referee-psi.vercel.app
A double-blind peer-review room where the page is the boundary between a browser AI agent and untrusted manuscript text. All manuscripts are fictional.
First seen 2026-11-05 · Last seen 2026-11-05 · Source: webmcp.com
Answer 3 · Act 3 · Transact 1 · Score: Not yet scored
Tool inspector
check_claim
Test a quote against the manuscript source WITHOUT recording anything. Use it when you are unsure a passage is verbatim, when an assert_finding was refused, or before you put a claim in your summary to the human. It applies the same verification the evidence gate applies, so a quote that passes here will pass there. It operates only on the public sections: it cannot confirm or deny anything about the nine blinded classes, and asking it to returns SECTION_NOT_FOUND on exactly the path an invented section name takes. It returns `result` as one of SUPPORTED, NOT_SUPPORTED or INDETERMINATE. NOT_SUPPORTED is information, not an error — it means the source does not support that wording, so change the wording rather than retrying it. INDETERMINATE means the check could not be completed and says nothing about the quote. It returns no offset and no score on any result, so it cannot be used to locate source text — do not probe.
Input schema
| Property | Type | Required | Description |
|---|---|---|---|
manuscript_id |
string | yes | — |
section |
string | yes | — |
evidence_quote |
string | yes | The passage you believe is verbatim. Under 40 normalized chars is refused. |
claim |
string | no | Optional context: what you intend to conclude. Recorded in the ledger, not scored. |
Raw JSON schema
{
"type": "object",
"properties": {
"manuscript_id": {
"type": "string",
"enum": [
"MS-101",
"MS-102",
"MS-103",
"MS-104",
"MS-105",
"MS-106",
"MS-107",
"MS-108",
"MS-109",
"MS-110",
"MS-111",
"MS-112"
]
},
"section": {
"type": "string",
"enum": [
"abstract",
"introduction",
"related_work",
"methods",
"results",
"discussion",
"limitations",
"data_availability"
]
},
"evidence_quote": {
"type": "string",
"minLength": 1,
"maxLength": 1200,
"description": "The passage you believe is verbatim. Under 40 normalized chars is refused."
},
"claim": {
"type": "string",
"minLength": 10,
"maxLength": 600,
"description": "Optional context: what you intend to conclude. Recorded in the ledger, not scored."
}
},
"required": [
"manuscript_id",
"section",
"evidence_quote"
],
"additionalProperties": false
}
Similar websites
pocketgull.app
Pocket Gull is a real-time clinical AI consult engine powered by Google Gemini. Features 10D multi-paradigm strategy, FHIR R4 export & 3D spatial anatomy.
webmcp-six-lac.vercel.app
AI Navigation Layer: Model Context Protocol (MCP) Semantic DOM integration for intelligent, safe, and observable web navigation.
ghqsandbox.net
Interactive math and physics tools an AI agent can drive directly through WebMCP.
app.formative.com
Formative classroom assessment and learning platform