Methodology — how this stays true

Most React resources rot silently: a listicle is true the day it's published and unaccountable after. react-brain is built around the opposite bet — every claim should be checkable, dated, and eventually scored. Five mechanisms do the work:

1. Verified facts, not vibes

Load-bearing facts (versions, maintenance status, breaking changes) are checked against primary sources — official blogs, changelogs, the npm registry — before they enter an entry, and stamped with a verified: date you can see on every page. Newsletter claims are treated as leads, never as facts.

2. Context-keyed recommendations

No entry says "X is best." Recommendations are written as context → choice clauses ("GraphQL backend → Apollo"; "P2P/serverless app → a client cache is N/A by design"). The same resolver that renders these pages powers the CLI and the doctor, so the advice can't drift between surfaces.

3. Adversarial challenge

High-stakes recommendations are periodically attacked: a challenger steelmans the case against the default and tests it with current evidence. Verdicts — SURVIVES / WEAKENED / OVERTURNED — require concrete evidence, and overturning demands the alternative win on a stated axis. Survived challenges are recorded on the entry; they're a stronger trust signal than editorial review.

4. Calibration — a published track record

Every default is logged as a dated, falsifiable prediction with a check-by horizon (shorter for fast-moving domains). Resolutions land in an append-only ledger and are scored per confidence tier on the scorecard. Unresolved predictions are never counted as wins — "unproven" is displayed as unproven.

5. External anchors + mechanized invariants

Editorial opinion is checked against the outside world: live npm downloads and publish dates flag claims that data contradicts (a library called "maintenance mode" that shipped last week gets its claim weakened — that has actually happened, see the scorecard). Structural rules — every entry has vetted reading, every category is reachable, no duplicate sources, schemas hold — run as an offline lint plus a golden-fixture eval in CI on every change.

Editorial line on articles

Great articles are cited with a short editorial annotation on why they matter and linked prominently to the original. Nothing is republished or paraphrased at length — the reading index exists to send you to the canonical source with context, not to replace it.

Sources & thanks

The corpus's discovery layer is six excellent newsletters, digested issue by issue — our genuine thanks to the people who curate them week after week:

To be precise about the division of labor: newsletters surface the leads — every load-bearing fact is then independently fetch-verified against primary sources (release notes, changelogs, the npm registry) before it enters the corpus, and the newsletters have not reviewed or endorsed anything here. The same gratitude extends to every author in the reading index, credited by name on their entries.

Everything above is inspectable: the corpus, the tools, the ledger, and the tests are all in the repo.