blobs/"
+}
+
+// countingStorerInternal wraps a storage.Storer and records the number
+// of Get calls per key, so the locality test can assert each blob is
+// fetched exactly once. Defined here (rather than reusing the one in
+// the integration_test package) because this test lives in package
+// vaultik for access to unexported cache internals.
+type countingStorerInternal struct {
+	storage.Storer
+	mu     sync.Mutex
+	counts map[string]int
+}
+
+func newCountingStorer(inner storage.Storer) *countingStorerInternal {
+	return &countingStorerInternal{Storer: inner, counts: make(map[string]int)}
+}
+
+func (c *countingStorerInternal) Get(ctx context.Context, key string) (io.ReadCloser, error) {
+	c.mu.Lock()
+	c.counts[key]++
+	c.mu.Unlock()
+	return c.Storer.Get(ctx, key)
+}
+
+func (c *countingStorerInternal) snapshot() map[string]int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	out := make(map[string]int, len(c.counts))
+	for k, v := range c.counts {
+		out[k] = v
+	}
+	return out
+}

internal/vaultik/restore_plan.go

@@ -0,0 +1,185 @@
+package vaultik
+
+import (
+	"context"
+	"fmt"
+	"math"
+	"os"
+
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/types"
+)
+
+// restorePlan orders restore-time file processing by blob locality. The
+// goal is to keep the blob disk cache occupancy as small as possible:
+// download one blob, drain every file referencing only that blob, let
+// the sweeper free the blob, then move on. Files that span multiple
+// blobs are processed when their full blob set is on disk.
+//
+// The plan keeps two indexes:
+//
+//   - fileBlobs: for each pending file, the set of blob hashes it
+//     still needs that are NOT yet in the cache. Files with an empty
+//     set are "ready" — they can be restored from the current cache
+//     with no further downloads.
+//   - blobFiles: for each blob, the set of pending files referencing
+//     it. Used to short-circuit "when this blob lands, which files
+//     become ready" without a global scan.
+type restorePlan struct {
+	fileBlobs map[types.FileID]map[string]struct{}
+	blobFiles map[string]map[types.FileID]struct{}
+	ready     []types.FileID
+	cached    map[string]struct{}
+}
+
+// newRestorePlan builds the file→blob index for the given files. Files
+// whose chunks reference no blobs (symlinks, directories) start in the
+// ready queue immediately.
+func newRestorePlan(
+	ctx context.Context,
+	repos *database.Repositories,
+	files []*database.File,
+	chunkToBlobMap map[string]*database.BlobChunk,
+	blobIDToHash map[string]string,
+) (*restorePlan, error) {
+	p := &restorePlan{
+		fileBlobs: make(map[types.FileID]map[string]struct{}, len(files)),
+		blobFiles: make(map[string]map[types.FileID]struct{}),
+		ready:     make([]types.FileID, 0, len(files)),
+		cached:    make(map[string]struct{}),
+	}
+	for _, f := range files {
+		if f.IsSymlink() || f.Mode&uint32(os.ModeDir) != 0 {
+			// No chunks to fetch — restore can run immediately.
+			p.fileBlobs[f.ID] = nil
+			p.ready = append(p.ready, f.ID)
+			continue
+		}
+		fileChunks, err := repos.FileChunks.GetByFileID(ctx, f.ID)
+		if err != nil {
+			return nil, fmt.Errorf("planning %s: %w", f.Path, err)
+		}
+		blobs := make(map[string]struct{})
+		for _, fc := range fileChunks {
+			bc, ok := chunkToBlobMap[fc.ChunkHash.String()]
+			if !ok {
+				return nil, fmt.Errorf("planning %s: chunk %s missing from blob map",
+					f.Path, fc.ChunkHash.String()[:16])
+			}
+			hash, ok := blobIDToHash[bc.BlobID.String()]
+			if !ok {
+				return nil, fmt.Errorf("planning %s: blob id %s missing from id-to-hash map",
+					f.Path, bc.BlobID)
+			}
+			blobs[hash] = struct{}{}
+		}
+		p.fileBlobs[f.ID] = blobs
+		for hash := range blobs {
+			set, ok := p.blobFiles[hash]
+			if !ok {
+				set = make(map[types.FileID]struct{})
+				p.blobFiles[hash] = set
+			}
+			set[f.ID] = struct{}{}
+		}
+		if len(blobs) == 0 {
+			p.ready = append(p.ready, f.ID)
+		}
+	}
+	return p, nil
+}
+
+// markBlobCached records that the named blob is now resident in the
+// disk cache and moves any pending file whose remaining-uncached-blobs
+// set just dropped to empty onto the ready queue.
+func (p *restorePlan) markBlobCached(blobHash string) {
+	if _, already := p.cached[blobHash]; already {
+		return
+	}
+	p.cached[blobHash] = struct{}{}
+	for fileID := range p.blobFiles[blobHash] {
+		blobs := p.fileBlobs[fileID]
+		delete(blobs, blobHash)
+		if len(blobs) == 0 {
+			p.ready = append(p.ready, fileID)
+		}
+	}
+}
+
+// popReady returns the next ready file, removing it from the queue. If
+// no file is ready, the second return value is false.
+func (p *restorePlan) popReady() (types.FileID, bool) {
+	if len(p.ready) == 0 {
+		return types.FileID{}, false
+	}
+	id := p.ready[0]
+	p.ready = p.ready[1:]
+	return id, true
+}
+
+// finishFile drops a restored file from both indexes so subsequent
+// planning calls don't reconsider it.
+func (p *restorePlan) finishFile(fileID types.FileID) {
+	for hash := range p.fileBlobs[fileID] {
+		if set, ok := p.blobFiles[hash]; ok {
+			delete(set, fileID)
+			if len(set) == 0 {
+				delete(p.blobFiles, hash)
+			}
+		}
+	}
+	delete(p.fileBlobs, fileID)
+	// Also scrub the file from any blobFiles entries where it might
+	// still appear even after its uncached-blob set was emptied.
+	for hash, set := range p.blobFiles {
+		if _, ok := set[fileID]; ok {
+			delete(set, fileID)
+			if len(set) == 0 {
+				delete(p.blobFiles, hash)
+			}
+		}
+	}
+}
+
+// pickNextDownload returns the pending file whose remaining-uncached
+// blob set is smallest (with ties broken by FileID string compare so
+// the choice is deterministic across runs). This file's blobs are
+// downloaded next, after which it — together with any other pending
+// files whose blob sets become empty — moves to the ready queue.
+//
+// The zero FileID return means nothing is pending.
+func (p *restorePlan) pickNextDownload() types.FileID {
+	var best types.FileID
+	bestCount := math.MaxInt
+	var bestID string
+	for id, blobs := range p.fileBlobs {
+		n := len(blobs)
+		if n == 0 {
+			// Already-ready files should have been popped via
+			// popReady; ignore here just in case.
+			continue
+		}
+		idStr := id.String()
+		if n < bestCount || (n == bestCount && (best.IsZero() || idStr < bestID)) {
+			best = id
+			bestCount = n
+			bestID = idStr
+		}
+	}
+	return best
+}
+
+// blobsNeeded returns the uncached blob hashes for fileID in any order.
+func (p *restorePlan) blobsNeeded(fileID types.FileID) []string {
+	blobs := p.fileBlobs[fileID]
+	out := make([]string, 0, len(blobs))
+	for h := range blobs {
+		out = append(out, h)
+	}
+	return out
+}
+
+// hasPending reports whether any unfinished files remain.
+func (p *restorePlan) hasPending() bool {
+	return len(p.fileBlobs) > 0
+}

