OpenScrub v1.0.19

OpenScrub v1.0.19

• Camera footage fixed — handheld/phone video was processed with
screen-scroll assumptions, causing blur smears at the frame edges and
stray blur before a face appeared. OpenScrub now detects camera
footage automatically, anchors detections to the frame, and tracks
moving faces frame-by-frame. Screen recordings behave exactly as
before (--scroll-track on/off to override).

• Ignore zones — draw "never blur" regions in the zone editor (watermark,
timestamp, your own webcam bubble). Ignore zones and detection zones
can't overlap: each clips against the other's edges while drawing.
They apply to every job; the old ignore-regions text field is gone.

OpenScrub v1.0.18

OpenScrub v1.0.18

• Fixed: a GPU/cuDNN failure could crash a job at the first OCR scan
(seen on the CUDA Docker image as CUDNN error 5003). The OCR engine
now self-tests at startup and falls back GPU → CPU → Tesseract, so
jobs always complete; the CUDA image's cuDNN mismatch is also fixed.
• Jobs can now be deleted from the web UI — a trash icon on each job
removes it and all of its files (with confirmation).

OpenScrub v1.0.17

OpenScrub v1.0.17

• The default MRN pattern is now generic — a standalone 6–10 digit
number (optional short letter prefix), flagged only when near an
MRN/chart/account label or 7+ digits long. Previously the default
encoded one specific EMR's numbering format. If your site has a known
MRN format, set it in the MRN regex field for best precision.

• First release with the NVIDIA GPU Docker image: CUDA-accelerated OCR
and NVENC encoding — ghcr.io/austinmabry/openscrub:cuda (Unraid-ready,
see the README's Docker section).

OpenScrub v1.0.16

OpenScrub v1.0.16

• Docker — every release now publishes a ready-to-run server image:
docker run -d -p 8384:8384
-v openscrub_data:/root/.local/share/OpenScrub
ghcr.io/austinmabry/openscrub:latest
Tesseract, FFmpeg, and the face model are included. CPU-only; your
data lives in the volume, so update by pulling the new tag.

• Detection Zones editor — the video preview now has a full timeline:
click or drag to scrub anywhere in the recording, with a hover time
indicator and current/duration readouts, so you can find the exact
layout you want to trace zones over.

OpenScrub v1.0.15

OpenScrub v1.0.15

• Verified gap bridging — when something disappears from detection and
comes back, OpenScrub no longer trusts a timer to decide whether to
keep it blurred in between. It checks the video itself: if the
content was really on screen the whole time, the blur bridges the
gap no matter how long; if the content truly went away, it never
bridges, no matter how short. The --bridge-gap setting still exists
but you shouldn't need to touch it anymore.

• OCR sanity check — if the first scan reads almost no text from a
clearly text-dense recording, the job log warns you immediately
(with the fix), instead of letting a broken OCR setup ruin a long
render silently.

• Optional vault auto-lock — encryption users can now have the vault
re-lock itself after a chosen number of idle minutes (no activity,
no running jobs). Off by default; enable it in App settings →
Encryption at rest.

Together with v1.0.14, the scanning engine now self-tunes its pacing,
finds true first frames however far back they are, and verifies gaps
against the pixels — fewer knobs, fewer excuses.

OpenScrub v1.0.14

OpenScrub v1.0.14

This release removes the two biggest "you should have tuned a knob"
excuses from the scanning engine.

• Deep backtrack — when something (a face, a name, a plate) turns out
to have been on screen longer than the backtrack buffer reaches,
OpenScrub no longer asks you to raise a setting. It now goes back to
the video file itself and searches — as far back as it needs to —
until it finds the true first frame, and starts the blur there.
Coverage of an object's first appearance no longer depends on any
window size.

• Self-tuning scan pacing — the scan interval now adapts to your
footage: it relaxes while the screen is static, tightens when a lot
is changing, and scans sooner during fast scrolling. Fixed pacing is
still available with --adaptive off. All safety guarantees are
unchanged: content that scrolls in unscanned is covered by safety
bands until it has been read.

Both are on by default. No settings to learn, nothing to configure.

Full test suite green (16 end-to-end tests, including a new one proving
the deep backtrack finds an onset beyond the buffer without
over-extending before the object existed).

OpenScrub v1.0.13

header now uses the real brand wordmark; typing the address without https:// automatically redirects to the secure URL.

OpenScrub v1.0.12

Homepage decluttered — server-level configuration (plate models, optional engines, encryption, HTTPS certificate) now lives in an App settings view behind the gear icon on the Job Settings card.

OpenScrub v1.0.11

Create your own regex detection categories from the web UI (named, unlimited, zoneable on the Detection Zones page), one-click CUDA PaddleOCR install, plus v1.0.10's zones-page logo fix and web-installable spaCy/PaddleOCR if you skipped that release.

OpenScrub v1.0.10

Web-installable spaCy NER and PaddleOCR (pip installs), fixed broken logo on the zones page, firewall rule now covers domain networks.