v3.2.2

Headline: make the dashboard readable. A four-persona review of the deployed
GitHub Pages dashboard (assessor, AI-agent, skeptic, first-time engineer) found
it unusable — an 18.8 MB self-analysis page, an inverted pyramid, no structured
feed, and no legible scope. This release fixes the scry-viz output. No analyzer
change.

Changed — scry-viz dashboard

• Capped the Program points section — was ~90% of the page bytes (the 18.8 MB
  culprit). Each function now shows a summary + the first 20 points + a pointer to
  the JSON feed; the whole HTML stays well under 1 MB (test-enforced).
• Collapsed the Guidance boilerplate — every DefiniteFault in full (proven
  bugs are never elided), other classes capped with a "… and N more" pointer, so
  the real fixes no longer drown under thousands of identical rows.
• Retitled from "verification dashboard" to "scry — a sound static analyzer
  for WebAssembly", with a lead that states what it catches and the soundness
  caveat: ⊤ / POTENTIAL-TRAP mean "unknown", never "safe".
• New Scope & limitations block — the mechanized-vs-γ-sweep-validated evidence
  distinction, and what scry does NOT prove (no WasmCert-Coq import yet;
  official-semantics proof covers i32.add only; memory content singleton-i32
  only; not a qualified tool — no TQL/TCL claim).

Added — structured guidance feed

• scry-viz::render_guidance_json + a <stem>.guidance.json written alongside
  the HTML (and emitted + sanity-checked in release.yml): the full, un-capped
  advisories + trap verdicts as machine-consumable JSON — the structured-primary
  feed an AI agent should consume instead of scraping HTML.

Notes

• FEAT-057 slice-2 reframed (rivet): a straight-line polyhedra pass is near-vacuous
  over Wasm's wrapping i32 (general-coefficient facts need no-wrap → bounds →
  guards → the fixpoint), so the polyhedra body is scoped to a fixpoint-integration
  arc, tracked honestly rather than shipped vacuous.