@kindgi/api 0.1.3 → 0.1.4-rc.1

package/dist/openapi/schemas.js

@@ -168,6 +168,16 @@ export const RunSchema = {
             description: 'Set on a child run: the node in the parent run that started it.',
         },
         agent: { $ref: '#/components/schemas/RunAgent' },
+        replayOf: {
+            type: 'string',
+            format: 'uuid',
+            description: 'Set on a replay run (an eval run re-running a past run): the run it replays.',
+        },
+        evalRunId: {
+            type: 'string',
+            description: 'Set on a replay run: the eval run that started it.',
+        },
+        versions: { $ref: '#/components/schemas/FlowVersionOverrides' },
         publicAccessToken: {
             type: 'string',
             description: 'Only in the response to `POST /v1/runs`, when the deployment issues public run tokens: a read-only token for this run (and its descendants) to hand to a browser, for `GET /v1/runs/{runId}/progress` and its stream.',
@@ -839,8 +849,20 @@ export const AgentSchema = {
         version: { type: 'string', description: 'Semver.' },
         name: { type: 'string' },
         description: { type: 'string' },
-        instructions: { type: 'string' },
+        instructions: {
+            oneOf: [{ type: 'string', minLength: 1 }, { $ref: '#/components/schemas/PromptRef' }],
+            description: 'The system prompt (a Liquid template), or a prompt block by range whose template and parameters are used instead (pinned at publish, `pins.prompts`).',
+        },
         parameters: { type: 'array', items: { $ref: '#/components/schemas/PromptParameter' } },
+        settings: {
+            type: 'array',
+            items: { $ref: '#/components/schemas/BlockRef' },
+            description: 'Settings blocks the agent reads, by range: tools read them as `ToolContext.settings[\'<id>\']`, templates as `settings["<id>"]`. Pinned at publish (`pins.settings`).',
+        },
+        modelSettings: {
+            $ref: '#/components/schemas/BlockRef',
+            description: "A model-settings block by range: its `temperature` and `maxOutputTokens` go into the turn's model calls. Pinned at publish.",
+        },
         capabilities: { type: 'array', items: { $ref: '#/components/schemas/Capability' } },
         tools: { type: 'array', items: { $ref: '#/components/schemas/ToolRef' } },
         retrieval: { type: 'array', items: { $ref: '#/components/schemas/RetrievalIntent' } },
@@ -860,6 +882,94 @@ export const AgentSchema = {
         tags: { type: 'array', items: { type: 'string' } },
         output: { $ref: '#/components/schemas/AgentOutputSpec' },
         toolErrors: { $ref: '#/components/schemas/ToolErrorsSpec' },
+        pins: { $ref: '#/components/schemas/AgentPins' },
+        derivedFrom: { $ref: '#/components/schemas/VersionDerivation' },
+        unregisteredAt: {
+            type: 'string',
+            format: 'date-time',
+            description: 'Present only on an unregistered version (`GET …/versions/{version}` reads those too). Unregister stops a version being chosen, not the pins that hold it: a new run naming it is refused, while a resumed run and a published version that pins it still run it.',
+        },
+        pinsDigest: {
+            type: 'string',
+            pattern: '^sha256:[0-9a-f]{64}$',
+            description: "Set by the runtime with `pins`: `sha256:<hex>` of the pins' canonical JSON (sorted keys, no whitespace). Two agent versions with the same digest run the same blocks.",
+        },
+    },
+};
+export const VersionDerivationSchema = {
+    description: "Set by the runtime on an agent or flow version a deploy registered in place of the definition's version, which was registered already with other pins or content (versions never change). Never in the publish body.",
+    type: 'object',
+    additionalProperties: false,
+    required: ['version', 'reason'],
+    properties: {
+        version: { type: 'string', description: 'The version the definition names.' },
+        reason: {
+            type: 'string',
+            enum: ['pins-changed', 'unpinned', 'version-taken', 'edited'],
+            description: "`pins-changed`: a block it uses has a new version; `unpinned`: the definition's version was published before pins existed; `version-taken`: the definition's version holds another definition; `edited`: derived from `version` with some data-block pins swapped (`POST /v1/agents/{agentId}/versions`).",
+        },
+        label: { type: 'string', description: 'For `edited`: a short label for the version.' },
+        by: { type: 'string', description: 'For `edited`: who derived it (`user:<id>`).' },
+    },
+};
+export const DeriveAgentVersionBodySchema = {
+    description: "Derive a new agent version from a pinned one with some data-block pins swapped: an expert's edit reaching an agent with no code change.",
+    type: 'object',
+    additionalProperties: false,
+    required: ['from', 'pins'],
+    properties: {
+        from: { type: 'string', description: 'The version to derive from (it must be pinned).' },
+        pins: { $ref: '#/components/schemas/AgentPinSwaps' },
+        label: { type: 'string', description: 'A short label for the new version.' },
+        projectId: {
+            type: 'string',
+            format: 'uuid',
+            description: "The agent's project, when the runtime doesn't record it on the version.",
+        },
+    },
+};
+const PinMapSchema = {
+    type: 'object',
+    additionalProperties: { type: 'string' },
+};
+export const AgentPinSwapsSchema = {
+    description: 'The data-block pins to swap, by block id → exact version. Only blocks the version already references; tool pins come from code.',
+    type: 'object',
+    additionalProperties: false,
+    properties: {
+        prompts: { ...PinMapSchema, description: 'Prompt block id → exact version.' },
+        settings: { ...PinMapSchema, description: 'Settings block id → exact version.' },
+    },
+};
+export const AgentPinsSchema = {
+    description: 'The exact block versions an agent version runs: its lockfile. Set by the runtime when the version is published, never in the publish body: each tool range resolves once to the version every run of that agent version uses, so a new tool version reaches the agent only through a new agent version. Absent on a version published before pins existed (its ranges resolve per run).',
+    type: 'object',
+    additionalProperties: false,
+    required: ['tools', 'prompts', 'settings'],
+    properties: {
+        tools: { ...PinMapSchema, description: 'Tool id → exact version.' },
+        prompts: { ...PinMapSchema, description: 'Prompt block id → exact version.' },
+        settings: { ...PinMapSchema, description: 'Settings block id → exact version.' },
+    },
+};
+export const PromptRefSchema = {
+    description: "A prompt block an agent's instructions come from, by id and semver range.",
+    type: 'object',
+    additionalProperties: false,
+    required: ['prompt', 'version'],
+    properties: {
+        prompt: { type: 'string', description: 'The prompt block id.' },
+        version: { type: 'string', description: 'A semver range (`^1.0.0`, `1.2.0`).' },
+    },
+};
+export const BlockRefSchema = {
+    description: 'A settings block an agent reads, by id and semver range.',
+    type: 'object',
+    additionalProperties: false,
+    required: ['id', 'version'],
+    properties: {
+        id: { type: 'string', description: 'The settings block id.' },
+        version: { type: 'string', description: 'A semver range (`^1.0.0`, `1.2.0`).' },
     },
 };
 export const PublishAgentBodySchema = {
@@ -883,8 +993,20 @@ export const PublishAgentBodySchema = {
         name: { type: 'string' },
