engineering intelligence · react & react native

React decisions, with receipts.

Hundreds of libraries, conflicting takes, advice that quietly rots. react-brain digests the ecosystem continuously and turns it into context-keyed recommendations — every load-bearing fact carries a verified-on date and a primary source, and every default is a forecast graded on the public scorecard.

42decisions 123verified readings 180libraries tracked 75%calibration · high tier

Analyze your project →Browse decisions

ecosystem pulse: last verification 2026-07-13 · 37 decisions re-verified this week · 6 sources digested

$ npx -y @heart-it/react-brain doctor .

ourpot (v1.4) · react-native · production · 41 deps

  state        Zustand              ✓ aligned
  lists        FlashList            ✓ aligned
  nav          Expo Router          ✓ aligned
  data         Hypercore (P2P)      ~ contextual
  animation    Reanimated 4         ✓ aligned
  rn-versions  0.83 → 0.86          ↗ review

2 source signals · 1 gap · docs: entries/rn-versions

how it works →

6 newsletters · release notes · npm · 34 shipped appsevery claim fetch-verified & dated42 context-keyed decisionsdoctor · migrate · review · MCPmethodology →

Browse by decision

React & language foundations3 App architecture9 UI12 Platform & native9 Build, test, observe, secure6 AI in React apps3

This week

recommendation of the week

Run strict TypeScript (strict:true, the TS 6 default) and export a small, deliberate public type surface from…

typescript · medium · verified 2026-07-13

Read the reasoning →

Quick start

$ npx -y @heart-it/react-brain doctor .      # assess a repo, ranked priorities
$ npx -y @heart-it/react-brain migrate .     # sequenced upgrade plan
$ npx -y @heart-it/react-brain review . --ci # block dead deps in CI

Organized by decision, not by article — 123 deep-dives cited as evidence, every default a dated prediction on the scorecard, and nothing asserted casually.

One corpus, every surface

# this website — press ⌘K for the console

# your terminal, zero install
$ npx -y @heart-it/react-brain doctor .

# your coding agent — 9 MCP tools
$ claude mcp add react-brain -- npx -y @heart-it/react-brain mcp

# your CI — dead/deprecated dep adds fail the build, with receipts
$ npx -y @heart-it/react-brain review . --base=origin/main --ci

package: npmjs.com/package/@heart-it/react-brain

Not another model answer

a model predicts from training datathe corpus fetch-verifies against primary sources, and stamps the date

a model's confidence is tonehere confidence is a forecast graded in public — the scorecard

a model's knowledge freezes at cutoffthe corpus is re-verified weekly from 6 cross-checked sources — changelog

Don’t take the framing on faith — the staleness benchmark measured it: claude-sonnet-5 alone answers 70% fresh and hedges a third of the time; with the corpus injected it hits 94% fresh and barely hedges at all. Deterministic rubric, conditions labeled, reproducible with your own key.

React decisions, with receipts.▊