Design-partner council¶
Orchid uses a small customer council to pressure-test product decisions before they become roadmap commitments. These are simulated design-partner roles for product discovery and review. They do not represent, speak for, or imply a relationship with any named company.
The council¶
| Role | Job to be done | Veto questions |
|---|---|---|
| Learning-program owner | Improve retained mastery without giving up the authored curriculum | Is the primary outcome independent and delayed? Is there a static control? |
| Learning designer/instructor | Keep pedagogical control and understand every adaptation | Can required units be locked, recommendations explained, and the authored path restored? |
| Platform engineer | Embed Orchid safely in an existing learning product | Are decisions/outcomes durable and idempotent? Can any learner path be replayed? |
| Learner advocate | Protect the learner experience | Does adaptation avoid repeated failure, excessive repetition, inaccessible content, or unexplained difficulty jumps? |
| Media-learning observer | Test future audio/video-learning extensions without diluting the product | Is this assessed, structured learning—or merely engagement optimization? |
The last role can be informed by a hypothetical Spotify- or YouTube-like platform. It must not be used to imply that those companies are customers or partners.
Review every proposed feature¶
Before accepting a feature, each council role answers these questions:
- Learning outcome: What delayed, independent mastery outcome could this improve? Immediate correctness and watch time are not enough.
- Curriculum control: Which eligibility, prerequisite, assessment-holdout, and override rules remain outside the model?
- Data contract: What exact event, catalog version, candidate set, decision ID, and outcome join are required?
- Failure mode: How does the learner fall back to an authored path if data is sparse, content is new, or the service is unavailable?
- Evidence: What static control and randomized experiment would make the result believable?
Reject features that cannot answer all five. Put general engagement ranking, unbounded content retrieval, and advanced policy learning in a separate future track unless they meet the same learning-evidence bar.
Current council decisions¶
Target customer¶
Start with a professional-certification, technical-skills, test-preparation, or other assessed-practice product. It should have a stable exercise bank, a skill map, several valid next exercises, automated scoring, and a delayed assessment.
Not the first market¶
Do not target a home feed, music radio, podcast discovery, or general video recommendation. Those systems optimize different objectives and need retrieval, implicit-feedback debiasing, multi-objective ranking, and real-time experiment infrastructure beyond Orchid's adaptive-practice scope.
Audio/video learning is in scope only when it is a bounded course with scored practice or checkpoints. A video view or audio completion is context, not proof of mastery.
Delivery gates¶
The council will not approve a production pilot until Orchid has:
- A versioned curriculum catalog with course/module, assessment-holdout, skill, difficulty, and prerequisite diagnostics.
- Durable, idempotent decision/outcome storage; deterministic replay; and a visible authored-path fallback/rollback.
- A reference event/service integration and recommendation explanations for learning designers.
- A pre-registered learner-level experiment against the current authored path, with delayed independent mastery as the primary outcome.
These delivery gates refine the product roadmap; they do not replace the existing data-readiness or learning-pilot requirements.