Most backup tools — free or paid — write your images and wish you luck. OPBS closes the loop: it actually restores your backups on a schedule and checks the result, repairs silent bit-rot before it becomes a failed restore, and keeps an off-box witness that catches ransomware rolling your history back. None of this is a paid tier.
Each layer catches what the one before it can't. Together they turn "the backup finished" into "the restore is known to work."
OPBS periodically takes your newest backup, restores it
onto a scratch disk you dedicate (its data gets overwritten),
then reads the filesystems back from the physical disk
and validates them: boot-sector 0x55AA signature, OEM id,
NTFS $MFT record 0 FILE magic and in-use
flag, FAT sector/cluster geometry. A restore that completes but
fails validation is a failed drill — surfaced, not swallowed.
Encrypted images are included when you configure the passphrase.
OPBS.exe --cli drill --dir D:\OPBS --disk 2 --verify
Each drill records a per-image timestamp
(opbs-drill-history.json); the Dashboard shows a
"drill-tested" badge. Failures alert only after N consecutive misses
(default 2) — no toast spam, but never a silent pass on a broken
chain. Drills always run with delta application and
pre-write verification enabled.
Hard drives corrupt themselves quietly. Every OPBS backup gets an
XOR-parity sidecar (.opar) — PAR2-style groups of 31
blocks over the stored frame bytes. A scheduled or idle scrub
re-reads every block, re-checks frame and raw CRCs, and
rebuilds any single corrupted block in place
from its parity group. It works on encrypted images too — no
passphrase needed to verify or repair.
OPBS.exe --cli scrub --dir D:\OPBS --repair
Scheduled under Settings → Scheduled Scrub, with a Dashboard
"Scrub Health" panel and per-image history in
opbs-scrub.json. Rot is fixed on the next scrub —
not discovered at restore time.
An OPBS repository keeps backups append-only behind an HMAC-SHA256 chained journal — the signing key never lives inside the repository. Images are locked for N days; pruning only ever touches expired ones, and unlocking is an audited, signed event. But a tamper-evident journal alone can't catch an attacker who rolls the whole thing back. So every repository backup can write an off-box anchor — a snapshot of the exact journal bytes to a local directory, S3 or SFTP that the attacker on the backup machine cannot also rewrite.
OPBS.exe --cli repo init D:\BackupRepo --lock-days 30 OPBS.exe --cli repo anchor D:\BackupRepo --to s3://my-witness/backups OPBS.exe --cli repo verify D:\BackupRepo --anchor s3://my-witness/backups
repo verify --anchor byte-compares the live journal
against the witness: a truncated journal (rollback), rewritten
records, a swapped header or a missing anchor all fail with
anchor-mismatch/anchor-missing. Pair it
with an S3 bucket under Object Lock and the images themselves are
WORM-protected in the cloud.
"The backup finished" is a status.
"The restore was tested last night" is a fact.
Drills make it a fact. Scrub keeps the data restorable. Anchors keep the history honest.
The three headline layers sit on a base that's standard in OPBS — also free.
Every written block is CRC-checked, and scheduled verification
re-reads images end to end, stamping a FLAG_VERIFIED
the Dashboard shows per image.
A manifest of the backup directory is rewritten after every backup and prune; drift detection flags missing delta bases, rewritten files and chain breaks before they bite.
A 0–100 reliability score per destination combining free space, chain integrity, verification, scrub and drill coverage, and SMART health — so weak drives are flagged before they fail you.
Run the free fleet server; each machine posts backup health, drill results and staleness. One green/amber/red row per machine — and webhook alerts the moment something turns critical.
repo init a repository directory, set an anchor target (S3/SFTP/local-off-box), and every repository backup re-anchors itself.fleet serve on any one machine and point the others' check-ins at it — the "did my fleet back up last night?" dashboard.