v0.6.0

Added

• EXPERIMENTAL presentation variant for detectPitfalls() —
  variant: 'summary' adds a one-to-two-sentence overall summary in the
  book's guide-not-judge voice (framing pitfalls around what the work's
  audience would misperceive, optionally noting one genuine strength) and a
  per-finding consequence rating (changes-takeaway / weakens-support /
  polish), and an avoided list — up to two catalog pitfalls the work
  visibly avoided, evidenced by a concrete countermeasure (a guard, a
  stated caveat, a deliberate choice), catalog-validated and disjoint from
  findings, with zero allowed. Explanations are capped at one to two
  artifact-specific sentences (the rule's general description is not
  restated per finding). Off by default; not yet exposed in the CLI or web
  app. A new dev-only A/B harness (evals/compare.mjs) runs the same
  artifacts through the variants and writes a side-by-side report — now
  including an avg-words-per-finding column — to evaluate fixes for
  finding-overload feedback before shipping any of them. (Earlier rounds
  trialed 'verdict' naming and a separate mandatory strengths field; both
  were cut after side-by-side review.)

• The web app now runs every scan with the summary presentation and presents
  results triaged instead of as a flat list: the overall summary leads, the
  most important finding is expanded under "Start here", remaining findings
  collapse to one-line rows grouped as Detected / Potential Pitfalls (with
  lower-confidence potential ones behind a "show more" toggle, matching the
  CLI default), consequence chips ("Changes the takeaway" / "Weakens
  support" / "Polish") replace severity badges, and visibly-avoided pitfalls
  close the report as earned credit.

• CLI: datapitfalls scan --summary opts a scan into the summary
  presentation — the report leads with the overall summary, findings carry
  consequence ratings, and avoided pitfalls close the output.

Changed

• Report labels reworded in the book's voice: findings group under "Detected
  Pitfalls" / "Potential Pitfalls" (instead of active/latent), with
  "Why it matters" / "Where it shows up" / "How to avoid it" field labels
  (instead of Why/Evidence/Fix). Machine-facing API fields (nature,
  severity, etc.) are unchanged.

• Public API reference (docs/API.md) documenting the supported
  library surface — detectPitfalls(), input/report types, formatReport,
  file routing, and taxonomy queries — plus an API-stability and semver policy,
  with a "Programmatic API" section linked from the README. Completes Phase 6 of
  the Roadmap.

Fixed

• VERSION is now read from package.json at runtime instead of a hardcoded
  literal, so the library and CLI always report the real published version.