# CurbSoftware Tech Innovations - Blog

- Site: https://blog.curbsoftware.com/

## Local-first is the whole product

- Link: https://blog.curbsoftware.com/local-first-is-the-product/
- Published: 2026-08-19T00:00:00+00:00
- Author: curbsoftware

Most software treats "works offline" as a degraded mode – a fallback you reach for
on a train. CurbApps inverts that. Local-first is not a fallback. It is the product.

## The default is your device

Install any CurbApps extension and it is immediately useful. Bookmark profiles, 
notes, one-time passwords, timers – they open from local storage, with no sign-up
and no round-trip to a server. There is no account wall, no onboarding email, no"
sync your data now" nag.

This matters for more than convenience. It changes what the software can know:

 * If your data never leaves your device, we cannot read it.
 * If we cannot read it, we cannot lose it, sell it, or hand it over.
 * If backup is your decision, uninstall is a clean break.

## The account exists for one thing

A CurbApps account does exactly one job: connect your devices when you want them
connected. Sign in, generate a key, switch sync on – and your encrypted data travels
to your other browsers and the phone. Switch it off and every device keeps working
from its local store.

The account never becomes a dependency. Local use is the stated spec of every shipped
product, not a marketing phrase.

## Why we think this is the right default

A product that works without you is rare. Most tools want your account first because
the account is the moat. We would rather the tool be the moat: something that is
genuinely useful the moment it is installed, and that earns the account when you
need the sync.

The trade-off is real. Without sync, your data lives only on the device, and backup
is your responsibility. We are honest about that in the terms – "local-first means
exactly what it says." We think a server that needs no trust is worth more than 
a server that demands it.

## What this means in practice

 * Install an extension and use it the same day, no account required.
 * Everything is stored locally, encrypted where it matters.
 * Sync is opt-in, end-to-end encrypted, and reversible.

Local-first is not a compromise we made for privacy. It is the reason the product
exists.

## How CurbApps encrypts your data

- Link: https://blog.curbsoftware.com/how-curbapps-encrypts-your-data/
- Published: 2026-08-18T00:00:00+00:00
- Author: curbsoftware

One claim keeps appearing in the CurbApps docs: "we cannot read your synced data."
That is not a slogan. It is a consequence of how the system is built. Here is the
model in plain language.

## The shape of it

Personal data such as a bookmark, note, OTP account, or vault entry is represented
as an _entity_. Before a synced entity leaves your device, it is encrypted with 
AES-256-GCM. The server receives ciphertext, stores ciphertext, and forwards ciphertext.
It does not receive an unwrapped data key, so it cannot read the plaintext.

## Where the key lives

The app generates the data-encryption key on your device. A master password derives
a separate wrapping key that protects it. Local auto-unlock keeps device-only key
material on that installation.

 * **Master password**: the data key is wrapped under a key derived from a password
   only you know. Unlocking requires it, and an encrypted recovery blob lets another
   signed-in device restore the workspace.
 * **Auto-unlock**: the device holds local key material without asking for a password.
   This mode is local-only and does not create the recovery material required by
   the current cloud sync path.

Use a master password when you want cloud sync, the web data manager, or recovery
on another device.

## What the server can see

The server can see that you synced _something_, at some size, at some time. It also
needs routing metadata such as the account, product scope, and entity type. It cannot
read encrypted fields such as titles, note content, passwords, OTP secrets, or tags.

## What this costs you

Zero-knowledge has a trade-off: there is no server-side "forgot password" path for
encrypted workspace data. Lose the master password and no support agent can decrypt
that data for you. The recovery blob helps a device restore the wrapped key, but
it does not reveal the password or plaintext to the server.

## One deliberate exception

One Premium feature, bookmark classification, sends a title and URL to a classifier.
Page content is never sent, and the result is encrypted before it syncs.

## In summary

Encrypt on the device, derive keys on the device, store ciphertext on the server.
Three decisions, one outcome: a sync service that does not need your trust, because
it cannot abuse it.

