Roadmap
Direction, not promises. No dates, and nothing here should be treated as a commitment you can plan around.
Why there are no dates. A roadmap with dates on it is a set of promises, and promises about software are broken more often than they are kept. What is useful to you is knowing what we think matters and in roughly what order — so that is what this is. The first group is deliberately the things we have already publicly admitted are missing; a roadmap that skips its own acknowledged gaps is a wish list.
Commitments we have already made publicly
These are gaps the Trust Center admits to. They are listed first because a roadmap that skips its own acknowledged omissions is a wish list.
A published Privacy Policy page
The Trust Center states plainly that one does not exist yet. It should.
Terms of Service
Expected of a platform taking payment from students; the footer links were removed rather than left pointing nowhere.
Verified student and parent stories
Published only when they meet the standard set out in the Trust Center — named, permissioned, including what did not work, and not cherry-picked.
Parent visibility
There is no parent login or automated report today. Any version of this has to respect that the student owns their preparation.
Directions we are actively working on
Work with real artefacts in the repository behind it, which is why it appears here rather than under "exploring".
Adaptive Verification
Distinguishing a confidently correct diagnosis from an uncertain one, so the platform can say how sure it is rather than only what it concluded. Phase 0 is applied; the architecture is documented.
Deeper failure observability
Continuing the work that turns a student-visible AI error into a diagnosable event, so reliability problems are found before they are reported.
Taxonomy depth
The 33-skill taxonomy is deliberately conservative. Unmapped detections are already logged, which is how we learn where it is too coarse.
Sustainable unit economics
The cost and coverage analytics exist so that the free tier can stay genuinely useful and prices can stay reasonable as the platform grows.
Directions we are considering
Genuinely undecided. Listed so the direction is visible, not so it can be counted on.
More Arabic and Franco depth
Language support exists; there is more to do in making explanation quality equal across all three.
Wider exam coverage within mathematics
Only where we can bring genuine specialist depth. Breadth without expertise would contradict the reason the platform works.
Support for schools and tutoring centres
Used by them already, without tooling built for them.
Published outcome evidence
The honest long-term goal of this Evidence Center: to be able to add an "outcome" evidence type alongside mechanism, research and record — with real data, properly collected.
What we will deliberately not build
- Other subjects — Not until we can bring the same depth of genuine subject expertise. Breadth without expertise contradicts the reason this works.
- An answer machine — The scope guard is a design decision, not a limitation to be removed later.
- Personalization by "learning style" — The evidence does not support the practice. See the research foundations.
- Score guarantees — No honest provider can offer one, and we will not publish improvement averages we cannot verify.
Holding us to it — the first group above is drawn from the gaps stated on the Trust Center. If they are still listed there in a year, that is a fair thing to ask about. What has already shipped