         description: { type: 'string' },
         projectId: ContentProjectIdProperty,
-        instructions: { type: 'string' },
+        instructions: {
+            oneOf: [{ type: 'string', minLength: 1 }, { $ref: '#/components/schemas/PromptRef' }],
+            description: 'The system prompt (a Liquid template), or a prompt block by range whose template and parameters are used instead (pinned at publish, `pins.prompts`).',
+        },
         parameters: { type: 'array', items: { $ref: '#/components/schemas/PromptParameter' } },
+        settings: {
+            type: 'array',
+            items: { $ref: '#/components/schemas/BlockRef' },
+            description: 'Settings blocks the agent reads, by range: tools read them as `ToolContext.settings[\'<id>\']`, templates as `settings["<id>"]`. Pinned at publish (`pins.settings`).',
+        },
+        modelSettings: {
+            $ref: '#/components/schemas/BlockRef',
+            description: "A model-settings block by range: its `temperature` and `maxOutputTokens` go into the turn's model calls. Pinned at publish.",
+        },
         capabilities: { type: 'array', items: { $ref: '#/components/schemas/Capability' } },
         tools: { type: 'array', items: { $ref: '#/components/schemas/ToolRef' } },
         retrieval: { type: 'array', items: { $ref: '#/components/schemas/RetrievalIntent' } },
@@ -1009,6 +1131,43 @@ export const FlowSchema = {
         edges: { type: 'array', items: { $ref: '#/components/schemas/FlowEdge' } },
         maxParallelism: { type: 'integer', minimum: 1 },
         metadata: { type: 'object', additionalProperties: true },
+        pins: { $ref: '#/components/schemas/FlowPins' },
+        pinsDigest: {
+            type: 'string',
+            pattern: '^sha256:[0-9a-f]{64}$',
+            description: "Set by the runtime with `pins`: `sha256:<hex>` of the pins' canonical JSON (sorted keys, no whitespace).",
+        },
+        derivedFrom: { $ref: '#/components/schemas/VersionDerivation' },
+        unregisteredAt: {
+            type: 'string',
+            format: 'date-time',
+            description: 'Present only on an unregistered version (`GET …/versions/{version}` reads those too). Unregister stops a version being chosen, not the pins that hold it: a new run naming it is refused, while a resumed run and a published version that pins it still run it.',
+        },
+    },
+};
+export const FlowPinsSchema = {
+    description: 'The exact tool and agent versions a flow version runs: its lockfile. Set by the runtime when the version is published, never in the publish body: each tool the flow runs, and each agent it runs at no named version, resolves once to its latest version then, which every run of that flow version uses. Absent on a version published before pins existed (it binds the latest versions per run).',
+    type: 'object',
+    additionalProperties: false,
+    required: ['tools', 'agents'],
+    properties: {
+        tools: { ...PinMapSchema, description: 'Tool id → exact version.' },
+        agents: {
+            ...PinMapSchema,
+            description: 'Agent id → exact version, for agent nodes that name no version.',
+        },
+    },
+};
+export const FlowVersionOverridesSchema = {
+    description: "Agents and tools a flow runs at other exact versions than the flow version's pins (\"this flow, with `acme.scorer` at 0.4.0\"), without publishing a new flow version: a comparison's flow candidate (`versions` on the start body and the run's `comparison`), and the flow runs that replay it (a run's `versions`). Each id must be an agent or tool the flow uses.",
+    type: 'object',
+    additionalProperties: false,
+    properties: {
+        tools: { ...PinMapSchema, description: 'Tool id → exact version.' },
+        agents: {
+            ...PinMapSchema,
+            description: 'Agent id → exact version, for agent nodes with or without a version of their own.',
+        },
     },
 };
 export const PublishFlowBodySchema = {
@@ -1708,190 +1867,620 @@ export const ObservationCollectionPageSchema = {
         hasMore: { type: 'boolean' },
     },
 };
-// ---------------- memory ----------------
-export const FactScopeSchema = {
-    type: 'object',
-    additionalProperties: true,
-    required: ['tenantId'],
-    description: 'Fact scope object. `tenantId` is required; every optional key narrows the fact (`userId`, `orgId`, `projectId`, `threadId`, `sessionId`). Additional keys accepted for forward compatibility.',
-    properties: {
-        tenantId: { type: 'string' },
-        userId: { type: 'string' },
-        orgId: { type: 'string' },
-        projectId: { type: 'string' },
-        threadId: { type: 'string' },
-        sessionId: { type: 'string' },
-    },
-};
-export const RetentionSchema = {
-    type: 'object',
-    additionalProperties: false,
-    properties: {
-        keepUntil: { type: 'string', format: 'date-time' },
-        keepDays: { type: 'integer', minimum: 1 },
-        legalHold: { type: 'boolean' },
-    },
-};
-export const SourceFreshnessSchema = {
-    type: 'object',
-    additionalProperties: false,
-    properties: {
-        ttlSeconds: { type: 'integer', minimum: 0 },
-        lastVerifiedAt: { type: 'string', format: 'date-time' },
-        etag: { type: 'string' },
-        sourceVersion: { type: 'string' },
-    },
-};
-export const SourceRefreshSchema = {
-    type: 'object',
-    additionalProperties: false,
-    required: ['strategy'],
-    properties: {
-        strategy: { type: 'string', enum: ['on-read', 'background', 'manual'] },
-        handler: { type: 'string' },
-        priority: { type: 'integer' },
-    },
-};
-export const FactSourceSchema = {
-    type: 'object',
-    additionalProperties: false,
-    required: ['kind', 'freshness', 'refresh'],
-    properties: {
-        kind: {
-            type: 'string',
-            enum: ['http-api', 'blob', 'mcp-tool', 'external-db', 'user-input'],
+// ---------------- judgments + judge classes ----------------
+export const JudgeClassScopeSchema = {
+    description: 'Where a judge class applies: the whole tenant, one project, or one agent in a project.',
+    oneOf: [
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['kind'],
+            properties: { kind: { type: 'string', const: 'tenant' } },
         },
-        uri: { type: 'string' },
-        freshness: { $ref: '#/components/schemas/SourceFreshness' },
-        refresh: { $ref: '#/components/schemas/SourceRefresh' },
-    },
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['kind', 'projectId'],
+            properties: {
+                kind: { type: 'string', const: 'project' },
+                projectId: { type: 'string', minLength: 1 },
+            },
+        },
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['kind', 'projectId', 'agentId'],
+            properties: {
+                kind: { type: 'string', const: 'agent' },
+                projectId: { type: 'string', minLength: 1 },
+                agentId: { type: 'string', minLength: 1 },
+            },
+        },
+    ],
 };
-export const FactSchema = {
+export const JudgeClassSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['id', 'type', 'scope', 'version', 'createdAt'],
+    required: ['id', 'tenantId', 'scope', 'name', 'weight', 'createdAt', 'updatedAt'],
     properties: {
-        id: { type: 'string', description: 'FactId.' },
-        type: {
+        id: { type: 'string' },
+        tenantId: { type: 'string' },
+        scope: { $ref: '#/components/schemas/JudgeClassScope' },
+        name: {
             type: 'string',
-            description: 'Fact type identifier (pack-defined; a few are framework-standard).',
+            description: 'The deployment\'s own word for the class: "expert", "user", "arbitrator".',