## Multi-Frame Super-Resolution, Compared: BasicVSR, BSRT, and Deep Burst SR

- Link: https://blog.curbsoftware.com/civicset-multi-frame-super-resolution-compared/
- Published: 2026-08-16T00:00:00+00:00
- Author: curbsoftware

Every camera that shoots a burst or a video clip is quietly collecting more detail
than any single frame contains. Hand motion, rolling shutter, and optical blur scatter
that detail across neighboring frames at sub-pixel offsets: each frame samples the
scene through a slightly different grid. Multi-frame super-resolution (MFSR) is 
the family of techniques that registers those frames against each other and fuses
the scattered detail back into one image. The models differ enormously in what they
accept as input and how they align it, and those two choices explain almost everything
else about them.

This article compares three reference points in the modern landscape: **BasicVSR/
BasicVSR++** for continuous video, **Deep Burst Super-Resolution (DBSR)** for RAW
sensor bursts, and the **Burst Super-Resolution Transformer (BSRT)**, which pushed
burst SR further with transformer-based fusion.

## The classical foundation: shift-and-add

Before deep learning, MFSR was a registration-and-averaging problem. Estimate the
sub-pixel translation of each frame against a reference, resample every frame onto
a common high-resolution grid, and accumulate. The astronomy community's drizzle
algorithm formalized this for undersampled telescope imagery, and the same idea (
cross-correlation shift estimation, bilinear resampling into a float accumulator,
then a final upscale) remains a respectable baseline when frames differ only by 
translation. Its ceiling is equally well understood: it models none of the optics,
the noise, or the demosaicing, and it breaks down the moment motion stops being 
a pure translation. Everything below exists to remove one of those limitations.

## BasicVSR and BasicVSR++: the video baseline done right

BasicVSR (CVPR 2021) is a study in restraint. The authors decompose video super-
resolution into four functionalities (**propagation, alignment, aggregation, and
upsampling**) and then assemble a bidirectional recurrent network almost entirely
from generic parts: residual blocks for features, a small optical-flow network (
SPyNet) for alignment, and straightforward feature fusion before upsampling. Because
the recurrent stream passes information forward and backward through the whole sequence,
each output frame can draw on every other frame, not just its temporal neighbors.
The paper's claim is efficiency: it outperformed the state of the art of its time
with fewer parameters and faster inference, and it doubles as a backbone you can
extend. Its sibling IconVSR adds two such extensions (an information-refill mechanism
and coupled propagation) for sharper 4× results at comparable speed.

BasicVSR++ (CVPR 2022) is where the extensions get systematic. It replaces the propagation
structure with **second-order grid propagation** (each step reaches across the temporal
grid two frames back, not one) and swaps flow-warped alignment for **flow-guided
deformable alignment**, letting a deformable convolution refine the optical-flow
estimate in feature space. The reported gain over BasicVSR is **0.82 dB PSNR at 
a similar parameter count**. On the standard REDS4 benchmark (BI degradation), MMagic's
published checkpoints put BasicVSR at **31.42 dB / 0.8909 SSIM** and BasicVSR++ 
at **32.39 dB / 0.9069**; trained on Vimeo-90K instead, BasicVSR reaches **37.20
dB** on Vimeo-90K-T (BI). In competition, BasicVSR++ took three championships and
one runner-up finish across the NTIRE 2021 Video Restoration and Enhancement tracks.
Both models are maintained today inside [MMagic](https://github.com/open-mmlab/mmagic)(
configs and pretrained weights for BasicVSR, IconVSR, BasicVSR++, and RealBasicVSR),
which has effectively become the reference implementation.

## Deep Burst Super-Resolution: treating the sensor honestly

DBSR (CVPR 2021, Bhat et al.) changes the input contract. Instead of processed sRGB
frames, it consumes a burst of **noisy, mosaicked RAW images** (14 frames in its
benchmarks) and must produce a clean, demosaicked, 4×-super-resolved RGB image in
one pass. That contract folds three problems into one network: super-resolution,
denoising, and demosaicing.

Architecturally, DBSR extracts deep features from every frame and **aligns those
embeddings explicitly using pixel-wise optical flow** (a pretrained PWC-Net), rather
than warping pixels before feature extraction. An attention-based fusion module 
then decides, per location, how much each aligned frame contributes. Training uses
bursts synthesized from the Zurich RAW-to-RGB dataset through an inverse camera 
pipeline (translation, downsampling, mosaicing, and realistic noise) and evaluation
runs on both that synthetic set (300 validation bursts; the paper reports **39.09
dB PSNR / 0.945 MS-SSIM**) and the real-world BurstSR set of Samsung Galaxy S8 RAW
bursts with DSLR ground truth (**47.76 dB** under the challenge's alignment-and-
color-corrected metric, not comparable to the synthetic number, but indicative of
how much information a real burst carries).

