Ledger and analytics

Every dollar in Meris is traceable to a row in Postgres, and every number on the Network page is computed from those rows.

Money in

AI labs buy prepaid credit. Each top-up is a row in credits. A lab's balance is the sum of its credits minus what it has spent.

Crediting contributors

Every minute, the server credits each GPU that is connected, serving and not stepped aside, for the time since the last pass:

1. Inserts jobs rows for the work: GPU, duration, price, contributor share and Meris fee.
2. Adds the 70% share to the contributor's earned_usd and to the worker's totals.
3. Adds the work to the current hour in network_hours.

All three use the same amounts, so balances, per-GPU totals and network totals agree.

Money out

A withdrawal locks the contributor's row, computes earned − paid − pending, checks the $5 minimum and inserts a payouts row as pending, for at most $50. Meris sends the stablecoin on the account's rail and records the transaction hash, which marks it paid. A payout that cannot be sent is marked rejected and its amount returns to the available balance.

Analytics

The Network page in the app and the public GET /api/network endpoint are built from:

Figure	Source
GPUs online, VRAM lent, jobs running	The live registry of connected workers
Paid to contributors, jobs	Sums of network_hours
Cumulative charts	network_hours, accumulated hour by hour
GPUs online over time	network_samples, one sample every five minutes
Live feed	The 20 latest completed jobs. Workers appear as a six-character hash, never by name or owner.
GPU types, regions and the globe	Paired workers; regions come from each worker's reported time zone