         },
-        scope: { $ref: '#/components/schemas/FactScope' },
-        version: {
-            type: 'integer',
-            minimum: 1,
-            description: 'Monotonic version within (scope, id). Supersession increments.',
+        weight: {
+            type: 'number',
+            minimum: 0,
+            description: 'How much a judgment of this class counts, relative to the others.',
         },
+        description: { type: 'string' },
         createdAt: { type: 'string', format: 'date-time' },
         updatedAt: { type: 'string', format: 'date-time' },
-        content: { description: 'Free-form structured payload.' },
-        contentRef: {
-            type: 'string',
-            description: '`blob://<provider>/<bucket>/<key>` when the payload is stored externally.',
-        },
-        contentHash: { type: 'string' },
-        size: { type: 'integer', minimum: 0 },
-        embeddingModel: { type: 'string' },
-        retention: { $ref: '#/components/schemas/Retention' },
-        source: { $ref: '#/components/schemas/FactSource' },
-        causedByLogId: { type: 'array', items: { type: 'string' } },
-        supersedes: {
+        unregisteredAt: {
             type: 'string',
-            description: 'FactId of the predecessor when this row supersedes another.',
+            format: 'date-time',
+            description: 'Set when the class was retired.',
         },
     },
 };
-export const FactCollectionPageSchema = {
+export const JudgeClassCollectionPageSchema = {
     type: 'object',
     additionalProperties: false,
     required: ['data', 'hasMore'],
     properties: {
-        data: { type: 'array', items: { $ref: '#/components/schemas/Fact' } },
-        nextCursor: {
-            type: 'string',
-            description: 'Opaque cursor for the next page. Absent when `hasMore: false`.',
-        },
+        data: { type: 'array', items: { $ref: '#/components/schemas/JudgeClass' } },
+        nextCursor: { type: 'string', description: 'Opaque cursor. Treat as opaque on the client.' },
         hasMore: { type: 'boolean' },
     },
 };
-export const WriteFactBodySchema = {
-    description: "Write a fact. `type` selects the retrieval-policy (which indexes populate); `scope.tenantId` MUST match the caller's tenant. Optional `retention` overrides tenant defaults; optional `contentHash` is a caller-supplied idempotence hint (runtime computes its own hash regardless).",
+export const CreateJudgeClassBodySchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['type', 'scope', 'content'],
+    required: ['scope', 'name', 'weight'],
     properties: {
-        type: { type: 'string', minLength: 1 },
-        scope: { $ref: '#/components/schemas/FactScope' },
-        content: { description: 'Free-form structured payload.' },
-        retention: { $ref: '#/components/schemas/Retention' },
-        contentHash: { type: 'string' },
+        scope: { $ref: '#/components/schemas/JudgeClassScope' },
+        name: { type: 'string', minLength: 1, maxLength: 100 },
+        weight: { type: 'number', minimum: 0 },
+        description: { type: 'string' },
     },
 };
-export const SupersedeFactResultSchema = {
+export const UpdateJudgeClassBodySchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['factId', 'superseded'],
+    minProperties: 1,
     properties: {
-        factId: { type: 'string' },
-        superseded: { type: 'boolean', const: true },
+        weight: { type: 'number', minimum: 0 },
+        description: { type: 'string' },
     },
 };
-export const RetrieveIntentSchema = {
-    description: 'Retrieval intent — mirrors `RetrievalIntent` from `@kindgi/agents`, widened for direct-HTTP use. `mode: "list"` returns a plain scoped list (no query). `mode: "keyword"` runs full-text search. `mode: "semantic"` runs vector similarity search — requires an embedding provider bound on the deployment; if unavailable, the route returns `400 bad-input`. `mode: "both"` unions keyword + semantic results, dedup by fact id.',
+export const UnregisterJudgeClassResultSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['mode'],
+    required: ['judgeClassId', 'unregistered'],
     properties: {
-        mode: { type: 'string', enum: ['list', 'keyword', 'semantic', 'both'] },
-        query: {
-            type: 'string',
-            description: 'Required for `keyword` / `semantic` / `both`; ignored for `list`.',
-        },
-        type: { type: 'string', minLength: 1 },
-        scope: { $ref: '#/components/schemas/FactScope' },
-        limit: { type: 'integer', minimum: 1 },
-        embeddingModel: { type: 'string' },
+        judgeClassId: { type: 'string' },
+        unregistered: { type: 'boolean', const: true },
     },
 };
-export const RetrieveMemoryBodySchema = RetrieveIntentSchema;
-export const RetrievalHitSchema = {
+export const JudgedSubjectSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['fact'],
+    required: ['kind', 'id', 'version'],
+    description: 'What a judged run ran: an agent at a version, or a flow at a version.',
     properties: {
-        fact: { $ref: '#/components/schemas/Fact' },
-        score: {
-            type: 'number',
-            description: 'Relevance score. Keyword mode returns an implementation-defined rank (higher = better). Semantic mode returns cosine similarity in [-1, 1] (higher = better). Absent for `list` mode.',
-        },
+        kind: { type: 'string', enum: ['agent', 'flow'] },
+        id: { type: 'string' },
+        version: { type: 'string' },
     },
 };
-export const RetrieveMemoryResultSchema = {
+export const JudgedItemSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['results'],
+    required: ['key'],
+    description: "The judged item of a run's output.",
     properties: {
-        results: { type: 'array', items: { $ref: '#/components/schemas/RetrievalHit' } },
+        key: { type: 'string', minLength: 1, description: "The caller's stable id for the item." },
+        pointer: {
+            type: 'string',
+            description: 'Where the item is in the run\'s output, as a JSON Pointer (RFC 6901), e.g. `/matches/2`. `""` is the whole output.',
+        },
+        rank: {
+            type: 'integer',
+            minimum: 0,
+            description: "The item's position in a ranked list (0 = first).",
+        },
     },
 };
-export const RevokeTokenResultSchema = {
+export const JudgmentAssertedBySchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['tokenId', 'revoked'],
+    required: ['kind', 'id'],
+    description: 'Who asserted a judgment: the authenticated caller, never a typed name.',
     properties: {
-        tokenId: { type: 'string', format: 'uuid' },
-        revoked: { type: 'boolean', const: true },
+        kind: { type: 'string', enum: ['user', 'service'] },
+        id: {
+            type: 'string',
+            description: 'A user id, or for a service token its token or session id.',
+        },
     },
 };
-// ---------------- signing keys (the deployment trust list) ----------------
-export const TrustedSigningKeySchema = {
+export const JudgmentSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['keyId', 'tenantId', 'algorithm', 'publicKey', 'createdAt'],
-    description: 'A public key the tenant trusts to sign deployments (`POST /v1/deployments`). A revoked key stays readable, with `revokedAt`, so the deployments it signed can be audited; it verifies no new ones.',
+    required: [
+        'id',
+        'tenantId',
+        'projectId',
+        'runId',
+        'subject',
+        'item',
+        'verdict',
+        'assertedBy',
+        'createdAt',
+    ],
     properties: {
-        keyId: { type: 'string', description: 'The signer key id a deployment envelope names.' },
