What should happen next?
The user can accept or override the suggestion; the final choice determines the item’s path.
Project note · 2026
Information Triage & Learning Assistant
FOMO Firewall
An information-triage prototype for knowledge workers that turns endless saving into three executable choices: Learn now, Later, or Ignore.
Why I started
Information overload pushes people to spend attention on repeated screening. More content creates more saves, but not more material that enters learning or work. The missing piece is not another summary—it is a low-cost, reversible way to decide what to do next.
The real question
How can a system recommend a next action with reasons, while making it easy for users to accept or override—and use those actions to improve later recommendations?
Choices I made
Every action should have a reason.
Start with RSS as a stable content input.
The first version needed to test whether triage changes behavior—not crawler coverage—so a stable input kept effort on the core journey.
Constrain output to Learn / Later / Ignore.
Each option maps to a next action, rather than returning a summary and leaving the original decision cost untouched.
Record acceptance and overrides instead of treating model accuracy as the only metric.
Adoption is closer to product value, while overrides reveal failures in rationale, preference, or category boundaries.
How the system moves
How an item moves from backlog to a clear action
Start / inputProcessDecisionOutcome
- Source
Item enters the inbox
An RSS item and its source metadata enter the triage queue.
- System
Add decision context
Combine topic, source, user intent, and durable preferences.
- Agent
Recommend with reasons
Return a reversible action rather than another summary.
Start a learning session
Progressive questions turn reading into recallable understanding.
Move to a deferred queue
Keep rationale and priority so the user does not screen from zero.
End this item’s journey
Remove it from attention while preserving the decision record.
Feedback loop
Write acceptance and overrides to memory
The next triage decision uses what was accepted, what changed, and why.
Business meaningThe product does not create value by reading on the user’s behalf. It reduces repeated judgment and uses every acceptance or override to improve the next triage decision.
What I made
From judgment to something that runs.
Triage
Combine topic, source, and personal intent into a recommendation with a rationale.
Learn
Use three progressively sharper prompts to turn reading into recallable understanding.
Remember
Separate immediate decisions, durable preferences, and configurable prompts so the system can adapt.
What the work left behind
- Core journey verified end to end in a personal environment
- Reproducible container image available
- Overrides recorded to test whether recommendations actually save judgment
Current evidence comes from personal use, not broad user validation. Override rate still needs a longer observation window.
The next step is not more summarization, but testing how different rationales affect trust and overrides.