DBSR also defined the benchmark it competes on: the [NTIRE 2021 Burst Super-Resolution Challenge](https://github.com/goutamgmb/NTIRE21_BURSTSR)
adopted its datasets and evaluation protocol, with DBSR serving as the challenge
baseline. The challenge itself was won by EBSR, a deformable-convolution approach
that, like several of its successors, borrowed DBSR's problem statement wholesale.

## BSRT: deformable alignment meets Swin Transformers

The Burst Super-Resolution Transformer (CVPR Workshops 2022, Luo et al.) starts 
from the same RAW-burst contract as DBSR but rebuilds the machinery. Alignment is
a **Pyramid Flow-Guided Deformable Convolution Network**: optical flow (again from
a pretrained PWC-Net) pre-aligns frames, and a pyramid of deformable convolutions
refines that alignment in feature space, aggregating texture across frames that 
flow alone cannot register. The backbone is built from **Swin Transformer blocks
and groups**, whose windowed attention captures long-range dependencies that convolutional
burst fusion misses. BSRT won the **real-world track of the NTIRE 2022 Burst Super-
Resolution Challenge** and set a new state of the art on both synthetic and real
BurstSR data, training on the NTIRE 2021 synthetic and real sets. An extended journal
version, BSRT++, followed in 2024. The [reference implementation](https://github.com/Algolzw/BSRT)
is MIT-licensed PyTorch, with one practical caveat: it builds DCNv2 (deformable 
convolution) from source, so reproducing it takes more than `pip install`.

## The wider map

Two neighbors are worth knowing. **EBSR**, the NTIRE 2021 winner mentioned above,
introduced the enhanced deformable-convolutional alignment grid that BSRT's flow-
guided variant builds upon. And in the satellite domain, the ESA Proba-V challenge
produced a family of non-sequential MFSR networks (DeepSum and its successors) that
register and fuse as few as nine low-resolution frames of the same ground patch;
a compact [PyTorch reimplementation](https://github.com/taynoel/Multi-Frame-Super-Resolution)
uses spatial-transformer registration nets with a two-stream fusion module, and 
is a readable example of the whole genre.

## Comparing them, dimension by dimension

### Input domain

 * **BasicVSR/++**: continuous sRGB video: frames are dense in time and already 
   processed.
 * **DBSR**: RAW Bayer bursts (RGGB, noisy, mosaicked), 14 frames, taken seconds
   apart.
 * **BSRT**: the same RAW burst contract as DBSR.
 * **Satellite MFSR**: short non-sequential bursts (9+ frames) of the same static
   target.

The RAW-versus-sRGB split is the most consequential line in the field: a RAW-domain
model inherits denoising and demosaicing as part of the task, and cannot be fed 
ordinary video without an inverse pipeline.

### Alignment mechanism

 * **BasicVSR**: optical-flow warping of features between recurrent steps.
 * **BasicVSR++**: flow-guided _deformable_ alignment: flow proposes, DCN refines.
 * **DBSR**: pixel-wise optical flow applied to deep feature embeddings.
 * **BSRT**: pyramid flow-guided deformable convolution: flow pre-alignment refined
   across scales.

The progression is consistent across both families: pure flow gives way to flow-
conditioned deformable convolution, because bursts and hard video motion violate
flow's smoothness assumptions.

### Aggregation and fusion

 * **BasicVSR/++**: implicit fusion inside the recurrent stream (plus IconVSR's 
   explicit information refills).
 * **DBSR**: attention-weighted per-location fusion of aligned embeddings.
 * **BSRT**: Swin Transformer burst fusion over aligned multi-frame features.

### What the network owes you

A video model owes one upsampled frame per input frame. A burst model owes a single
clean RGB image (super-resolution _plus_ denoising _plus_ demosaicing) from the 
whole burst. The second contract is harder, which is why RAW burst models can post
PSNR numbers that look (and are) incomparable with sRGB video benchmarks.