-        tenantId: { type: 'string', format: 'uuid' },
+        id: { type: 'string' },
+        tenantId: { type: 'string' },
+        projectId: { type: 'string' },
+        runId: { type: 'string' },
+        subject: { $ref: '#/components/schemas/JudgedSubject' },
+        item: { $ref: '#/components/schemas/JudgedItem' },
+        verdict: { type: 'string', enum: ['yes', 'no'] },
+        reason: { type: 'string' },
+        judgeClassId: {
+            type: 'string',
+            description: 'The judge class the judgment is recorded under. Absent when unclassified (counts with weight 1).',
+        },
+        assertedBy: { $ref: '#/components/schemas/JudgmentAssertedBy' },
+        participantId: {
+            type: 'string',
+            description: "The app's opaque id for its end user who judged, when an app judged on their behalf.",
+        },
+        createdAt: { type: 'string', format: 'date-time' },
+        unregisteredAt: {
+            type: 'string',
+            format: 'date-time',
+            description: 'Set when the judgment was removed or superseded.',
+        },
+        supersededBy: { type: 'string', description: 'The judgment that replaced this one.' },
+    },
+};
+export const JudgedRunContextSchema = {
+    type: 'object',
+    additionalProperties: false,
+    description: 'What a judged run needs besides its input to be replayed, captured when it was first judged. For an agent turn: the conversation before it, what its retrievals returned, and the decision at its session approval gate. For a flow run: its tool calls with their results.',
+    properties: {
+        history: {
+            type: 'array',
+            items: {},
+            description: "The conversation's messages before the turn, oldest first (at most the last 200).",
+        },
+        historyTruncated: {
+            type: 'boolean',
+            description: 'Whether older messages were left out of `history`.',
+        },
+        retrieved: { description: "What the turn's retrievals returned." },
+        sessionApproval: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['approved'],
+            description: "The reviewer's decision at the turn's session approval gate, when the turn waited on one. A replay of the turn follows it.",
+            properties: {
+                approved: { type: 'boolean' },
+                rationale: { type: 'string', description: "The reviewer's reason for a rejection." },
+            },
+        },
+        flow: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['calls', 'steps'],
+            description: "For a flow run: what it did, kept at its first judgment so it can be replayed. Every tool call it made with its result (at its tool nodes, in its agent steps' turns and in its sub-flows), at most 500, and its agent steps.",
+            properties: {
+                calls: {
+                    type: 'array',
+                    items: {
+                        type: 'object',
+                        additionalProperties: false,
+                        required: ['runId', 'toolId'],
+                        properties: {
+                            runId: {
+                                type: 'string',
+                                description: "The run that made it: the flow run, a sub-flow's, or an agent step's turn.",
+                            },
+                            nodeId: {
+                                type: 'string',
+                                description: 'The tool node that made it, or the agent step whose turn did.',
+                            },
+                            scope: { type: 'string', description: 'The loop iteration, in a loop body.' },
+                            toolId: { type: 'string' },
+                            arguments: {},
+                            result: {},
+                        },
+                    },
+                },
+                steps: {
+                    type: 'array',
+                    items: {
+                        type: 'object',
+                        additionalProperties: false,
+                        required: ['runId', 'agentId', 'agentVersion'],
+                        properties: {
+                            runId: { type: 'string' },
+                            nodeId: { type: 'string' },
+                            scope: { type: 'string' },
+                            agentId: { type: 'string' },
+                            agentVersion: { type: 'string' },
+                            retrieved: { description: "What the step's turn retrieved." },
+                        },
+                    },
+                },
+                truncated: { type: 'boolean', description: 'More calls were made than were kept.' },
+            },
+        },
+    },
+};
+export const JudgedRunCopySchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['runId', 'subject', 'input', 'output', 'capturedAt'],
+    description: "The stored copy of a judged run's input and output, taken when it was first judged.",
+    properties: {
+        runId: { type: 'string' },
+        subject: { $ref: '#/components/schemas/JudgedSubject' },
+        input: {},
+        context: {
+            $ref: '#/components/schemas/JudgedRunContext',
+        },
+        output: {},
+        capturedAt: { type: 'string', format: 'date-time' },
+    },
+};
+export const JudgmentWithCopiesSchema = {
+    description: 'A judgment with the stored copies of what was judged.',
+    allOf: [
+        { $ref: '#/components/schemas/Judgment' },
+        {
+            type: 'object',
+            required: ['run'],
+            properties: {
+                run: { $ref: '#/components/schemas/JudgedRunCopy' },
+                itemValue: {
+                    description: "The judged item's value, when the judgment pointed at it.",
+                },
+            },
+        },
+    ],
+};
+export const JudgmentCollectionPageSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['data', 'hasMore'],
+    properties: {
+        data: { type: 'array', items: { $ref: '#/components/schemas/Judgment' } },
+        nextCursor: { type: 'string', description: 'Opaque cursor. Treat as opaque on the client.' },
+        hasMore: { type: 'boolean' },
+    },
+};
+export const CreateJudgmentBodySchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['runId', 'item', 'verdict'],
+    properties: {
+        runId: { type: 'string', minLength: 1 },
+        item: { $ref: '#/components/schemas/JudgedItem' },
+        verdict: { type: 'string', enum: ['yes', 'no'] },
+        reason: { type: 'string', maxLength: 4000 },
+        judgeClassId: {
+            type: 'string',
+            minLength: 1,
+            description: 'Optional. When given it must exist and apply to the run.',
+        },
+        participantId: {
+            type: 'string',
+            minLength: 1,
+            description: "An app's opaque id for its end user, when judging on their behalf.",
+        },
+    },
+};
+export const UnregisterJudgmentResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['judgmentId', 'unregistered'],
+    properties: {
+        judgmentId: { type: 'string' },
+        unregistered: { type: 'boolean', const: true },
+    },
+};
+export const JudgedItemSummarySchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['key', 'yes', 'no', 'yesWeight', 'totalWeight', 'reasons'],
+    description: "The judgments of one item of a case's output, summed up.",
+    properties: {
+        key: { type: 'string' },
+        pointer: { type: 'string' },
+        rank: { type: 'integer', minimum: 0 },
+        yes: { type: 'integer', minimum: 0, description: 'How many judgments said yes.' },