internal/vaultik/restore_sweeper.go

@@ -0,0 +1,118 @@
+package vaultik
+
+import (
+	"context"
+	"fmt"
+
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+)
+
+// restoreSweeper frees cached blobs once all files that reference any of
+// their chunks have been restored. It works as follows:
+//
+//  1. Callers add a file's ID to an in-memory restored set via
+//     fileRestored once the file is fully written to disk.
+//  2. After each chunk is restored, chunkRestored accumulates a running
+//     byte count.
+//  3. When the accumulator crosses a threshold (one hundredth of the
+//     configured blob size — so a sweep runs about a hundred times per
+//     blob's worth of restored bytes), the sweeper iterates every key in
+//     the cache. For each cached blob it asks the DB which files
+//     reference any chunk in that blob, then compares that list against
+//     the in-memory restored set. If any referencing file is missing
+//     from the set the blob is kept; otherwise the cache entry is
+//     deleted.
+//
+// All DB reads happen against the snapshot's temporary metadata DB,
+// which is local, indexed, and not under contention — the queries are
+// cheap and run at most once per blob per sweep interval.
+type restoreSweeper struct {
+	ctx        context.Context
+	repos      *database.Repositories
+	cache      *blobDiskCache
+	threshold  int64
+	bytesAccum int64
+	restored   map[string]struct{}
+}
+
+// newRestoreSweeper returns a sweeper that triggers eviction every
+// `threshold` bytes restored. Callers should pass blob_size_limit/100.
+func newRestoreSweeper(ctx context.Context, repos *database.Repositories, cache *blobDiskCache, threshold int64) *restoreSweeper {
+	if threshold <= 0 {
+		threshold = 1
+	}
+	return &restoreSweeper{
+		ctx:       ctx,
+		repos:     repos,
+		cache:     cache,
+		threshold: threshold,
+		restored:  make(map[string]struct{}),
+	}
+}
+
+// fileRestored records a file as fully restored. After this call, any
+// blob whose only remaining references come from files in the restored
+// set may be evicted on the next sweep.
+func (s *restoreSweeper) fileRestored(fileID string) {
+	s.restored[fileID] = struct{}{}
+}
+
+// chunkRestored accounts n bytes against the sweep threshold and runs a
+// sweep if the threshold has been crossed since the last sweep.
+func (s *restoreSweeper) chunkRestored(n int64) {
+	s.bytesAccum += n
+	if s.bytesAccum < s.threshold {
+		return
+	}
+	s.bytesAccum = 0
+	s.sweep()
+}
+
+// sweep deletes any cached blob whose chunks are no longer referenced
+// by an unrestored file. Per-blob DB failures are logged and the blob
+// is kept — we'd rather hold a blob longer than risk a re-download.
+func (s *restoreSweeper) sweep() {
+	for _, blobHash := range s.cache.Keys() {
+		needed, err := s.blobStillNeeded(blobHash)
+		if err != nil {
+			log.Debug("sweeper referencing-files query failed", "blob_hash", blobHash[:16], "error", err)
+			continue
+		}
+		if !needed {
+			s.cache.Delete(blobHash)
+		}
+	}
+}
+
+// blobStillNeeded returns true if any file that references a chunk in
+// this blob has not yet been restored. On any error the function
+// returns true — keeping the blob is always the safe answer because we
+// can't prove we're done with it.
+func (s *restoreSweeper) blobStillNeeded(blobHash string) (bool, error) {
+	rows, err := s.repos.DB().Conn().QueryContext(s.ctx, `
+		SELECT DISTINCT fc.file_id
+		FROM file_chunks fc
+		JOIN blob_chunks bc ON bc.chunk_hash = fc.chunk_hash
+		JOIN blobs b ON b.id = bc.blob_id
+		WHERE b.blob_hash = ?
+	`, blobHash)
+	if err != nil {
+		return true, fmt.Errorf("querying referencing files: %w", err)
+	}
+	defer func() { _ = rows.Close() }()
+
+	for rows.Next() {
+		var fileID string
+		if err := rows.Scan(&fileID); err != nil {
+			return true, fmt.Errorf("scanning file_id: %w", err)
+		}
+		if _, ok := s.restored[fileID]; !ok {
+			return true, nil
+		}
+	}
+	if err := rows.Err(); err != nil {
+		return true, err
+	}
+	return false, nil
+}

