Earning Rewards

Per-task formula

reward = reward_weight(model)
       × completed_count
       × { 0 if runtime == "echo", else 1 }

claimable_total = sum(reward over all of wallet's nodes)
                + floor(api_request_count / 10)

Each completed task gives a fractional accrual based on the model that served it. A wallet can run multiple nodes (browser + native); their accruals are summed at claim time.

Model multipliers

Heavier models give more reward per task because they cost more compute to run. Current weights:

llama-3.2-1b   → 0.6 pts/task
qwen-2.5-1.5b  → 0.7
gemma-2-2b     → 0.8
llama-3.2-3b   → 1.0  (baseline)
phi-3.5        → 1.0
qwen-2.5-3b    → 1.1
hermes-3-8b    → 1.6
mistral-7b     → 1.8

Anti-sybil

Spend pending points as credits

Convert pending points to API credits at 1 pt = 10 credits. Run a node for a few hours, then test your own AI app on those credits without paying. Converting resets the pending counter but never touches your lifetime score.

• POST /api/credits/topup with your session cookie
• Server zeroes out provider counters that fed the conversion
• Credits land in your balance immediately — no on-chain wait

Season 0 & the leaderboard

Alongside the pending balance, every point you earn is added to your lifetime score — it never decreases, not even when you convert pending points to credits. Lifetime score ranks every wallet on the public leaderboard (/app → Leaderboard or GET /v1/leaderboard).

• Pending points — convertible to credits, resets on conversion
• Lifetime points — Season 0 score, feeds the leaderboard
• Top ranks are the natural allowlist for future reward programs