+        no: { type: 'integer', minimum: 0, description: 'How many judgments said no.' },
+        yesWeight: {
+            type: 'number',
+            description: 'The weight behind "yes" (an unclassified judgment counts 1).',
+        },
+        totalWeight: { type: 'number', description: 'The weight behind all judgments of the item.' },
+        reasons: {
+            type: 'array',
+            description: 'The reasons given, newest first.',
+            items: {
+                type: 'object',
+                additionalProperties: false,
+                required: ['verdict', 'reason'],
+                properties: {
+                    verdict: { type: 'string', enum: ['yes', 'no'] },
+                    reason: { type: 'string' },
+                },
+            },
+        },
+    },
+};
+export const JudgedEvalCaseSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['caseId', 'subject', 'input', 'output', 'items'],
+    description: "One case of a `judged` eval suite: a copy of a judged run with its items' judgments summed up.",
+    properties: {
+        caseId: { type: 'string', description: "The judged run's id." },
+        subject: { $ref: '#/components/schemas/JudgedSubject' },
+        input: {},
+        context: { $ref: '#/components/schemas/JudgedRunContext' },
+        output: {},
+        items: { type: 'array', items: { $ref: '#/components/schemas/JudgedItemSummary' } },
+    },
+};
+export const JudgedEvalCaseCollectionPageSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['data', 'hasMore'],
+    properties: {
+        data: { type: 'array', items: { $ref: '#/components/schemas/JudgedEvalCase' } },
+        nextCursor: { type: 'string', description: 'Opaque cursor. Treat as opaque on the client.' },
+        hasMore: { type: 'boolean' },
+    },
+};
+export const BuildJudgedSuiteBodySchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['version', 'projectId'],
+    description: 'Name the agent (`agentId`) or the flow (`flowId`) whose judged runs to use.',
+    properties: {
+        version: { type: 'string', description: 'Semver version to publish, e.g. `1.0.0`.' },
+        projectId: { type: 'string', minLength: 1 },
+        agentId: { type: 'string', minLength: 1 },
+        agentVersion: { type: 'string', minLength: 1, description: 'Needs `agentId`.' },
+        flowId: { type: 'string', minLength: 1 },
+        since: {
+            type: 'string',
+            format: 'date-time',
+            description: 'Runs first judged at or after this time.',
+        },
+        until: {
+            type: 'string',
+            format: 'date-time',
+            description: 'Runs first judged before this time.',
+        },
+        judgeClassIds: {
+            type: 'array',
+            items: { type: 'string', minLength: 1 },
+            description: 'Count only judgments recorded under these judge classes.',
+        },
+        minJudgments: {
+            type: 'integer',
+            minimum: 1,
+            description: 'Leave out runs with fewer counted judgments. Default 1.',
+        },
+        description: { type: 'string' },
+    },
+};
+export const BuildJudgedSuiteResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['suiteId', 'version', 'kind', 'caseCount', 'truncated'],
+    properties: {
+        suiteId: { type: 'string' },
+        version: { type: 'string' },
+        kind: { type: 'string', enum: ['judged'] },
+        caseCount: { type: 'integer', minimum: 0 },
+        truncated: {
+            type: 'boolean',
+            description: 'Whether more judged runs matched than the 1000 cases a set holds.',
+        },
+    },
+};
+// ---------------- memory ----------------
+export const FactScopeSchema = {
+    type: 'object',
+    additionalProperties: true,
+    required: ['tenantId'],
+    description: 'Fact scope object. `tenantId` is required; every optional key narrows the fact (`userId`, `orgId`, `projectId`, `threadId`, `sessionId`). Additional keys accepted for forward compatibility.',
+    properties: {
+        tenantId: { type: 'string' },
+        userId: { type: 'string' },
+        orgId: { type: 'string' },
+        projectId: { type: 'string' },
+        threadId: { type: 'string' },
+        sessionId: { type: 'string' },
+    },
+};
+export const RetentionSchema = {
+    type: 'object',
+    additionalProperties: false,
+    properties: {
+        keepUntil: { type: 'string', format: 'date-time' },
+        keepDays: { type: 'integer', minimum: 1 },
+        legalHold: { type: 'boolean' },
+    },
+};
+export const SourceFreshnessSchema = {
+    type: 'object',
+    additionalProperties: false,
+    properties: {
+        ttlSeconds: { type: 'integer', minimum: 0 },
+        lastVerifiedAt: { type: 'string', format: 'date-time' },
+        etag: { type: 'string' },
+        sourceVersion: { type: 'string' },
+    },
+};
+export const SourceRefreshSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['strategy'],
+    properties: {
+        strategy: { type: 'string', enum: ['on-read', 'background', 'manual'] },
+        handler: { type: 'string' },
+        priority: { type: 'integer' },
+    },
+};
+export const FactSourceSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['kind', 'freshness', 'refresh'],
+    properties: {
+        kind: {
+            type: 'string',
+            enum: ['http-api', 'blob', 'mcp-tool', 'external-db', 'user-input'],
+        },
+        uri: { type: 'string' },
+        freshness: { $ref: '#/components/schemas/SourceFreshness' },
+        refresh: { $ref: '#/components/schemas/SourceRefresh' },
+    },
+};
+export const FactSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['id', 'type', 'scope', 'version', 'createdAt'],
+    properties: {
+        id: { type: 'string', description: 'FactId.' },
+        type: {
+            type: 'string',
+            description: 'Fact type identifier (pack-defined; a few are framework-standard).',
+        },
+        scope: { $ref: '#/components/schemas/FactScope' },
+        version: {
+            type: 'integer',
+            minimum: 1,
+            description: 'Monotonic version within (scope, id). Supersession increments.',
+        },
+        createdAt: { type: 'string', format: 'date-time' },
+        updatedAt: { type: 'string', format: 'date-time' },
+        content: { description: 'Free-form structured payload.' },
+        contentRef: {
+            type: 'string',
+            description: '`blob://<provider>/<bucket>/<key>` when the payload is stored externally.',
+        },
+        contentHash: { type: 'string' },
+        size: { type: 'integer', minimum: 0 },
+        embeddingModel: { type: 'string' },
+        retention: { $ref: '#/components/schemas/Retention' },
+        source: { $ref: '#/components/schemas/FactSource' },
+        causedByLogId: { type: 'array', items: { type: 'string' } },
+        supersedes: {
+            type: 'string',
+            description: 'FactId of the predecessor when this row supersedes another.',
+        },
+    },
+};
+export const FactCollectionPageSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['data', 'hasMore'],
+    properties: {
+        data: { type: 'array', items: { $ref: '#/components/schemas/Fact' } },
+        nextCursor: {
+            type: 'string',
+            description: 'Opaque cursor for the next page. Absent when `hasMore: false`.',
+        },
+        hasMore: { type: 'boolean' },
+    },
+};
+export const WriteFactBodySchema = {
+    description: "Write a fact. `type` selects the retrieval-policy (which indexes populate); `scope.tenantId` MUST match the caller's tenant. Optional `retention` overrides tenant defaults; optional `contentHash` is a caller-supplied idempotence hint (runtime computes its own hash regardless).",
