docs: bootstrap tutor planning

.planning/ROADMAP.md

@@ -0,0 +1,98 @@
+# Roadmap: Tutor Platform
+
+## Milestone 1: Job-Seeker Interview Tutor MVP
+
+### Phase 1: Go Backend Foundation and Workflow Boundary
+
+**Goal:** Establish the Go service skeleton and typed workflow boundary for
+internalized `agent-farm-go` patterns.
+
+**Requirements:** BACK-01, BACK-02, BACK-03, BACK-04, BACK-05
+
+**Success Criteria:**
+
+- Go backend scaffold exists with clear module boundaries.
+- No manually authored source file exceeds 600 lines.
+- Workflow interfaces are typed and isolated from HTTP handlers.
+- Runtime config can identify the `third-one` / `deepseek-v4-flash` target.
+- Basic build/test command is documented in `AGENTS.md`.
+
+### Phase 2: Diagnostic Interview Loop
+
+**Goal:** Prove the first job-seeker loop from role selection through graded
+diagnostic interview.
+
+**Requirements:** INT-01, INT-02, INT-03, INT-04, INT-05, INT-06
+
+**Success Criteria:**
+
+- User can choose target role and stack.
+- Backend can create a diagnostic session.
+- System produces role-specific interview questions.
+- User answers are graded through typed workflow results.
+- Grading evidence and original answer are persisted.
+
+### Phase 3: Learner Memory
+
+**Goal:** Convert graded answer evidence into structured learner memory.
+
+**Requirements:** MEM-01, MEM-02, MEM-03, MEM-04, MEM-05
+
+**Success Criteria:**
+
+- Learner profile is persisted.
+- Concept mastery updates require evidence.
+- Misconceptions link to supporting answers.
+- Session context and durable memory remain separate.
+- Memory extraction workflow emits typed candidates.
+
+### Phase 4: Progression and Gamified Learning Routine
+
+**Goal:** Make readiness and next challenge visible without empty rewards.
+
+**Requirements:** PROG-01, PROG-02, PROG-03, PROG-04, PROG-05
+
+**Success Criteria:**
+
+- Readiness map displays concept states.
+- Challenge ladder exists for the first backend interview track.
+- Next challenge is selected from learner memory and grading evidence.
+- Boss question unlocks after prerequisite stability.
+- Streak/reward behavior avoids punitive and random-reward mechanics.
+
+### Phase 5: Source-Backed Ontology Builder
+
+**Goal:** Start material ingestion and ontology candidate generation.
+
+**Requirements:** ONTO-01, ONTO-02, ONTO-03, ONTO-04
+
+**Success Criteria:**
+
+- User/operator can add source material.
+- Concepts, prerequisites, rubrics, and question candidates carry provenance.
+- Missing prerequisites and weak areas are flagged.
+- Generated/inferred content is not promoted as canonical automatically.
+
+### Phase 6: Visual Teaching Asset Pipeline
+
+**Goal:** Generate reviewable teaching asset candidates from ontology concepts.
+
+**Requirements:** ASSET-01, ASSET-02, ASSET-03
+
+**Success Criteria:**
+
+- Asset prompt generation contract exists.
+- Generated assets store prompt lineage, source concept, source evidence, model
+  config, and review state.
+- Actual image model identifier is verified before production image calls.
+
+## Parking Lot
+
+- General student mode.
+- Teacher/parent dashboards.
+- School tenant administration.
+- Company-specific interview packs.
+- Human ontology review console.
+
+---
+*Roadmap created: 2026-04-26 after initial product planning and Go backend decision.*