### Efficiency and practicality

BasicVSR's founding claim was efficiency against then-SOTA, and it still holds: 
it remains the sensible default when compute matters. BasicVSR++ buys 0.82 dB for
a similar parameter count. The burst models are a different regime (trained on multi-
GPU setups with patch-based crops) and BSRT adds a from-source DCNv2 build to the
bill.

### Code, weights, and licenses

 * [BasicVSR / IconVSR](https://github.com/ckkelvinchan/BasicVSR-IconVSR) (Apache-
   2.0), with maintained configs and weights in [MMagic](https://github.com/open-mmlab/mmagic).
 * [DBSR](https://github.com/goutamgmb/deep-burst-sr) and the [NTIRE21 challenge toolkit](https://github.com/goutamgmb/NTIRE21_BURSTSR).
 * [BSRT](https://github.com/Algolzw/BSRT) (MIT).

## Further reading

 * [BasicVSR: The Search for Essential Components in Video Super-Resolution and Beyond](https://arxiv.org/abs/2012.02181),
   Chan et al., CVPR 2021, plus the [project page](https://ckkelvinchan.github.io/projects/BasicVSR/).
 * [Deep Burst Super-Resolution](https://arxiv.org/abs/2101.10997), Bhat et al.,
   CVPR 2021.
 * [BSRT: Improving Burst Super-Resolution with Swin Transformer and Flow-Guided Deformable Alignment](https://arxiv.org/abs/2204.08332),
   Luo et al., CVPR Workshops 2022.
 * MMagic config readmes for [BasicVSR](https://github.com/open-mmlab/mmagic/blob/main/configs/basicvsr/README.md)
   and [BasicVSR++](https://github.com/open-mmlab/mmagic/blob/main/configs/basicvsr_pp/README.md)
   carry the full benchmark tables and training recipes.

The unifying lesson across all of them is the one the classical shift-and-add baseline
already taught: alignment quality bounds everything downstream. What changed is 
that alignment moved into feature space, became learnable, and now carries most 
of the performance gains the field has made since.

## Introducing CurbApps Mobile

- Link: https://blog.curbsoftware.com/introducing-curbapps-mobile/
- Published: 2026-08-12T00:00:00+00:00
- Author: curbsoftware

CurbApps started as a set of browser extensions, tools that live where the work 
happens. The phone was always the missing surface. CurbApps Mobile is our work in
progress for that space.

## What it is

CurbApps Mobile is a signed-in companion with five tabs:

 * **Page**: synced bookmark profiles plus live TOTP and HOTP codes.
 * **Place**: recent notes, a focused field-based composer, and synced popup-tool
   preferences.
 * **Tube**: the encrypted CurbTube watchlist.
 * **Vault**: passwords and other encrypted vault entries shared with CurbPage and
   CurbPlace.
 * **Metrics**: world clocks and current currency reference rates.

## How it fits

The app uses the same account, license key, and encrypted entity model as the extensions.
It requires sign-in. Cloud sync also requires recoverable master-password material,
so a local auto-unlock key by itself is not enough.

The mobile surfaces stay focused. The Place composer does not replace CurbPlace’s
Plate editor, folder manager, or conflict-resolution screen. Mobile shows the enabled
CurbPlace popup tools, but it does not execute the tool catalog.

## Built offline-first

Changes queue locally while the phone is offline. Sync runs only while the app is
in the foreground, then pauses when the app is backgrounded. Master-password sessions
lock when the app leaves the active AppState. Biometric unlock and a configurable
lock timer are not part of the current build.

## Where to get it

The app is not yet published in the App Store or Google Play. Contributors and invited
testers can run development builds from the monorepo. We will add public store links
after the releases are live.

This first release is a companion, not a full copy of every extension screen. That
smaller surface is intentional, and the documentation will grow with the shipped
app.

## How to modernize a public-sector risk register

- Link: https://blog.curbsoftware.com/riskcurb-how-to-modernize-a-public-sector-risk-register/
- Published: 2026-05-20T00:00:00+00:00
- Author: curbsoftware

Most public agencies still run their risk register in a spreadsheet. It works, until
it doesn't. Versions drift, owners change, ratings go stale, and nobody is quite
sure which file is current before a committee meeting. Modernizing the register 
is less about new software and more about giving risk a single, accountable home.