+    type: 'object',
+    additionalProperties: false,
+    required: ['type', 'scope', 'content'],
+    properties: {
+        type: { type: 'string', minLength: 1 },
+        scope: { $ref: '#/components/schemas/FactScope' },
+        content: { description: 'Free-form structured payload.' },
+        retention: { $ref: '#/components/schemas/Retention' },
+        contentHash: { type: 'string' },
+    },
+};
+export const SupersedeFactResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['factId', 'superseded'],
+    properties: {
+        factId: { type: 'string' },
+        superseded: { type: 'boolean', const: true },
+    },
+};
+export const RetrieveIntentSchema = {
+    description: 'Retrieval intent — mirrors `RetrievalIntent` from `@kindgi/agents`, widened for direct-HTTP use. `mode: "list"` returns a plain scoped list (no query). `mode: "keyword"` runs full-text search. `mode: "semantic"` runs vector similarity search — requires an embedding provider bound on the deployment; if unavailable, the route returns `400 bad-input`. `mode: "both"` unions keyword + semantic results, dedup by fact id.',
+    type: 'object',
+    additionalProperties: false,
+    required: ['mode'],
+    properties: {
+        mode: { type: 'string', enum: ['list', 'keyword', 'semantic', 'both'] },
+        query: {
+            type: 'string',
+            description: 'Required for `keyword` / `semantic` / `both`; ignored for `list`.',
+        },
+        type: { type: 'string', minLength: 1 },
+        scope: { $ref: '#/components/schemas/FactScope' },
+        limit: { type: 'integer', minimum: 1 },
+        embeddingModel: { type: 'string' },
+    },
+};
+export const RetrieveMemoryBodySchema = RetrieveIntentSchema;
+export const RetrievalHitSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['fact'],
+    properties: {
+        fact: { $ref: '#/components/schemas/Fact' },
+        score: {
+            type: 'number',
+            description: 'Relevance score. Keyword mode returns an implementation-defined rank (higher = better). Semantic mode returns cosine similarity in [-1, 1] (higher = better). Absent for `list` mode.',
+        },
+    },
+};
+export const RetrieveMemoryResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['results'],
+    properties: {
+        results: { type: 'array', items: { $ref: '#/components/schemas/RetrievalHit' } },
+    },
+};
+export const RevokeTokenResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['tokenId', 'revoked'],
+    properties: {
+        tokenId: { type: 'string', format: 'uuid' },
+        revoked: { type: 'boolean', const: true },
+    },
+};
+// ---------------- signing keys (the deployment trust list) ----------------
+export const TrustedSigningKeySchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['keyId', 'tenantId', 'algorithm', 'publicKey', 'createdAt'],
+    description: 'A public key the tenant trusts to sign deployments (`POST /v1/deployments`). A revoked key stays readable, with `revokedAt`, so the deployments it signed can be audited; it verifies no new ones.',
+    properties: {
+        keyId: { type: 'string', description: 'The signer key id a deployment envelope names.' },
+        tenantId: { type: 'string', format: 'uuid' },
         algorithm: { type: 'string', enum: ['ed25519'] },
         publicKey: {
             type: 'string',
@@ -2663,6 +3252,13 @@ export const ProviderMetadataSchema = {
             type: 'boolean',
             description: "A fallback serves a capability only when no other provider satisfies it (e.g. `kindgi dev`'s scripted `dev-echo`); an agent turn routed to one carries a `fallback-provider` warning. Absent = `false`.",
         },
+        labels: {
+            type: 'object',
+            maxProperties: 32,
+            propertyNames: { pattern: '^[a-z0-9]([a-z0-9._/-]{0,61}[a-z0-9])?$' },
+            additionalProperties: { type: 'string', maxLength: 256 },
+            description: 'Bookkeeping, such as who manages the provider; the router ignores labels. At most 32 keys; a key is 1-63 lowercase letters and digits, with `.`, `-`, `_` or `/` inside; a value is at most 256 characters. The convention key `kindgi.com/managed-by` names the manager (`kindgi-dev`, `kindgi-deploy:<environment>`). Out of bounds: `400 invalid-provider`, reason `invalid-labels`.',
+        },
     },
 };
 export const ProviderCollectionPageSchema = {
@@ -3194,6 +3790,16 @@ export const CostAggregateResultSchema = {
         groups: {
             type: 'array',
             items: { $ref: '#/components/schemas/CostAggregateGroup' },
+            description: 'The most expensive groups first (`totalUsd` descending, ties by key), at most `limit`.',
+        },
+        totalGroups: {
+            type: 'integer',
+            minimum: 0,
+            description: 'How many groups there were before the `limit` cap. Absent from a runtime before 0.1.5.',
+        },
+        truncated: {
+            type: 'boolean',
+            description: '`true` when there were more groups than `limit`: `groups` holds the most expensive ones, and `totalUsd` / `totalRecords` / `tokens` still cover every record. Absent from a runtime before 0.1.5.',
         },
         totalUsd: { type: 'number', minimum: 0 },
         totalRecords: { type: 'integer', minimum: 0 },
@@ -3455,7 +4061,7 @@ export const ReinstatePolicyVersionResultSchema = {
  */
 export const EvalKindSchema = {
     type: 'string',
-    enum: ['accuracy', 'pairwise', 'regression', 'human-review', 'benchmark', 'custom'],
+    enum: ['accuracy', 'pairwise', 'regression', 'human-review', 'benchmark', 'custom', 'judged'],
 };
 /**
  * Wire shape for a single evaluation suite. `spec` is a JSON object
@@ -3469,80 +4075,196 @@ export const EvalKindSchema = {
 export const EvalSuiteSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['id', 'tenantId', 'version', 'kind', 'spec'],
+    required: ['id', 'tenantId', 'version', 'kind', 'spec'],
+    properties: {
+        id: { type: 'string', minLength: 1 },
+        tenantId: { type: 'string', format: 'uuid' },
+        version: {
+            type: 'string',
+            pattern: '^\\d+\\.\\d+\\.\\d+$',
+            description: 'Semver — publishing a modified suite produces a new version.',
+        },
+        kind: { $ref: '#/components/schemas/EvalKind' },
+        description: { type: 'string' },
+        spec: {
+            type: 'object',
+            additionalProperties: true,
+            description: 'Kind-specific suite body. For `accuracy`, typically `{ cases: [{ input, expectedOutput }], grader?: { adapterId, config? } }`. For `pairwise`, typically `{ prompts, variantA, variantB }`. For `regression`, typically `{ baseline, cases }`. For `human-review`, typically `{ rubric, reviewerRole }`. For `benchmark`, typically `{ benchmark: { name, version } }`. For `custom`, typically `{ handler: { modulePath, entrypointPath }, cases }`.',
+        },
+    },
+};
+export const EvalSuiteCollectionPageSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['data', 'hasMore'],
+    properties: {
+        data: {
+            type: 'array',
+            items: { $ref: '#/components/schemas/EvalSuite' },
+        },
+        nextCursor: { type: 'string' },
+        hasMore: { type: 'boolean' },
+    },
+};
+export const PublishEvalSuiteBodySchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['id', 'version', 'kind', 'spec', 'projectId'],
+    properties: {
+        id: { type: 'string', minLength: 1 },
+        projectId: ContentProjectIdProperty,
+        tenantId: {
+            type: 'string',
+            format: 'uuid',
+            description: 'Optional. When present, must match the caller tenant (server-derived from the token). Cross-tenant publish is rejected.',
