What's shipped, honestly.

We are opening the gates, slowly, to the first Fluera beta cohort. iOS (iPadOS) and macOS ship first — Android, Windows and Linux follow as we work through the alpha backlog.

What’s included

• Infinite canvas with 13 pressure-sensitive brushes
• Full 12-step learning cycle, with all four pillar features enabled
• Spaced repetition scheduler with personalised FSRS weights
• Local-first SQLCipher-encrypted storage, opt-in cloud sync
• End-to-end encrypted P2P collaboration (Visit, Teaching, Duel modes)

What’s not yet

• Time Travel audio-sync is still alpha on iOS only
• Android release in the next beta cycle
• Education audit-log dashboard (Phase 2)

How to get in

Join the beta waitlist. We onboard in batches of around fifty a week and read every reply.

Ghost Map ships in its first usable form. After Step 3 of the cycle, students can trigger a comparison overlay: their canvas versus an AI-generated ideal map. Mismatches pulse visually.

New

• Ghost Map overlay with red (missing), yellow (mis-linked), blue (added) cues
• Confidence slider on every Socratic reveal — primes hypercorrection
• On-canvas attempt flow: students write the correction by hand before the ideal reveals
• FAB with Confronta/Rivela buttons during attempts

Improved

• Gemini positioning is non-deterministic; we added an anti-overlap post-pass (350px min spacing)
• Drawing pipeline no longer blocks during inline attempt
• Label text is now anchored at fixed screen size (160px, 1 line); rectangles scale with canvas

Known issues

• Gemini occasionally bunches nodes on one cluster — the post-pass sometimes pushes them far from related groups
• No auto-detect for when a student finishes writing an attempt (manual confirm only)
• Visual polish on ghost nodes (animations, transitions) will improve in v0.10

The default FSRS scheduler gets most of the way there. The last 20% — the part that distinguishes a good spaced-repetition system from a great one — is personalisation.

New

• Eight FSRS weights (w0–w7) calibrated per-user via gradient descent on review history
• Pedagogical modifiers stacked on top:
  • Hypercorrection bonus — confidence-5 wrong answers get shorter next intervals (and a priority badge)
  • Peek malus — premature reveals shorten the next interval (no free lookup)
  • Response-time signal — slow correct responses treated as weaker than fast correct responses
  • Breadcrumb scaffolding — hint chains scale the interval by how much scaffolding was needed
• Per-subject weight calibration (medicine vs math vs languages)

Improved

• Review-queue ordering now uses topological sort of prerequisites, not flat date-based
• Post-review feedback celebrates effort rather than correctness (growth mindset microcopy)

Thanks

To everyone in the closed alpha who generated enough review data for the gradient descent to actually converge. The scheduler is better because your reviews were imperfect.