Rating, XP, difficulty and weekly rank answer different questions. These are the rules the app and backend are expected to share.
Rating starts at 1000, never falls below 100 and uses K factors 32, 24 and 16. Solved result, challenge difficulty and completion time affect each change. There is no human opponent, and rating does not drive adaptive difficulty.
A solved challenge grants each inferred skill exactly one award: 60/100/140/180/220 XP for difficulty 1–5. Each of ten skills has levels 1–5; the next threshold is round(100 × level^1.5), and level 5 displays 100% complete.
Difficulty uses recent result and relevant skill progress to choose a suitable next step. It neither grants a second XP award nor reads the Elo-inspired rating.
The weekly board counts solutions finalized by the server from Monday UTC. Only the canonical solved_count field determines public position; client-reported progress never updates it.
The first valid practice event in a local calendar day can advance the streak once. Achievements recognize reachable skill, challenge and consistency milestones; their legacy xpReward field is not a spendable or applied reward.
Core progress is persisted and synced best-effort for signed-in users. Local state can remain available during an outage, so a pending sync is not a promise of perfect real-time replication.