+        },
+        version: { type: 'string', pattern: '^\\d+\\.\\d+\\.\\d+$' },
+        kind: { $ref: '#/components/schemas/EvalKind' },
+        description: { type: 'string' },
+        spec: { type: 'object', additionalProperties: true },
+    },
+};
+export const PublishEvalSuiteResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['suiteId', 'version'],
+    properties: {
+        suiteId: { type: 'string' },
+        version: { type: 'string' },
+    },
+};
+export const UnregisterEvalSuiteResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['suiteId', 'version', 'unregistered'],
+    properties: {
+        suiteId: { type: 'string' },
+        version: { type: 'string' },
+        unregistered: { type: 'boolean', const: true },
+    },
+};
+export const ReinstateEvalSuiteVersionResultSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['suiteId', 'version', 'wasTombstoned'],
+    properties: {
+        suiteId: { type: 'string' },
+        version: { type: 'string' },
+        wasTombstoned: { type: 'boolean' },
+    },
+};
+// ---------------- data blocks ----------------
+export const BlockKindSchema = {
+    type: 'string',
+    enum: ['prompt', 'settings'],
+    description: '`prompt`: a Liquid template an agent renders as its instructions. `settings`: a JSON object tools and templates read.',
+};
+export const PromptBlockContentSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['template'],
+    properties: {
+        template: {
+            type: 'string',
+            minLength: 1,
+            description: "Liquid, rendered as an agent's instructions are (same parameters and auto-injected variables).",
+        },
+        parameters: { type: 'array', items: { $ref: '#/components/schemas/PromptParameter' } },
+    },
+};
+export const SettingsBlockContentSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['values'],
     properties: {
-        id: { type: 'string', minLength: 1 },
-        tenantId: { type: 'string', format: 'uuid' },
-        version: {
-            type: 'string',
-            pattern: '^\\d+\\.\\d+\\.\\d+$',
-            description: 'Semver — publishing a modified suite produces a new version.',
+        values: {
+            type: 'object',
+            additionalProperties: true,
+            description: 'What tools read (`ToolContext.settings[<block id>]`) and templates read (`settings.<block id>.<key>`).',
         },
-        kind: { $ref: '#/components/schemas/EvalKind' },
-        description: { type: 'string' },
-        spec: {
+        schema: {
             type: 'object',
             additionalProperties: true,
-            description: 'Kind-specific suite body. For `accuracy`, typically `{ cases: [{ input, expectedOutput }], grader?: { adapterId, config? } }`. For `pairwise`, typically `{ prompts, variantA, variantB }`. For `regression`, typically `{ baseline, cases }`. For `human-review`, typically `{ rubric, reviewerRole }`. For `benchmark`, typically `{ benchmark: { name, version } }`. For `custom`, typically `{ handler: { modulePath, entrypointPath }, cases }`.',
+            description: "JSON Schema (draft 2020-12) the values must satisfy. A later version's values must satisfy the latest version's schema too.",
         },
     },
 };
-export const EvalSuiteCollectionPageSchema = {
+export const BlockSchema = {
+    description: 'A data block version: a prompt or settings, versioned like a tool (immutable versions, soft unregister). An agent version pins the block versions it uses when it is published. Belongs to one project and is authorized through it.',
     type: 'object',
     additionalProperties: false,
-    required: ['data', 'hasMore'],
+    required: ['id', 'version', 'kind', 'content', 'projectId', 'publishedAt'],
     properties: {
-        data: {
-            type: 'array',
-            items: { $ref: '#/components/schemas/EvalSuite' },
+        id: { type: 'string', description: 'Dotted lowercase id (e.g. `acme.intake-prompt`).' },
+        version: { type: 'string', pattern: '^\\d+\\.\\d+\\.\\d+$' },
+        kind: { $ref: '#/components/schemas/BlockKind' },
+        description: { type: 'string' },
+        content: {
+            oneOf: [
+                { $ref: '#/components/schemas/PromptBlockContent' },
+                { $ref: '#/components/schemas/SettingsBlockContent' },
+            ],
+            description: '`PromptBlockContent` for a prompt, `SettingsBlockContent` for settings.',
+        },
+        projectId: { type: 'string', format: 'uuid' },
+        publishedAt: { type: 'string', format: 'date-time' },
+        unregisteredAt: {
+            type: 'string',
+            format: 'date-time',
+            description: 'Present only on an unregistered version; agent versions that pin it still read it.',
         },
+    },
+};
+export const BlockCollectionPageSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['data', 'hasMore'],
+    properties: {
+        data: { type: 'array', items: { $ref: '#/components/schemas/Block' } },
         nextCursor: { type: 'string' },
         hasMore: { type: 'boolean' },
     },
 };
-export const PublishEvalSuiteBodySchema = {
+export const PublishBlockBodySchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['id', 'version', 'kind', 'spec', 'projectId'],
+    required: ['projectId', 'id', 'version', 'kind', 'content'],
     properties: {
-        id: { type: 'string', minLength: 1 },
         projectId: ContentProjectIdProperty,
-        tenantId: {
-            type: 'string',
-            format: 'uuid',
-            description: 'Optional. When present, must match the caller tenant (server-derived from the token). Cross-tenant publish is rejected.',
-        },
+        id: { type: 'string', minLength: 1 },
         version: { type: 'string', pattern: '^\\d+\\.\\d+\\.\\d+$' },
-        kind: { $ref: '#/components/schemas/EvalKind' },
+        kind: { $ref: '#/components/schemas/BlockKind' },
         description: { type: 'string' },
-        spec: { type: 'object', additionalProperties: true },
+        content: {
+            oneOf: [
+                { $ref: '#/components/schemas/PromptBlockContent' },
+                { $ref: '#/components/schemas/SettingsBlockContent' },
+            ],
+        },
     },
 };
-export const PublishEvalSuiteResultSchema = {
+export const PublishBlockResultSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['suiteId', 'version'],
-    properties: {
-        suiteId: { type: 'string' },
-        version: { type: 'string' },
-    },
+    required: ['blockId', 'version'],
+    properties: { blockId: { type: 'string' }, version: { type: 'string' } },
 };
-export const UnregisterEvalSuiteResultSchema = {
+export const UnregisterBlockResultSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['suiteId', 'version', 'unregistered'],
+    required: ['blockId', 'version', 'unregistered'],
     properties: {
-        suiteId: { type: 'string' },
+        blockId: { type: 'string' },
         version: { type: 'string' },
         unregistered: { type: 'boolean', const: true },
     },
 };
-export const ReinstateEvalSuiteVersionResultSchema = {
+export const ReinstateBlockResultSchema = {
     type: 'object',
     additionalProperties: false,
-    required: ['suiteId', 'version', 'wasTombstoned'],
+    required: ['blockId', 'version', 'wasTombstoned'],
     properties: {
-        suiteId: { type: 'string' },
+        blockId: { type: 'string' },
         version: { type: 'string' },
         wasTombstoned: { type: 'boolean' },
     },
@@ -3575,6 +4297,368 @@ export const EvalRunFlowRefSchema = {
         version: { type: 'string', pattern: '^\\d+\\.\\d+\\.\\d+$' },
     },
 };
+export const EvalBaselineSchema = {
