How Arbiter sources data, what runs on your machine, what reaches our server, and where we draw hard lines. This is our operating posture and a set of commitments — not legal advice.
Arbiter does not automate the game. No botting, no auto-trading, no clicking, no input injection, no reading game memory. Every trade is read by you and executed by you, by hand. The tool does math; you make the decision and do the trade. This is the operative line under Grinding Gear Games' Terms of Service, and we hold it deliberately.
The client reads the trade in front of you — optical character recognition of your own screen, the same pixels you are already looking at — and sends that read to our server, which does the gold-and-currency math against the freshest market picture it holds and answers with the verdict.
We compute the verdict server-side, and we are explicit about why:
The consequence, which we own rather than hide: Arbiter is online-only. If the server is unreachable you get a clear error and a retry — never a stale answer dressed up as a fresh one.
OCR of your screen is equivalent to you reading your own screen. We do not hook the game, inspect its process, or read its memory.
One narrow, deliberate exception on files: if you switch on map auto-tracking for the Loot
Valuer, the client follows the game's plain-text event log (Client.txt) — the file the game
writes precisely so companion tools can follow it — to notice when you return to your hideout. Read-only,
that one file, only when you enable the feature, and nothing from it leaves your machine.
A broad market scan (movers, spreads, cross-hub mispricing, mistrade watchlist) computed on our server from market data and surfaced in the dashboard. Discovery points you at where to look; the verdict on your own read decides whether to act.
The client talks to the server through exactly one module, and the upload set is pinned by an automated test that fails the build if anything new tries to cross. What may go up, always to your own account, only to answer your own requests:
Your reads, observations, and ledger are stored under your account and used to serve you. We do not sell them, share them, or pool them. If a shared "crowd" signal ever ships, it will be strictly opt-in and structural — a market plus a coarse movement bucket, carrying no price, no direction, no size — with a server-side validator that rejects anything richer than that schema. Your edge is yours.
Public market-data APIs, polled politely by a single ingest worker, well under published rate limits, with backoff and retry. Our database is the cache — one fetch serves all users; we do not fan out duplicate requests per user. We do not scrape the website, drive the trade UI, or circumvent any access control or rate limit. Where official API programs exist we pursue sanctioned access and operate within its granted terms.
The radar is offered on the standard market-data pattern: Free sees the previous snapshot; Pro sees the latest. At serve time, Free requests are answered with the snapshot before the newest and Pro requests with the newest. The data age you see is always computed from our true fetch time, never a lagging upstream timestamp, so a snapshot's age is exactly what it says.
Client.txt
event log, read-only, for the map auto-tracking you switch on.)Ask in the founding Discord or write to the contact address on the front page.