internal/vaultik/restore_sweeper_integration_test.go

@@ -0,0 +1,248 @@
+package vaultik_test
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"math/rand"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/config"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/ui"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestRestoreSweeperEvictsBlobs exercises the reference-counted blob
+// disk cache eviction during restore.
+//
+// The scenario: 30 unique 1 MB random files plus 10 duplicates of those
+// (40 files total, 30 MB of unique content) get backed up with a 10 MB
+// blob_size_limit. After backup the snapshot's encrypted blobs are
+// restored through Vaultik.Restore, and per-key Get counts on the
+// storage layer are recorded. Each blob in the snapshot MUST be
+// downloaded exactly once — re-downloads would mean the sweeper either
+// evicted a blob that was still needed (LRU regression) or that the
+// cache held nothing at all (broken cache).
+//
+// The duplicates ensure deduplicated files share blobs with their
+// originals; the sweeper must keep each blob alive until BOTH the
+// original AND every duplicate referencing its chunks have been
+// restored.
+func TestRestoreSweeperEvictsBlobs(t *testing.T) {
+	log.Initialize(log.Config{})
+
+	fs := afero.NewOsFs()
+	tempDir, err := os.MkdirTemp("", "vaultik-sweeper-")
+	require.NoError(t, err)
+	defer func() { _ = os.RemoveAll(tempDir) }()
+
+	dataDir := filepath.Join(tempDir, "source")
+	storeDir := filepath.Join(tempDir, "remote")
+	restoreDir := filepath.Join(tempDir, "restored")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	require.NoError(t, fs.MkdirAll(dataDir, 0o755))
+
+	// Generate 30 unique 1 MB random files. The PRNG seed is fixed so
+	// failures are reproducible; the entropy is what matters here — the
+	// FastCDC chunker needs realistic-looking data to pick chunk
+	// boundaries naturally.
+	const (
+		uniqueFiles    = 30
+		duplicateFiles = 10
+		fileSize       = 1 * 1024 * 1024
+	)
+	rng := rand.New(rand.NewSource(42))
+
+	type sourceFile struct {
+		path string
+		data []byte
+	}
+	uniques := make([]sourceFile, 0, uniqueFiles)
+	expected := make(map[string][]byte, uniqueFiles+duplicateFiles)
+
+	for i := 0; i < uniqueFiles; i++ {
+		data := make([]byte, fileSize)
+		_, err := rng.Read(data)
+		require.NoError(t, err)
+		path := filepath.Join(dataDir, fmt.Sprintf("unique-%02d.bin", i))
+		require.NoError(t, afero.WriteFile(fs, path, data, 0o644))
+		uniques = append(uniques, sourceFile{path: path, data: data})
+		expected[path] = data
+	}
+
+	// Pick 10 of the originals and copy each to a fresh path so the
+	// chunker dedups them against the originals' blobs.
+	for i, idx := range rng.Perm(uniqueFiles)[:duplicateFiles] {
+		src := uniques[idx]
+		dstPath := filepath.Join(dataDir, fmt.Sprintf("dup-%02d.bin", i))
+		require.NoError(t, afero.WriteFile(fs, dstPath, src.data, 0o644))
+		expected[dstPath] = src.data
+	}
+
+	chunkSize := int64(64 * 1024)
+	maxBlobSize := int64(10 * 1024 * 1024)
+
+	storer, err := storage.NewFileStorer(storeDir)
+	require.NoError(t, err)
+
+	agePublicKey := "age1ezrjmfpwsc95svdg0y54mums3zevgzu0x0ecq2f7tp8a05gl0sjq9q9wjg"
+	ageSecretKey := "AGE-SECRET-KEY-19CR5YSFW59HM4TLD6GXVEDMZFTVVF7PPHKUT68TXSFPK7APHXA2QS2NJA5"
+
+	cfg := &config.Config{
+		AgeRecipients:    []string{agePublicKey},
+		AgeSecretKey:     ageSecretKey,
+		CompressionLevel: 3,
+		Hostname:         "test-host",
+		BlobSizeLimit:    config.Size(maxBlobSize),
+	}
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+	defer func() { _ = db.Close() }()
+
+	repos := database.NewRepositories(db)
+
+	sm := snapshot.NewSnapshotManager(snapshot.SnapshotManagerParams{
+		Repos:   repos,
+		Storage: storer,
+		Config:  cfg,
+	})
+	sm.SetFilesystem(fs)
+
+	scanner := snapshot.NewScanner(snapshot.ScannerConfig{
+		FS:               fs,
+		Storage:          storer,
+		ChunkSize:        chunkSize,
+		MaxBlobSize:      maxBlobSize,
+		CompressionLevel: cfg.CompressionLevel,
+		AgeRecipients:    cfg.AgeRecipients,
+		Repositories:     repos,
+	})
+
+	snapshotID, err := sm.CreateSnapshotWithName(ctx, cfg.Hostname, "sweeper", "test-version", "test-git")
+	require.NoError(t, err)
+
+	scanResult, err := scanner.Scan(ctx, dataDir, snapshotID)
+	require.NoError(t, err)
+	require.Equal(t, uniqueFiles+duplicateFiles, scanResult.FilesScanned)
+	require.Greater(t, scanResult.BlobsCreated, 1, "30 MB of unique data at 10 MB blob size should yield multiple blobs")
+
+	require.NoError(t, sm.CompleteSnapshot(ctx, snapshotID))
+	require.NoError(t, sm.ExportSnapshotMetadata(ctx, dbPath, snapshotID))
+
+	// Count blobs actually present on disk; this is the ground-truth
+	// figure each blob's GET count must equal exactly once.
+	blobCount := countBlobsOnDisk(t, storeDir)
+	require.Greater(t, blobCount, 1, "expected more than one blob")
+	t.Logf("backup produced %d blobs from %d files (%d unique + %d duplicates)",
+		blobCount, uniqueFiles+duplicateFiles, uniqueFiles, duplicateFiles)
+
+	// Force restore to operate without the source-side index, exactly
+	// as a real restore on a fresh machine would.
+	require.NoError(t, db.Close())
+
+	counter := newCountingStorer(storer)
+
+	restoreVaultik := &vaultik.Vaultik{
+		Config:  cfg,
+		Storage: counter,
+		Fs:      fs,
+		Stdout:  io.Discard,
+		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
+	}
+	restoreVaultik.SetContext(ctx)
+
+	require.NoError(t, restoreVaultik.Restore(&vaultik.RestoreOptions{
+		SnapshotID: snapshotID,
+		TargetDir:  restoreDir,
+	}))
+
+	// Verify every restored file byte-matches its source.
+	for origPath, want := range expected {
+		restoredPath := filepath.Join(restoreDir, origPath)
+		got, err := afero.ReadFile(fs, restoredPath)
+		require.NoErrorf(t, err, "restored file missing: %s", restoredPath)
+		require.Equalf(t, want, got, "byte mismatch for %s", origPath)
+	}
+
+	// Each blob must have been downloaded exactly once. >1 means the
+	// sweeper evicted a still-needed blob; 0 means the cache silently
+	// stopped being consulted.
+	blobDownloads := 0
+	for key, count := range counter.snapshot() {
+		if !strings.HasPrefix(key, "blobs/") {
+			continue
+		}
+		assert.Equalf(t, 1, count,
+			"blob %s should have been downloaded exactly once during restore, got %d", key, count)
+		blobDownloads++
+	}
+	assert.Equal(t, blobCount, blobDownloads,
+		"every blob on disk should have been fetched exactly once during restore")
+	t.Logf("restore downloaded %d blobs, each exactly once", blobDownloads)
+}
+
+// countingStorer wraps a Storer and records the number of Get calls per
+// key. Used to verify that the restore-side blob cache + sweeper avoid
+// re-downloading blobs that are evicted while still needed.
+type countingStorer struct {
+	storage.Storer
+	mu     sync.Mutex
+	counts map[string]int
+}
+
+func newCountingStorer(inner storage.Storer) *countingStorer {
+	return &countingStorer{Storer: inner, counts: make(map[string]int)}
+}
+
+func (c *countingStorer) Get(ctx context.Context, key string) (io.ReadCloser, error) {
+	c.mu.Lock()
+	c.counts[key]++
+	c.mu.Unlock()
+	return c.Storer.Get(ctx, key)
+}
+
+func (c *countingStorer) snapshot() map[string]int {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	out := make(map[string]int, len(c.counts))
+	for k, v := range c.counts {
+		out[k] = v
+	}
+	return out
+}
+
+// countBlobsOnDisk walks the blobs/ tree of a FileStorer-backed store
+// and returns the total number of blob files. Used to ground-truth the
+// expected number of restore-time downloads.
+func countBlobsOnDisk(t *testing.T, storeDir string) int {
+	t.Helper()
+	count := 0
+	root := filepath.Join(storeDir, "blobs")
+	err := filepath.Walk(root, func(_ string, info os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+		if !info.IsDir() {
+			count++
+		}
+		return nil
+	})
+	require.NoError(t, err)
+	return count
+}

internal/vaultik/snapshot.go

@@ -8,16 +8,13 @@ import (
 	"regexp"
 	"sort"
 	"strings"
-	"sync"
 	"text/tabwriter"
 	"time"
 
 	"github.com/dustin/go-humanize"
-	"golang.org/x/sync/errgroup"
 	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/log"
 	"sneak.berlin/go/vaultik/internal/snapshot"
-	"sneak.berlin/go/vaultik/internal/types"
 )
 
 // SnapshotCreateOptions contains options for the snapshot create command
@@ -92,8 +89,13 @@ func (v *Vaultik) CreateSnapshot(opts *SnapshotCreateOptions) error {
 		}
 	}
 
