Review and the badge
Platform checks, operator review, review statuses and what the "Tested on listed cases" badge does and does not mean.
Review is how a version earns the "Tested on listed cases" badge. The platform runs its own checks on the fixed version, you pin one of your successful tests, and a separate operator decides. Review is optional: any tested version can be published without it and is then shown to buyers as "Creator test only". This page explains the checks, the operator's role, the review statuses and what the badge covers. It is for creators preparing a version for the marketplace.
Where review stands today#
| Agent | Platform checks | Can earn the badge |
|---|---|---|
| Template agents (Token Researcher, Transaction Inspector, Deployer Investigator, Wallet Analyst, Solana Transaction Inspector) | Platform-selected synthetic cases for the kind | Yes |
| Workflows (builder, template duplicate, assistant, Record & replay) | The workflow platform checks | Not yet, see below |
| Watchtower | Not reviewed in Studio; runs through monitors | No |
Review a template agent#
Open Review for the version (from the test page's Continue to review, the Versions table's Review button, or /studio/review). The page states "Approval applies to this fixed version only." and has two parts.
1. Platform checks#
Choose Run platform checks. The platform runs its own synthetic cases against this exact version; your own example cases never replace or skip them. The cases depend on the kind:
| Kind | Synthetic cases exercise |
|---|---|
| Token Researcher | Token metadata, non-standard decimals, proxy and owner gaps, missing contracts, provider failures, the wrong chain and a changing snapshot. |
| Transaction Inspector | Unlimited approvals, unknown selectors, revert reasons, native transfers, contract creation, provider failures and the wrong chain. |
| Deployer Investigator | Direct and factory deployments, unindexed creations, infrastructure funders, truncated history, provider failures and the wrong chain. Ownership must never be inferred. |
| Wallet Analyst | Balances and transfers, assets moving both ways in one transaction, self-transfers, truncated history, missing token metadata, provider failures and the wrong chain. Trades and results must never be claimed. |
| Solana Transaction Inspector | Legacy and v0 payloads, lookup tables, unknown programs, failed simulations, read limits, the wrong cluster and provider failures. |
Each run is listed as "N/M cases · passed" or "failed"; choose Inspect to see every case. The checks use fixtures: "They do not establish live coverage."
2. Your test#
Choose a completed private test of this exact version under Creator test to submit. Inspect creator report shows it. If the test is partial, tick "I have reviewed the coverage gaps and want this partial report included in the operator review." Then choose Submit for operator review. The confirmation reads "Submitted for operator review. This version and its evidence are fixed."
Review a workflow#
The workflow review page has the same two parts:
- Workflow platform checks (Run workflow platform checks): "The platform runs each declared sample input against fixture chain data and checks the result contract, the permission summary, declared tools and chains against recorded coverage, and that findings come from tool output. AI analysis is checked without calling a model and needs a per-run cost cap; custom transforms run only in the isolated sandbox. Name-only or prompt-only workflows fail."
- Your creator test: choose a succeeded test of this version, then Submit for operator review.
The checks are listed by name when they fail:
| Check | Fails when |
|---|---|
manifest-and-registry | A tool is not approved, not on a declared chain, or used with unknown settings or output fields. |
deterministic-capabilities | AI analysis has no per-run cost cap, or a step cannot run on this server. |
tool-bound-findings | The workflow runs no tool, or no required finding is bound to tool output. |
permission-summary | A tool is granted but never used, or the workflow uses web reads, secret references or connectors, which the fixtures cannot evaluate. |
declared-tool-coverage | A tool has no verified coverage on a declared chain. |
declared-sample-inputs | The version declares no sample inputs (see above). |
declared-sample-input:(id) | One declared sample input does not produce a complete report that meets the result contract. |
What the operator reviews#
A separate operator (never the version's owner) reviews the fixed version together with the pinned evidence: the passing platform checks and your submitted test, with the test's stored reads. The operator records one of four decisions, each with a written reason: approve, reject, pause or revoke. "Platform cases are fixtures; approval is not an audit or a claim that anything is safe."
Your submission is pinned. Once submitted, a version cannot swap its evidence: "This immutable version already has a pinned review; create a new version for changed evidence."
Review statuses#
| Status | Buyers see | What it means for you |
|---|---|---|
| Not submitted | "Creator test only" | You can publish with Publish without review. |
| Pending | "Creator test only · review pending" | Publish now as "Creator test only", or wait for approval. |
| Approved | "Operator reviewed" | Publishing adds the "Tested on listed cases" badge. If the version is already listed, Publish to the marketplace on its card adds the badge. |
| Rejected | "Rejected in operator review" | The version cannot take new purchases. |
| Paused | "Paused by operator review" | The version cannot take new purchases. |
| Revoked | "Revoked by operator review" | The version cannot take new purchases. |
On Publish, each version's card says where its review stands, for example "Operator review rejected this version, so it cannot take new purchases." Fix the cause in the draft and submit a new version.
What the badge means#
"Tested on listed cases" means that this exact version passed the platform test cases listed on its page and that an operator reviewed one saved creator test. "The badge covers the listed platform checks and your submitted test only." It is not an audit, a security score, a verified-safe claim or an investment-safety claim. On a listing, the Testing tab shows the platform cases and the review scope: "Operator review: platform test cases plus a separate saved creator test. Not an audit, security score or investment-safety claim." See Badges and review.
Limits and special cases#
- You can run platform checks at most ten times in any 24 hours: "The daily limit of ten immutable-version fixture evaluations is reached."
- Recorded agents need two passing replays on unseen inputs before they can be submitted.
- A paused, revoked or deprecated version cannot be checked again.
- The operator cannot review their own version: "An operator cannot review their own version."