Provider path: register, stake, jobs, earnings — and the slash rules · content/guides/selling-compute.md

Selling Compute — turn your rig into a QALS earner

The provider path: register, stake, take jobs, get paid — with the trust mechanics (escrow, slash, reputation) doing the policing so you don't have to trust anyone.

✓ verified 2026-09-12 — the full provider lifecycle passed live: qmarket/test_market_v2.sh v2 SCENARIO PASS (register with stake → job under escrow → dispute → 50/50 slash → delist below rep 40 → second provider's reputation recovery on clean jobs). Marketplace :8824 green; live /providers shows real state (minirig-inference, rep 101, jobs_done 1, 4c earned).

What you need

• A machine with compute to sell (local GPU rig, or any remote box — RunPod-style included).
• A loopd identity with a stake — skin in the game, a bond you can lose for cheating.
• qmarket running: python3 qmarket/qmarket.py (:8824), provider agent: python3 qmarket/provider.py run (:8850).

The model in four lines

1. Register with name, job kind, price, and a stake (bond).
2. A buyer's job payment sits in escrow (a ledger hold) while your machine works.
3. Your output hash is attested and anchored; the metered amount settles against the hold at your price — 0% marketplace fee (founder policy, 2026-09-12; the token spread is the margin).
4. Cheat (faked output), and a dispute triggers verification — recompute vs anchor — then a 50/50 slash of your stake, reputation −25, and delisting below rep 40. Honest jobs build reputation back.

Steps

1. Price your kind first (the bench oracle)

python3 qmarket/bench.py        # measure, don't guess
# reference ladder: A100 272c/h · 4090 110c/h · L4 69c/h

2. Register with a stake

python3 qmarket/provider.py register --name my-rig --kind gpu-render \
    --price 110 --stake 2000 --description "4090 box, night hours"

Your stake is a loopd agent you own — the market can slash it, nothing else. What success looks like: you appear in GET :8824/providers with your price, reputation 100, stake listed.

3. Take jobs

Start the provider agent; it polls the market and executes jobs it's registered for (local adapter, or remote automation — the RunPod/external-rig pattern maps onto the same provider loop, artifacts moving via object storage, output hash anchored at settle).

python3 qmarket/provider.py run        # serves the adapter on :8850

Buyers post jobs; the flow on their side is hold → execute → settle → anchor. From the provider side you see the job, run it, return output; the escrowed money settles at metered × your price.

4. Watch earnings land

curl -s http://127.0.0.1:8824/earnings/<provider-id>   # earnings accrue AU$1:1 in credit

Earnings are plain loopd credit — spendable in chat, redeemable for services, capless like any wallet balance.

What success looks like

• You in /providers: {"name":"my-rig","reputation":100,"stake_cents":2000,…}.
• A job: hold placed on the buyer (you can see it in their feed as a 115% pre-auth), your machine runs, settle receipt r_… at metered cents, output hash anchored on Qalnet.
• The proof suite's whole arc passing: bash qmarket/test_market_v2.sh (30 assertions) — including the cheating provider getting slashed twice and delisted.

Pricing your reputation (the rules that protect buyers — and honest sellers)

Troubleshooting

Companions: sell-compute funnel (the dedicated seller page — pro and home-rig tracks + earnings math) · P2P compute market (the full design) · compute market v2 · competitors 2026 · how it works §5.