+	// Terminus must obey the --cron invariant: silent on total
+	// success only. UI.Complete is dropped in cron/quiet mode (that's
+	// the success path), but if any warnings fired during the run we
+	// emit the summary via UI.Warning so cron actually delivers
+	// something for the user to look at.
 	if v.UI.WarningCount() > 0 {
-		v.UI.Complete("Finished (with %d warnings).", v.UI.WarningCount())
+		v.UI.Warning("Finished with %d warning(s) — review the output above.", v.UI.WarningCount())
 	} else {
 		v.UI.Complete("Finished successfully.")
 	}
@@ -306,23 +308,13 @@ func (v *Vaultik) finalizeSnapshotMetadata(snapshotID string, stats *snapshotSta
 	return nil
 }
 
-// formatUploadSpeed formats bytes uploaded and duration into a human-readable speed string
-func formatUploadSpeed(bytesUploaded int64, duration time.Duration) string {
+// uploadSpeed returns the average network upload rate as a colorized
+// bits/sec string, or "N/A" when there's no usable data.
+func (v *Vaultik) uploadSpeed(bytesUploaded int64, duration time.Duration) string {
 	if bytesUploaded <= 0 || duration <= 0 {
-		return "N/A"
-	}
-	bytesPerSec := float64(bytesUploaded) / duration.Seconds()
-	bitsPerSec := bytesPerSec * 8
-	switch {
-	case bitsPerSec >= 1e9:
-		return fmt.Sprintf("%.1f Gbit/s", bitsPerSec/1e9)
-	case bitsPerSec >= 1e6:
-		return fmt.Sprintf("%.0f Mbit/s", bitsPerSec/1e6)
-	case bitsPerSec >= 1e3:
-		return fmt.Sprintf("%.0f Kbit/s", bitsPerSec/1e3)
-	default:
-		return fmt.Sprintf("%.0f bit/s", bitsPerSec)
+		return v.UI.Speed(0)
 	}
+	return v.UI.Speed(float64(bytesUploaded) / duration.Seconds())
 }
 
 // printSnapshotSummary prints the comprehensive snapshot completion summary
@@ -342,7 +334,7 @@ func (v *Vaultik) printSnapshotSummary(snapshotID string, startTime time.Time, s
 	}
 
 	v.UI.Complete("Created snapshot %s.", v.UI.Snapshot(snapshotID))
-	filesMsg := fmt.Sprintf("Files: %s examined, %s processed, %s unchanged",
+	filesMsg := fmt.Sprintf("Files: %s examined, %s backed up, %s unchanged",
 		v.UI.Count(stats.totalFiles),
 		v.UI.Count(totalFilesChanged),
 		v.UI.Count(stats.totalFilesSkipped))
@@ -351,7 +343,7 @@ func (v *Vaultik) printSnapshotSummary(snapshotID string, startTime time.Time, s
 	}
 	v.UI.Detail("%s.", filesMsg)
 
-	dataMsg := fmt.Sprintf("Data: %s total (%s processed)",
+	dataMsg := fmt.Sprintf("Data: %s total (%s backed up)",
 		v.UI.Size(totalBytesAll),
 		v.UI.Size(stats.totalBytes))
 	if stats.totalBytesDeleted > 0 {
@@ -368,7 +360,7 @@ func (v *Vaultik) printSnapshotSummary(snapshotID string, startTime time.Time, s
 			stats.totalBlobsUploaded,
 			v.UI.Size(stats.totalBytesUploaded),
 			v.UI.Duration(stats.uploadDuration),
-			formatUploadSpeed(stats.totalBytesUploaded, stats.uploadDuration))
+			v.uploadSpeed(stats.totalBytesUploaded, stats.uploadDuration))
 	}
 	v.UI.Detail("Snapshot create duration: %s.", v.UI.Duration(snapshotDuration))
 }
@@ -388,25 +380,43 @@ func (v *Vaultik) getSnapshotBlobSizes(snapshotID string) (compressed int64, unc
 	return compressed, uncompressed
 }
 
-// ListSnapshots lists all snapshots
+// ListSnapshots prints the table of snapshots, plus any reconciliation
+// warnings/notes between the local index and the backup destination
+// store.
+//
+// The local index database is always the primary source for the
+// table — it has the human snapshot IDs, timestamps, and per-snapshot
+// stats.
+//
+// If an age secret key is configured AND remote listing succeeds, we
+// cross-reference: any local snapshot whose hashed key isn't visible
+// remotely gets a "local-only" cleanup hint, and any remote key that
+// doesn't correspond to a known local snapshot gets reported in a
+// NOTE.
+//
+// If no age key is set the local machine is assumed write-only
+// (backup-only), so we skip remote listing entirely — there's no
+// value showing keys the user couldn't restore anyway.
+//
+// If remote listing fails (unmounted volume, permission denied,
+// network), we degrade to local-only with a warning. List never
+// fails just because the destination is unreachable.
 func (v *Vaultik) ListSnapshots(jsonOutput bool) error {
 	log.Info("Listing snapshots")
-	remoteSnapshots, err := v.listRemoteSnapshotIDs()
+
+	localSnaps, err := v.Repositories.Snapshots.ListRecent(v.ctx, 10000)
 	if err != nil {
-		return err
+		return fmt.Errorf("listing local snapshots: %w", err)
 	}
 
-	localSnapshotMap, err := v.reconcileLocalWithRemote(remoteSnapshots)
-	if err != nil {
-		return err
+	snapshots := make([]SnapshotInfo, 0, len(localSnaps))
+	for _, ls := range localSnaps {
+		if ls.CompletedAt == nil {
+			continue
+		}
+		snapshots = append(snapshots, v.snapshotInfoFromLocal(ls))
 	}
 
-	snapshots, err := v.buildSnapshotInfoList(remoteSnapshots, localSnapshotMap)
-	if err != nil {
-		return err
-	}
-
-	// Sort by timestamp (newest first)
 	sort.Slice(snapshots, func(i, j int) bool {
 		return snapshots[i].Timestamp.After(snapshots[j].Timestamp)
 	})
@@ -421,159 +431,85 @@ func (v *Vaultik) ListSnapshots(jsonOutput bool) error {
 		return err
 	}
 
-	// Warn about local snapshots that don't exist in remote storage.
-	var stale []string
-	for id := range localSnapshotMap {
-		if !remoteSnapshots[id] {
-			stale = append(stale, id)
+	if v.Config.AgeSecretKey == "" {
+		return nil
+	}
+
+	remoteKeys, err := v.listAllRemoteSnapshotKeys()
+	if err != nil {
+		v.UI.Warning("Could not list backup destination store: %v.", err)
+		return nil
+	}
+
+	localKeys := make(map[string]string, len(localSnaps))
+	for _, ls := range localSnaps {
+		if ls.CompletedAt == nil {
+			continue
+		}
+		localKeys[snapshot.RemoteSnapshotKey(ls.ID.String())] = ls.ID.String()
+	}
+	remoteSet := make(map[string]bool, len(remoteKeys))
+	for _, k := range remoteKeys {
+		remoteSet[k] = true
+	}
+
+	var localOnly []string
+	for key, humanID := range localKeys {
+		if !remoteSet[key] {
+			localOnly = append(localOnly, humanID)
 		}
 	}
-	if len(stale) > 0 {
-		v.UI.Warning("%d local snapshot record(s) not found in backup destination store:", len(stale))
-		for _, id := range stale {
+	var remoteOnlyCount int
+	for key := range remoteSet {
+		if _, ok := localKeys[key]; !ok {
+			remoteOnlyCount++
+		}
+	}
+
+	if len(localOnly) > 0 {
+		v.UI.Warning("%d local snapshot record(s) not found in backup destination store:", len(localOnly))
+		for _, id := range localOnly {
 			v.UI.Info("%s", v.UI.Snapshot(id))
 		}
 		v.UI.Info("Run 'vaultik snapshot cleanup' to remove stale local records.")
 	}
+	if remoteOnlyCount > 0 {
+		v.UI.Notice("NOTE: %d remote snapshot(s) found in backup destination store but not in local database.", remoteOnlyCount)
+	}
 