## Start with one register, not a platform

You do not need to connect everything on day one. Pick the register that matters
most, whether enterprise risk, a major capital project, or a department, and move
it first. A single source of truth for one program is more valuable than a half-
finished rollout across ten.

## Make ownership non-negotiable

Every risk should carry an owner, an accountable department, a rating, and a next-
review date. These four fields turn a static list into a living register. When ownership
is explicit, reviews stop being a scramble and start being a routine.

## Connect treatment to the risk

A register that only records risks is half a system. The value appears when each
risk links to its controls and treatments, with owners, due dates, and evidence.
That linkage is what makes reporting trustworthy: leadership can see not just what
the risks are, but what is being done about them.

## Keep the migration honest

Moving off spreadsheets is a good moment to retire risks that no longer apply, merge
duplicates, and re-rate what is left. Bring the data in clean rather than importing
years of drift. A smaller, accurate register beats a large, stale one.

## Where this leads

Once a single register is connected, controls, claims context, and reporting follow
naturally. The register stops being a document people maintain and becomes the operating
view the whole program works from.

## How to modernize a public-sector risk register

- Link: https://blog.curbsoftware.com/claimscurb-how-to-modernize-a-public-sector-risk-register/
- Published: 2026-05-20T00:00:00+00:00
- Author: curbsoftware

Most public agencies still run their risk register in a spreadsheet. It works, until
it doesn't. Versions drift, owners change, ratings go stale, and nobody is quite
sure which file is current before a committee meeting. Modernizing the register 
is less about new software and more about giving risk a single, accountable home.

## Start with one register, not a platform

You do not need to connect everything on day one. Pick the register that matters
most, whether enterprise risk, a major capital project, or a department, and move
it first. A single source of truth for one program is more valuable than a half-
finished rollout across ten.

## Make ownership non-negotiable

Every risk should carry an owner, an accountable department, a rating, and a next-
review date. These four fields turn a static list into a living register. When ownership
is explicit, reviews stop being a scramble and start being a routine.

## Connect treatment to the risk

A register that only records risks is half a system. The value appears when each
risk links to its controls and treatments, with owners, due dates, and evidence.
That linkage is what makes reporting trustworthy: leadership can see not just what
the risks are, but what is being done about them.

## Keep the migration honest

Moving off spreadsheets is a good moment to retire risks that no longer apply, merge
duplicates, and re-rate what is left. Bring the data in clean rather than importing
years of drift. A smaller, accurate register beats a large, stale one.

## Where this leads

Once a single register is connected, controls, claims context, and reporting follow
naturally. The register stops being a document people maintain and becomes the operating
view the whole program works from.

## Connecting claims and risk management

- Link: https://blog.curbsoftware.com/riskcurb-connecting-claims-and-risk-management/
- Published: 2026-04-28T00:00:00+00:00
- Author: curbsoftware

In most public agencies, claims live in one system and risk lives in another. The
claims team tracks incidents, reserves, and renewals. The risk team maintains the
register. The two rarely meet, which means the clearest evidence of where things
actually go wrong never reaches the people deciding how to prevent it.

## Loss history is a risk signal

Every claim is a record of something that already happened. Patterns in claims, 
such as a recurring property loss or a service area that keeps generating liability,
are direct evidence about operational risk. When claims connect back to the relevant
risks and controls, that history stops being a closed file and becomes a signal 
that shapes treatment.

## Connect, don't merge

Connecting claims to risk does not mean RiskCurb becomes a system of record for 
insurance or legal work. It means an incident or claim can be linked to the risks,
assets, and service areas it relates to. The claims team keeps their workflow; the
risk team gains context they never had.

## What this changes in practice

 * Reserve and renewal reviews can reference the operational risks behind the numbers.
 * A spike in a loss theme can trigger a treatment review, not just a payout.
 * Reporting to leadership can show exposure and claims trends side by side.

## A measured approach

Coverage context, reserves, and deductibles belong with the people accountable for
them. The goal is not to overclaim an insurance capability, but to make sure loss
history informs decisions instead of sitting in a separate silo. Connected claims
context is one of the most practical ways to make a risk program smarter over time.

