Offline-first or always online: what is left of an app without a connection?
Using Notion, Obsidian, Google Maps, Toast and Figma, we examine which work should survive disconnection, where a live server decision matters and what each choice costs.
ARKTELOS LAB / ENGINEERING LABORATORY
Not an opinion about technology, but engineering work that can be checked.
Dart and Flutter architecture, integrations, packages, AI agents, and experiments — with exact experiment revisions, explicit limitations, and reproducible evidence.
PUBLIC CATALOGUE
Articles and issues
Research articles are bound to immutable revisions and Evidence Signatures. Issues bring published work together in one contents page.
Two devices changed the same record. What exactly needs syncing?
A phone and tablet edit one shopping-list record. How to preserve a pending local edit through revision checks, server conflicts, user choice and retries after a lost response.
A complex interface: when does animation help, and when does it get in the way?
How polished motion can confirm an action too soon: cart updates, stale delivery quotes, reduced-motion preferences and task-based evaluation of animation.
The tests are green. Why does the broken code still pass?
A Dart experiment shows why tests with full line coverage missed reversed pickup-point order, a radius boundary, and unavailable slots—and which assertions caught those faults.
September 2026. Between code and outcome
The September ArkTelos Lab issue: twelve articles on AI agents, architectural boundaries, lifetimes, accessibility and performance.
Upgrade a dependency or stay on the old version: what is the team actually deciding?
Upgrading flutter_secure_storage on Android: why an intermediate release does not guarantee migration, what the app must check, and when deferral is justified.
Beyond the screen: helping people understand and act
One class booking: typography, accessible buttons, gesture alternatives, feedback and the limits of local AI. A Flutter fixture separates what code checks from what requires real assistive technology.
JSON parsing moved to an isolate. Which problem did that solve?
Offloading JSON parsing can delay the result while reducing animation stalls. A Flutter experiment on M1 shows why the optimization criterion must be defined before choosing an API.
Cached data is not necessarily current data. What should a repository report?
The catalog is already visible, but its refresh fails. A Dart/Flutter experiment separates content, confirmation age, network failures and persistence failures, while Ark MVP carries those distinctions into screen state.
The agent has read the instructions. How can it inspect a running Flutter app?
The analyzer reports no issues and the model holds new orders, yet the screen keeps loading. An experiment examines what widget tests and MCP observations establish, and why an available tool does not necessarily become part of an agent’s verification.
The user signed out. What is still alive in get_it?
What closing a get_it scope does to retained references, asynchronous initialization and disposal failures. Four code examples examine where session cleanup ends and application responsibility begins.
The error was caught. Who decides what happens next?
Why catching an exception does not establish a save outcome. A Dart experiment with ark_error_manager separates diagnostics, user notifications and retry decisions.
CODE / DATA / AGENTS
Experiments do not disappear behind conclusions
When a record relies on code, a benchmark, or an agent run, its environment, commands, results, and limitations are available from the article.
Reproducibility rules and the Evidence Signature →ARKTELOS CHANNELS
News and engineering material, without mixing languages.
The main channels cover ArkTelos development. ArkTelos Lab publishes architecture analysis, experiments, and reproducible research.