 	return nil
 }
 
-// listRemoteSnapshotIDs returns a set of snapshot IDs found in remote storage
-func (v *Vaultik) listRemoteSnapshotIDs() (map[string]bool, error) {
-	remoteSnapshots := make(map[string]bool)
-	objectCh := v.Storage.ListStream(v.ctx, "metadata/")
+// snapshotInfoFromLocal builds a SnapshotInfo row from a local snapshot
+// record. Failures from any per-snapshot stat query degrade that
+// column to its snapshot-row fallback but never fail the listing.
+func (v *Vaultik) snapshotInfoFromLocal(ls *database.Snapshot) SnapshotInfo {
+	idStr := ls.ID.String()
 
-	for object := range objectCh {
-		if object.Err != nil {
-			return nil, fmt.Errorf("listing remote snapshots: %w", object.Err)
-		}
-
-		parts := strings.Split(object.Key, "/")
-		if len(parts) >= 2 && parts[0] == "metadata" && parts[1] != "" {
-			if strings.HasPrefix(parts[1], ".") {
-				continue
-			}
-			remoteSnapshots[parts[1]] = true
-		}
-	}
-
-	return remoteSnapshots, nil
-}
-
-// reconcileLocalWithRemote builds a map of local snapshots keyed by ID for cross-referencing with remote
-func (v *Vaultik) reconcileLocalWithRemote(remoteSnapshots map[string]bool) (map[string]*database.Snapshot, error) {
-	localSnapshots, err := v.Repositories.Snapshots.ListRecent(v.ctx, 10000)
+	totalSize, err := v.Repositories.Snapshots.GetSnapshotTotalCompressedSize(v.ctx, idStr)
 	if err != nil {
-		return nil, fmt.Errorf("listing local snapshots: %w", err)
+		log.Warn("Failed to get total compressed size", "id", idStr, "error", err)
+		totalSize = ls.BlobSize
 	}
 
-	localSnapshotMap := make(map[string]*database.Snapshot)
-	for _, s := range localSnapshots {
-		localSnapshotMap[s.ID.String()] = s
+	uncompressedSize, err := v.Repositories.Snapshots.GetSnapshotUncompressedChunkSize(v.ctx, idStr)
+	if err != nil {
+		log.Warn("Failed to get uncompressed chunk size", "id", idStr, "error", err)
 	}
 
-	return localSnapshotMap, nil
-}
-
-// buildSnapshotInfoList constructs SnapshotInfo entries from remote IDs and local data
-func (v *Vaultik) buildSnapshotInfoList(remoteSnapshots map[string]bool, localSnapshotMap map[string]*database.Snapshot) ([]SnapshotInfo, error) {
-	snapshots := make([]SnapshotInfo, 0, len(remoteSnapshots))
-
-	// remoteOnly collects snapshot IDs that need a manifest download.
-	var remoteOnly []string
-
-	for snapshotID := range remoteSnapshots {
-		if localSnap, exists := localSnapshotMap[snapshotID]; exists && localSnap.CompletedAt != nil {
-			totalSize, err := v.Repositories.Snapshots.GetSnapshotTotalCompressedSize(v.ctx, snapshotID)
-			if err != nil {
-				log.Warn("Failed to get total compressed size", "id", snapshotID, "error", err)
-				totalSize = localSnap.BlobSize
-			}
-
-			snapshots = append(snapshots, SnapshotInfo{
-				ID:             localSnap.ID,
-				Timestamp:      localSnap.StartedAt,
-				CompressedSize: totalSize,
-			})
-		} else {
-			timestamp, err := parseSnapshotTimestamp(snapshotID)
-			if err != nil {
-				log.Warn("Failed to parse snapshot timestamp", "id", snapshotID, "error", err)
-				continue
-			}
-
-			// Pre-add with zero size; will be filled by concurrent downloads.
-			snapshots = append(snapshots, SnapshotInfo{
-				ID:             types.SnapshotID(snapshotID),
-				Timestamp:      timestamp,
-				CompressedSize: 0,
-			})
-			remoteOnly = append(remoteOnly, snapshotID)
-		}
+	newChunkSize, err := v.Repositories.Snapshots.GetSnapshotNewChunkSize(v.ctx, idStr)
+	if err != nil {
+		log.Warn("Failed to get new chunk size", "id", idStr, "error", err)
 	}
 
-	// Download manifests concurrently for remote-only snapshots.
-	if len(remoteOnly) > 0 {
-		// maxConcurrentManifestDownloads bounds parallel manifest fetches to
-		// avoid overwhelming the S3 endpoint while still being much faster
-		// than serial downloads.
-		const maxConcurrentManifestDownloads = 10
-
-		type manifestResult struct {
-			snapshotID string
-			size       int64
-		}
-
-		var (
-			mu      sync.Mutex
-			results []manifestResult
-		)
-
-		g, gctx := errgroup.WithContext(v.ctx)
-		g.SetLimit(maxConcurrentManifestDownloads)
-
-		for _, sid := range remoteOnly {
-			g.Go(func() error {
-				manifestPath := fmt.Sprintf("metadata/%s/manifest.json.zst", sid)
-				reader, err := v.Storage.Get(gctx, manifestPath)
-				if err != nil {
-					return fmt.Errorf("downloading manifest for %s: %w", sid, err)
-				}
-				defer func() { _ = reader.Close() }()
-
-				manifest, err := snapshot.DecodeManifest(reader)
-				if err != nil {
-					return fmt.Errorf("decoding manifest for %s: %w", sid, err)
-				}
-
-				mu.Lock()
-				results = append(results, manifestResult{
-					snapshotID: sid,
-					size:       manifest.TotalCompressedSize,
-				})
-				mu.Unlock()
-				return nil
-			})
-		}
-
-		if err := g.Wait(); err != nil {
-			return nil, fmt.Errorf("fetching manifest sizes: %w", err)
-		}
-
-		// Build a lookup from results and patch the pre-added entries.
-		sizeMap := make(map[string]int64, len(results))
-		for _, r := range results {
-			sizeMap[r.snapshotID] = r.size
-		}
-		for i := range snapshots {
-			if sz, ok := sizeMap[string(snapshots[i].ID)]; ok {
-				snapshots[i].CompressedSize = sz
-			}
-		}
+	return SnapshotInfo{
+		ID:               ls.ID,
+		Timestamp:        ls.StartedAt,
+		CompressedSize:   totalSize,
+		UncompressedSize: uncompressedSize,
+		NewChunkSize:     newChunkSize,
+		LocallyTracked:   true,
 	}
-
-	return snapshots, nil
 }
 
