Can I run it myself — do I need keys? +
Yes, and no keys. On a fresh clone, npm run grade grades the bundled hero mesh against the public sample profile sample-game-prop@1 and prints a real per-predicate verdict — with zero keys. Keys are only needed to sign a verdict and to settle on-chain. The verification is the part you can check yourself.
Do I have to use your acceptance thresholds? +
No — bring your own. Pass any AcceptanceProfile / GradePlan JSON via the CLI --profile flag, or inline in the HTTP /grade body (no registration). Our tuned thresholds are never required to run the grader; they're the moat, gitignored.
What does the grader actually check? +
A deterministic floor of seven checks — triangle budget, watertight/manifold, valid GLB, UVs present, texture-size budget, file-size budget, and named editable parameters — plus an opt-in semantic (render + CLIP) gate that confirms it's the right asset. Every settling check is reproducible code; same asset + profile ⇒ same verdict and the same predicateDigest.
Which chain does settlement run on? +
Testnet only, on three chains across two rails: BNB Chain and Base Sepolia (EVM, EIP-191) and a Solana devnet program (ed25519). The same canonical signed verdict settles on any of them — verification produces one verdict, settlement is pluggable per chain. No mainnet.
Where can I actually use Grain? +
Three real surfaces over one engine: grain-mcp (any agent — Cursor, Claude Code), a "Grade with Grain" Blender add-on (the artist's loop), and a Minds Grader Skill (the Mind's loop, live on both bazaars) — plus two original, credited methodology Skills that design the profile and author a passing asset.
Is it just a linear request → grade → settle loop? +
No. A Mind-supervised swarm of role companions (3D Artist, Technical Artist, Programmer) each verify + settle their slice, and the verified asset only enters the scene behind a causal hash gate — runnable in-browser. See the demo → A Godot grounding demo even loads the same GLB into a real engine only if the verdict passes and the bytes' hash matches.
What stops a compromised grader from stealing funds? +
The contract. The grader key's only power is one binary verdict on a pre-funded order; release and refund go only to the buyer or seller recorded at funding. No owner, setter, or withdraw. A compromised key can mis-grade, never exfiltrate — and the property holds on both the EVM escrow and the Solana program.
Isn't a 3D validator commodity? Where's the moat? +
The predicate code is commodity — and open. The defensible asset is the acceptance-profile data: what real buyers (TADs, QA leads, marketplace inspectors) actually accept. Accumulated from real rejection data, not scrapeable, and mapped to willingness-to-pay. And 3D is just the wedge — the substrate (escrow, custody boundary, signed verdict, profiles, Skill) is vertical-agnostic.
If everyone vibe-codes from AI, doesn't verification homogenize the art out of games? +
No — that fear is backwards. What flattens games into formulaic, borrowed-from-the-past output isn't a check; it's the conformance grind — the triangle budget, the manifold, the valid GLB, the re-export, the QA pass — eating the hours that should go to art. Grain is not a taste engine, an art judge, or a generator: it grades only the commodity floor and signs a verdict that's identical no matter how distinctive the work is. It never rules on whether a thing is beautiful, original, or good. The spec is the floor, not the ceiling — Grain clears the grind so the artistry has room.
Built on Minds by Animoca Brands
