Research & trends

Useful research, connected to things people can try.

Brali does not collect papers for decoration. Research can strengthen a hack, challenge an old claim, suggest a new protocol, or show that something should stay uncertain.

Evidence boundary · 18 Aug 2026

Retrieval practice: memory is not application

Active recall remains useful for retention. Two 2026 studies show why Brali now separates tested-material memory, later learning, and actual application instead of calling all of them one testing effect.

Evidence note · 18 Aug 2026

Habits usually take longer than 21 days

A recent systematic review found wide variation in habit-formation time. Fixed streak targets are a poor substitute for a routine that actually fits.

AI note · 18 Aug 2026

RAG helps, but it is not a trust button

Retrieval can improve AI answers, but weak retrieval and source selection can also make them worse. Better data still needs better evaluation.

Living research pipeline

Brali keeps the research process as data instead of hiding it inside one prompt.

  1. Search lenses: research-queries.json maps maintained scholarly queries to Domains, Topics, Methods, and Lenses.
  2. Discovery queue: the weekly Research Scout deduplicates recent scholarly metadata into research-candidates.json.
  3. Evidence decision: an Evidence Reviewer reads the actual source and records the narrow claim it supports, challenges, or fails to support in evidence-decisions.json.
  4. Library change: only then can a candidate strengthen an existing hack, propose a new atomic hack, or become part of an executable protocol.

See the agents and skills → · See the knowledge contracts →

Why the queue is deliberately untrusted

Search metadata is useful for discovery, not for conclusions. A title can sound relevant while the population, intervention, outcome, effect size, or limitations say something much narrower. Candidate records therefore cannot receive Brali's reviewed evidence state without an explicit source review and evidence decision.

What belongs here

Current questions

From research to the library

A research note or candidate does not automatically become a trusted protocol. It can improve an existing hack, add or remove a claim, change an evidence state after review, create a new atomic hack proposal, suggest a protocol composed from existing actions, become a challenge-existing correction that narrows a claim, or remain a watch decision when the evidence is informative but not yet actionable.