 // printSnapshotTable renders the snapshot list as a formatted table
@@ -603,18 +539,27 @@ func (v *Vaultik) printSnapshotTable(snapshots []SnapshotInfo) error {
 	if _, err := fmt.Fprintln(w, "REMOTE SNAPSHOTS:"); err != nil {
 		return err
 	}
-	if _, err := fmt.Fprintln(w, "SNAPSHOT ID\tTIMESTAMP\tCOMPRESSED SIZE"); err != nil {
+	if _, err := fmt.Fprintln(w, "SNAPSHOT ID\tTIMESTAMP\tCOMPRESSED SIZE\tUNCOMPRESSED SIZE\tNEW CHUNK SIZE"); err != nil {
 		return err
 	}
-	if _, err := fmt.Fprintln(w, "───────────\t─────────\t───────────────"); err != nil {
+	if _, err := fmt.Fprintln(w, "───────────\t─────────\t───────────────\t─────────────────\t──────────────"); err != nil {
 		return err
 	}
 
+	const remoteOnlyCell = "<remote only>"
 	for _, snap := range snapshots {
-		if _, err := fmt.Fprintf(w, "%s\t%s\t%s\n",
+		uncompressed := remoteOnlyCell
+		newChunks := remoteOnlyCell
+		if snap.LocallyTracked {
+			uncompressed = formatBytes(snap.UncompressedSize)
+			newChunks = formatBytes(snap.NewChunkSize)
+		}
+		if _, err := fmt.Fprintf(w, "%s\t%s\t%s\t%s\t%s\n",
 			snap.ID,
 			snap.Timestamp.Format("2006-01-02 15:04:05"),
-			formatBytes(snap.CompressedSize)); err != nil {
+			formatBytes(snap.CompressedSize),
+			uncompressed,
+			newChunks); err != nil {
 			return err
 		}
 	}
@@ -750,14 +695,23 @@ func (v *Vaultik) confirmAndExecutePurge(toDelete []SnapshotInfo, force, quiet b
 		if err := v.deleteSnapshotFromLocalDB(snapshotID); err != nil {
 			log.Error("Failed to delete from local database", "snapshot_id", snapshotID, "error", err)
 		}
-		if err := v.deleteSnapshotFromRemote(snapshotID); err != nil {
+		if err := v.deleteRemoteSnapshotByKey(snapshot.RemoteSnapshotKey(snapshotID)); err != nil {
 			return fmt.Errorf("deleting snapshot %s from remote: %w", snapshotID, err)
 		}
 	}
 
+	// Tidy up local DB orphans now so users don't have to run a
+	// separate command after a purge. Guarded against nil for tests
+	// that don't wire up a SnapshotManager.
+	if v.SnapshotManager != nil {
+		if err := v.SnapshotManager.CleanupOrphanedData(v.ctx); err != nil {
+			log.Warn("Failed to clean up orphaned local data after purge", "error", err)
+		}
+	}
+
 	if !quiet {
 		v.printfStdout("Deleted %d snapshot(s)\n", len(toDelete))
-		v.printlnStdout("\nNote: Run 'vaultik prune' to clean up unreferenced blobs.")
+		v.printlnStdout("\nNote: Run 'vaultik prune' to clean up unreferenced remote blobs.")
 	}
 
 	return nil
@@ -786,8 +740,9 @@ func (v *Vaultik) VerifySnapshotWithOptions(snapshotID string, opts *VerifyOptio
 
 	v.printVerifyHeader(snapshotID, opts)
 
-	// Download and parse manifest
-	manifest, err := v.downloadManifest(snapshotID)
+	// Download and parse manifest. The caller supplies a human
+	// snapshot ID; we hash it to address remote storage.
+	manifest, err := v.downloadManifestByKey(snapshot.RemoteSnapshotKey(snapshotID))
 	if err != nil {
 		if opts.JSON {
 			result.Status = "failed"
@@ -902,12 +857,18 @@ func (v *Vaultik) outputVerifyJSON(result *VerifyResult) error {
 
 // CleanupLocalSnapshots removes local snapshot records that have no
 // corresponding metadata in remote storage. These are typically left
-// behind by incomplete or interrupted backups.
+// behind by incomplete or interrupted backups. Each local snapshot's
+// human ID is hashed via RemoteSnapshotKey and compared against the
+// remote listing.
 func (v *Vaultik) CleanupLocalSnapshots() error {
-	remoteSnapshots, err := v.listRemoteSnapshotIDs()
+	remoteKeys, err := v.listAllRemoteSnapshotKeys()
 	if err != nil {
 		return err
 	}
+	remoteSet := make(map[string]bool, len(remoteKeys))
+	for _, k := range remoteKeys {
+		remoteSet[k] = true
+	}
 
 	localSnapshots, err := v.Repositories.Snapshots.ListRecent(v.ctx, 10000)
 	if err != nil {
@@ -917,7 +878,7 @@ func (v *Vaultik) CleanupLocalSnapshots() error {
 	var removed int
 	for _, snap := range localSnapshots {
 		id := snap.ID.String()
-		if !remoteSnapshots[id] {
+		if !remoteSet[snapshot.RemoteSnapshotKey(id)] {
 			v.printfStdout("Removing stale local record: %s\n", id)
 			if err := v.deleteSnapshotFromLocalDB(id); err != nil {
 				log.Error("Failed to delete local snapshot", "snapshot_id", id, "error", err)
@@ -937,8 +898,12 @@ func (v *Vaultik) CleanupLocalSnapshots() error {
 
 // Helper methods that were previously on SnapshotApp
 
-func (v *Vaultik) downloadManifest(snapshotID string) (*snapshot.Manifest, error) {
-	manifestPath := fmt.Sprintf("metadata/%s/manifest.json.zst", snapshotID)
+// downloadManifestByKey fetches the manifest at
+// metadata/<remoteKey>/manifest.json.zst. The remoteKey is the double-
+// SHA256 derivation produced by snapshot.RemoteSnapshotKey, not the
+// human snapshot ID. Callers that have a human ID must hash first.
+func (v *Vaultik) downloadManifestByKey(remoteKey string) (*snapshot.Manifest, error) {
+	manifestPath := fmt.Sprintf("metadata/%s/manifest.json.zst", remoteKey)
 
 	reader, err := v.Storage.Get(v.ctx, manifestPath)
 	if err != nil {
@@ -1073,12 +1038,22 @@ func (v *Vaultik) RemoveSnapshot(snapshotID string, opts *RemoveOptions) (*Remov
 	// If --remote, also remove from remote storage
 	if opts.Remote {
 		log.Info("Removing snapshot metadata from remote storage", "snapshot_id", snapshotID)
-		if err := v.deleteSnapshotFromRemote(snapshotID); err != nil {
+		if err := v.deleteRemoteSnapshotByKey(snapshot.RemoteSnapshotKey(snapshotID)); err != nil {
 			return result, fmt.Errorf("removing from remote storage: %w", err)
 		}
 		result.RemoteRemoved = true
 	}
 