## Connecting claims and risk management

- Link: https://blog.curbsoftware.com/claimscurb-connecting-claims-and-risk-management/
- Published: 2026-04-28T00:00:00+00:00
- Author: curbsoftware

In most public agencies, claims live in one system and risk lives in another. The
claims team tracks incidents, reserves, and renewals. The risk team maintains the
register. The two rarely meet, which means the clearest evidence of where things
actually go wrong never reaches the people deciding how to prevent it.

## Loss history is a risk signal

Every claim is a record of something that already happened. Patterns in claims, 
such as a recurring property loss or a service area that keeps generating liability,
are direct evidence about operational risk. When claims connect back to the relevant
risks and controls, that history stops being a closed file and becomes a signal 
that shapes treatment.

## Connect, don't merge

Connecting claims to risk does not mean RiskCurb becomes a system of record for 
insurance or legal work. It means an incident or claim can be linked to the risks,
assets, and service areas it relates to. The claims team keeps their workflow; the
risk team gains context they never had.

## What this changes in practice

 * Reserve and renewal reviews can reference the operational risks behind the numbers.
 * A spike in a loss theme can trigger a treatment review, not just a payout.
 * Reporting to leadership can show exposure and claims trends side by side.

## A measured approach

Coverage context, reserves, and deductibles belong with the people accountable for
them. The goal is not to overclaim an insurance capability, but to make sure loss
history informs decisions instead of sitting in a separate silo. Connected claims
context is one of the most practical ways to make a risk program smarter over time.

## Reporting risk to executive leadership

- Link: https://blog.curbsoftware.com/riskcurb-reporting-risk-to-executive-leadership/
- Published: 2026-04-02T00:00:00+00:00
- Author: curbsoftware

Ask most risk managers about board reporting and you will hear the same thing: it
is rebuilt by hand every cycle. Numbers are copied from spreadsheets, slides are
reformatted, and by the time the report is approved it is already out of date. The
problem is not effort. It is that reporting is disconnected from the data the team
works in.

## What leadership actually wants

Boards and councils do not want every risk. They want a clear picture of exposure,
trend, and what is being done about it:

 * Where is exposure concentrated, and is it rising or falling?
 * Which treatments are on track, and which are overdue?
 * How do strategic risks connect to the objectives leadership cares about?

## Report from live data, not a rebuild

When reporting draws from the same connected register the team maintains, the quarterly
rebuild disappears. Trend and exposure views reflect the current state. Overdue 
controls surface automatically. The report becomes a view of reality rather than
a snapshot someone assembled by hand.

## Keep judgment with people

AI can draft a leadership summary or surface a trend, and that saves real time. 
But the framing, the emphasis, and the approval stay with the risk owners and the
team presenting to leadership. Assistive reporting accelerates the work without 
replacing the judgment that makes it credible.

## The shift

Good risk reporting is not a separate artifact. It is the same operating view, presented
for the audience that needs it. Once reporting is connected to live data, leadership
gets a more honest picture, and the team gets its quarter back.

## Reporting risk to executive leadership

- Link: https://blog.curbsoftware.com/claimscurb-reporting-risk-to-executive-leadership/
- Published: 2026-04-02T00:00:00+00:00
- Author: curbsoftware

Ask most risk managers about board reporting and you will hear the same thing: it
is rebuilt by hand every cycle. Numbers are copied from spreadsheets, slides are
reformatted, and by the time the report is approved it is already out of date. The
problem is not effort. It is that reporting is disconnected from the data the team
works in.

## What leadership actually wants

Boards and councils do not want every risk. They want a clear picture of exposure,
trend, and what is being done about it:

 * Where is exposure concentrated, and is it rising or falling?
 * Which treatments are on track, and which are overdue?
 * How do strategic risks connect to the objectives leadership cares about?

## Report from live data, not a rebuild

When reporting draws from the same connected register the team maintains, the quarterly
rebuild disappears. Trend and exposure views reflect the current state. Overdue 
controls surface automatically. The report becomes a view of reality rather than
a snapshot someone assembled by hand.

## Keep judgment with people

AI can draft a leadership summary or surface a trend, and that saves real time. 
But the framing, the emphasis, and the approval stay with the risk owners and the
team presenting to leadership. Assistive reporting accelerates the work without 
replacing the judgment that makes it credible.

## The shift

Good risk reporting is not a separate artifact. It is the same operating view, presented
for the audience that needs it. Once reporting is connected to live data, leadership
gets a more honest picture, and the team gets its quarter back.