+	// Clean up the local rows that just became orphaned (files, chunks,
+	// blob_chunks, blobs no longer referenced by any snapshot). This
+	// used to be a separate `vaultik snapshot prune` step; running it
+	// inline means `snapshot remove` leaves no ghost rows behind.
+	if v.SnapshotManager != nil {
+		if err := v.SnapshotManager.CleanupOrphanedData(v.ctx); err != nil {
+			log.Warn("Failed to clean up orphaned local data after removal", "error", err)
+		}
+	}
+
 	// Output result
 	if opts.JSON {
 		return result, v.outputRemoveJSON(result)
@@ -1088,20 +1063,48 @@ func (v *Vaultik) RemoveSnapshot(snapshotID string, opts *RemoveOptions) (*Remov
 	v.printfStdout("Removed snapshot '%s' from local database\n", snapshotID)
 	if opts.Remote {
 		v.printlnStdout("Removed snapshot metadata from remote storage")
-		v.printlnStdout("\nNote: Blobs were not removed. Run 'vaultik prune' to remove orphaned blobs.")
+		v.printlnStdout("\nNote: Remote blobs were not removed. Run 'vaultik prune' to remove orphaned blobs.")
 	}
 
 	return result, nil
 }
 
-// RemoveAllSnapshots removes all snapshots from local database and optionally from remote
+// RemoveAllSnapshots removes every snapshot known to the local
+// database from the local index, and (with --remote) every snapshot
+// metadata directory in remote storage. Both sides are processed so a
+// "remove --all" leaves nothing behind, even when the local DB and
+// remote storage have diverged.
 func (v *Vaultik) RemoveAllSnapshots(opts *RemoveOptions) (*RemoveResult, error) {
-	snapshotIDs, err := v.listAllRemoteSnapshotIDs()
+	localSnaps, err := v.localSnapshotIDs()
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("listing local snapshots: %w", err)
 	}
 
-	if len(snapshotIDs) == 0 {
+	// remoteKeys is the set of metadata/<key>/ subdirectories on the
+	// destination store; failures are downgraded to a warning so a
+	// permission-denied or unreachable remote can't block a local-only
+	// remove.
+	remoteKeys, remoteErr := v.listAllRemoteSnapshotKeys()
+	if remoteErr != nil {
+		log.Warn("Could not list remote snapshots", "error", remoteErr)
+		v.UI.Warning("Could not list remote snapshots: %v.", remoteErr)
+	}
+
+	// Anything visible on the remote that doesn't correspond to a
+	// known local human ID is treated as an orphan key — handled only
+	// when --remote is in effect.
+	knownLocalKeys := make(map[string]string, len(localSnaps))
+	for _, id := range localSnaps {
+		knownLocalKeys[snapshot.RemoteSnapshotKey(id)] = id
+	}
+	var orphanRemoteKeys []string
+	for _, key := range remoteKeys {
+		if _, known := knownLocalKeys[key]; !known {
+			orphanRemoteKeys = append(orphanRemoteKeys, key)
+		}
+	}
+
+	if len(localSnaps) == 0 && len(orphanRemoteKeys) == 0 {
 		if !opts.JSON {
 			v.printlnStdout("No snapshots found")
 		}
@@ -1109,19 +1112,42 @@ func (v *Vaultik) RemoveAllSnapshots(opts *RemoveOptions) (*RemoveResult, error)
 	}
 
 	if opts.DryRun {
-		return v.handleRemoveAllDryRun(snapshotIDs, opts)
+		return v.handleRemoveAllDryRun(localSnaps, orphanRemoteKeys, opts)
 	}
 
-	return v.executeRemoveAll(snapshotIDs, opts)
+	return v.executeRemoveAll(localSnaps, orphanRemoteKeys, opts)
 }
 
-// listAllRemoteSnapshotIDs collects all unique snapshot IDs from remote storage
-func (v *Vaultik) listAllRemoteSnapshotIDs() ([]string, error) {
-	log.Info("Listing all snapshots")
+// localSnapshotIDs returns every snapshot ID present in the local
+// index database, sorted for deterministic iteration. Empty slice if
+// the database has no Repositories (e.g. tests).
+func (v *Vaultik) localSnapshotIDs() ([]string, error) {
+	if v.Repositories == nil {
+		return nil, nil
+	}
+	snaps, err := v.Repositories.Snapshots.ListRecent(v.ctx, 100000)
+	if err != nil {
+		return nil, err
+	}
+	ids := make([]string, 0, len(snaps))
+	for _, s := range snaps {
+		ids = append(ids, s.ID.String())
+	}
+	sort.Strings(ids)
+	return ids, nil
+}
+
+// listAllRemoteSnapshotKeys collects the hashed remote keys
+// (subdirectories under metadata/) currently present in the
+// destination store. Returns (nil, err) when the store cannot be
+// listed; callers must treat that as "no remote info available," not
+// fatal.
+func (v *Vaultik) listAllRemoteSnapshotKeys() ([]string, error) {
+	log.Info("Listing all remote snapshots")
 	objectCh := v.Storage.ListStream(v.ctx, "metadata/")
 
 	seen := make(map[string]bool)
-	var snapshotIDs []string
+	var keys []string
 	for object := range objectCh {
 		if object.Err != nil {
 			return nil, fmt.Errorf("listing remote snapshots: %w", object.Err)
@@ -1134,30 +1160,36 @@ func (v *Vaultik) listAllRemoteSnapshotIDs() ([]string, error) {
 				continue
 			}
 			if strings.HasSuffix(object.Key, "/") || strings.Contains(object.Key, "/manifest.json.zst") {
-				sid := parts[1]
-				if !seen[sid] {
-					seen[sid] = true
-					snapshotIDs = append(snapshotIDs, sid)
+				key := parts[1]
+				if !seen[key] {
+					seen[key] = true
+					keys = append(keys, key)
 				}
 			}
 		}
 	}
 
-	return snapshotIDs, nil
+	return keys, nil
 }
 
 // handleRemoveAllDryRun handles the dry-run mode for removing all snapshots
-func (v *Vaultik) handleRemoveAllDryRun(snapshotIDs []string, opts *RemoveOptions) (*RemoveResult, error) {
-	result := &RemoveResult{
-		DryRun:           true,
-		SnapshotsRemoved: snapshotIDs,
+func (v *Vaultik) handleRemoveAllDryRun(localSnaps, orphanRemoteKeys []string, opts *RemoveOptions) (*RemoveResult, error) {
+	result := &RemoveResult{DryRun: true}
+	result.SnapshotsRemoved = append(result.SnapshotsRemoved, localSnaps...)
+	if opts.Remote {
+		result.SnapshotsRemoved = append(result.SnapshotsRemoved, orphanRemoteKeys...)
 	}
 	if !opts.JSON {
-		v.printfStdout("Would remove %d snapshot(s):\n", len(snapshotIDs))
-		for _, id := range snapshotIDs {
+		v.printfStdout("Would remove %d local snapshot(s):\n", len(localSnaps))
+		for _, id := range localSnaps {
 			v.printfStdout("  %s\n", id)
 		}
-		if opts.Remote {
+		if opts.Remote && len(orphanRemoteKeys) > 0 {
+			v.printfStdout("Would also remove %d orphan remote snapshot key(s):\n", len(orphanRemoteKeys))
+			for _, key := range orphanRemoteKeys {
+				v.printfStdout("  %s\n", key)
+			}
+		} else if opts.Remote {
 			v.printlnStdout("Would also remove from remote storage")
 		}
 		v.printlnStdout("[Dry run - no changes made]")
@@ -1168,17 +1200,19 @@ func (v *Vaultik) handleRemoveAllDryRun(snapshotIDs []string, opts *RemoveOption
 	return result, nil
 }
 
-// executeRemoveAll removes all snapshots from local database and optionally from remote storage
-func (v *Vaultik) executeRemoveAll(snapshotIDs []string, opts *RemoveOptions) (*RemoveResult, error) {
+// executeRemoveAll deletes every local snapshot (and, with --remote,
+// every corresponding remote metadata directory plus any orphan remote
+// keys that don't match a local snapshot).
+func (v *Vaultik) executeRemoveAll(localSnaps, orphanRemoteKeys []string, opts *RemoveOptions) (*RemoveResult, error) {
 	// --all requires --force
 	if !opts.Force {
 		return nil, fmt.Errorf("--all requires --force")
 	}
 
-	log.Info("Removing all snapshots", "count", len(snapshotIDs))
+	log.Info("Removing all snapshots", "local_count", len(localSnaps), "orphan_remote_count", len(orphanRemoteKeys))
 
 	result := &RemoveResult{}
-	for _, snapshotID := range snapshotIDs {
+	for _, snapshotID := range localSnaps {
 		log.Info("Removing snapshot", "snapshot_id", snapshotID)
 
 		if err := v.deleteSnapshotFromLocalDB(snapshotID); err != nil {
@@ -1187,7 +1221,7 @@ func (v *Vaultik) executeRemoveAll(snapshotIDs []string, opts *RemoveOptions) (*
 		}
 
 		if opts.Remote {
-			if err := v.deleteSnapshotFromRemote(snapshotID); err != nil {
+			if err := v.deleteRemoteSnapshotByKey(snapshot.RemoteSnapshotKey(snapshotID)); err != nil {
 				log.Error("Failed to remove from remote", "snapshot_id", snapshotID, "error", err)
 				continue
 			}
@@ -1196,10 +1230,29 @@ func (v *Vaultik) executeRemoveAll(snapshotIDs []string, opts *RemoveOptions) (*
 		result.SnapshotsRemoved = append(result.SnapshotsRemoved, snapshotID)
 	}
 
+	if opts.Remote {
+		for _, key := range orphanRemoteKeys {
+			log.Info("Removing orphan remote snapshot", "remote_key", key)
+			if err := v.deleteRemoteSnapshotByKey(key); err != nil {
+				log.Error("Failed to remove orphan from remote", "remote_key", key, "error", err)
+				continue
+			}
+			result.SnapshotsRemoved = append(result.SnapshotsRemoved, key)
+		}
+	}
+
 	if opts.Remote {
 		result.RemoteRemoved = true
 	}
 
+	// Clean up everything that just became orphaned locally so the
+	// index database doesn't carry 39k ghost rows after a wipe.
+	if v.SnapshotManager != nil {
+		if err := v.SnapshotManager.CleanupOrphanedData(v.ctx); err != nil {
+			log.Warn("Failed to clean up orphaned local data after bulk removal", "error", err)
+		}
+	}
+
 	if opts.JSON {
 		return result, v.outputRemoveJSON(result)
 	}
@@ -1207,7 +1260,7 @@ func (v *Vaultik) executeRemoveAll(snapshotIDs []string, opts *RemoveOptions) (*
 	v.printfStdout("Removed %d snapshot(s)\n", len(result.SnapshotsRemoved))
 	if opts.Remote {
 		v.printlnStdout("Removed snapshot metadata from remote storage")
-		v.printlnStdout("\nNote: Blobs were not removed. Run 'vaultik prune' to remove orphaned blobs.")
+		v.printlnStdout("\nNote: Remote blobs were not removed. Run 'vaultik prune' to remove orphaned blobs.")
 	}
 
 	return result, nil
@@ -1236,9 +1289,13 @@ func (v *Vaultik) deleteSnapshotFromLocalDB(snapshotID string) error {
 	return nil
 }
 
-// deleteSnapshotFromRemote removes snapshot metadata files from remote storage
-func (v *Vaultik) deleteSnapshotFromRemote(snapshotID string) error {
-	prefix := fmt.Sprintf("metadata/%s/", snapshotID)
+// deleteRemoteSnapshotByKey removes everything under
+// metadata/<remoteKey>/ on the destination store. The argument is a
+// remote key (double-SHA256 derivation), not a human snapshot ID;
+// callers that have a human ID must hash via snapshot.RemoteSnapshotKey
+// first.
+func (v *Vaultik) deleteRemoteSnapshotByKey(remoteKey string) error {
+	prefix := fmt.Sprintf("metadata/%s/", remoteKey)
 	objectCh := v.Storage.ListStream(v.ctx, prefix)
 
 	var objectsToDelete []string

internal/vaultik/vaultik.go

@@ -44,6 +44,13 @@ type Vaultik struct {
 	// writer wrapping Stdout; the cli layer replaces it with a discarding
 	// writer in --cron mode.
 	UI *ui.Writer
+
+	// restoreCacheObserver, if non-nil, is invoked once with the
+	// restore-side blob disk cache immediately after the cache is
+	// created and again immediately before it is closed. Only
+	// internal-package tests set this; the type is unexported so
+	// callers outside this package can't reach it.
+	restoreCacheObserver func(*blobDiskCache)
 }
 
 // VaultikParams contains all parameters for New that can be provided by fx
@@ -147,11 +154,6 @@ func (v *Vaultik) printlnStdout(args ...any) {
 	_, _ = fmt.Fprintln(v.Stdout, args...)
 }
 
-// printfStderr writes formatted output to stderr.
-func (v *Vaultik) printfStderr(format string, args ...any) {
-	_, _ = fmt.Fprintf(v.Stderr, format, args...)
-}
-
 // scanStdin reads a line of input from stdin.
 func (v *Vaultik) scanStdin(a ...any) (int, error) {
 	return fmt.Fscanln(v.Stdin, a...)

internal/vaultik/verify.go

@@ -106,8 +106,11 @@ func (v *Vaultik) RunDeepVerify(snapshotID string, opts *VerifyOptions) error {
 
 // loadVerificationData downloads manifest, database, and blob list for verification
 func (v *Vaultik) loadVerificationData(snapshotID string, opts *VerifyOptions, result *VerifyResult) (*snapshot.Manifest, *tempDB, []snapshot.BlobInfo, error) {
+	// All remote paths use the hashed key derived from the human ID.
+	remoteKey := snapshot.RemoteSnapshotKey(snapshotID)
+
 	// Download manifest
-	manifestPath := fmt.Sprintf("metadata/%s/manifest.json.zst", snapshotID)
+	manifestPath := fmt.Sprintf("metadata/%s/manifest.json.zst", remoteKey)
 	log.Info("Downloading manifest", "path", manifestPath)
 	if !opts.JSON {
 		v.printfStdout("Downloading manifest...\n")
@@ -136,7 +139,7 @@ func (v *Vaultik) loadVerificationData(snapshotID string, opts *VerifyOptions, r
 	}
 
 	// Download and decrypt database
-	dbPath := fmt.Sprintf("metadata/%s/db.zst.age", snapshotID)
+	dbPath := fmt.Sprintf("metadata/%s/db.zst.age", remoteKey)
 	log.Info("Downloading encrypted database", "path", dbPath)
 	dbReader, err := v.Storage.Get(v.ctx, dbPath)
 	if err != nil {