+    description: "What a comparison compares the candidate against: `'recorded'` (each case's recorded output, what was judged), a version (`{ agentId, version }`, replayed under the same rules), or the version live in a scope (`{ live: { projectId?, segments? } }`). Only `'recorded'` runs today; the others are refused when the run starts.",
+    oneOf: [
+        { type: 'string', enum: ['recorded'] },
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['agentId', 'version'],
+            properties: { agentId: { type: 'string' }, version: { type: 'string' } },
+        },
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['live'],
+            properties: {
+                live: {
+                    type: 'object',
+                    additionalProperties: false,
+                    properties: {
+                        projectId: { type: 'string' },
+                        segments: { type: 'object', additionalProperties: { type: 'string' } },
+                    },
+                },
+            },
+        },
+    ],
+};
+export const EvalComparisonSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['baseline', 'reads', 'repetitions', 'k'],
+    description: "A comparison eval run's settings (a `judged` suite).",
+    properties: {
+        baseline: { $ref: '#/components/schemas/EvalBaseline' },
+        reads: {
+            type: 'string',
+            enum: ['recorded', 'live'],
+            description: "Whether replayed reads use the past run's results when it has them (`recorded`), or run live.",
+        },
+        repetitions: { type: 'integer', minimum: 1, maximum: 10 },
+        k: { type: 'integer', minimum: 1, maximum: 100 },
+        versions: { $ref: '#/components/schemas/FlowVersionOverrides' },
+    },
+};
+// ---------------- comparison results (a `judged` eval run's `result`) ----------------
+const nullableNumber = { type: ['number', 'null'] };
+export const ComparisonMetricSchema = {
+    description: "One metric, the recorded runs beside the candidate. `null` where a side had no judged evidence; `n` / `weight` are the candidate's evidence (cases with judged items, and the judgment weight behind them), `baselineN` / `baselineWeight` the recorded side's.",
+    type: 'object',
+    additionalProperties: false,
+    required: [
+        'baseline',
+        'candidate',
+        'delta',
+        'n',
+        'weight',
+        'baselineN',
+        'baselineWeight',
+        'direction',
+    ],
+    properties: {
+        baseline: nullableNumber,
+        candidate: nullableNumber,
+        delta: nullableNumber,
+        n: { type: 'integer', minimum: 0 },
+        weight: { type: 'number', minimum: 0 },
+        baselineN: { type: 'integer', minimum: 0 },
+        baselineWeight: { type: 'number', minimum: 0 },
+        direction: { type: 'string', enum: ['higher'] },
+        k: {
+            type: 'integer',
+            minimum: 1,
+            description: '`weightedPrecisionAtK`: the ranked items it looked at.',
+        },
+        spread: {
+            type: 'number',
+            description: "With more than one repetition: the candidate's max − min across them.",
+        },
+    },
+};
+export const ComparisonCandidateSchema = {
+    description: 'What ran on the cases: an agent version, or a flow version (with any versions it swapped in).',
+    oneOf: [
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['kind', 'agentId', 'version'],
+            properties: {
+                kind: { type: 'string', enum: ['agent'] },
+                agentId: { type: 'string' },
+                version: { type: 'string' },
+            },
+        },
+        {
+            type: 'object',
+            additionalProperties: false,
+            required: ['kind', 'flowId', 'version'],
+            properties: {
+                kind: { type: 'string', enum: ['flow'] },
+                flowId: { type: 'string' },
+                version: { type: 'string' },
+                versions: { $ref: '#/components/schemas/FlowVersionOverrides' },
+            },
+        },
+    ],
+};
+export const JudgedComparisonSummarySchema = {
+    description: 'What a comparison concluded: the candidate beside the recorded runs, the case counts, and the metrics. What a promotion gate reads.',
+    type: 'object',
+    additionalProperties: false,
+    required: [
+        'evalRunId',
+        'status',
+        'completedAt',
+        'suite',
+        'candidate',
+        'baseline',
+        'scope',
+        'cases',
+        'diverged',
+        'refusedWrites',
+        'errors',
+        'stopped',
+        'reads',
+        'sampling',
+        'repetitions',
+        'metrics',
+    ],
+    properties: {
+        evalRunId: { type: 'string' },
+        status: { type: 'string', enum: ['completed', 'partial', 'failed'] },
+        completedAt: { type: 'string', format: 'date-time' },
+        suite: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['id', 'version'],
+            properties: { id: { type: 'string' }, version: { type: 'string' } },
+        },
+        candidate: { $ref: '#/components/schemas/ComparisonCandidate' },
+        baseline: {
+            description: "What the candidate was compared with: `recorded` (the test set's recorded runs, with the versions that served them), or another version.",
+            oneOf: [
+                {
+                    type: 'object',
+                    additionalProperties: false,
+                    required: ['kind', 'versions'],
+                    properties: {
+                        kind: { type: 'string', enum: ['recorded'] },
+                        versions: {
+                            type: 'array',
+                            items: {
+                                type: 'object',
+                                additionalProperties: false,
+                                required: ['version', 'cases'],
+                                properties: {
+                                    agentId: { type: 'string' },
+                                    flowId: { type: 'string' },
+                                    version: { type: 'string' },
+                                    cases: { type: 'integer', minimum: 0 },
+                                },
+                            },
+                        },
+                    },
+                },
+                {
+                    type: 'object',
+                    additionalProperties: false,
+                    required: ['kind', 'agentId', 'version', 'via'],
+                    properties: {
+                        kind: { type: 'string', enum: ['version'] },
+                        agentId: { type: 'string' },
+                        version: { type: 'string' },
+                        via: { type: 'string', enum: ['explicit', 'live'] },
+                        liveScope: { type: 'object', additionalProperties: true },
+                    },
+                },
+            ],
+        },
+        scope: {
+            type: 'object',
+            additionalProperties: false,
+            description: "Where the test set's judgments came from.",
+            properties: { projectId: { type: 'string' } },
+        },
+        cases: { type: 'integer', minimum: 0 },
+        diverged: {
+            type: 'integer',
+            minimum: 0,
+            description: "Cases where a read with no recording ran live under `reads: 'recorded'`.",
+        },
+        refusedWrites: {
+            type: 'integer',
+            minimum: 0,
+            description: 'Tool calls refused across the cases (what the candidate would have done).',
+        },
+        errors: { type: 'integer', minimum: 0, description: 'Cases none of whose repetitions ran.' },
+        stopped: {
+            type: 'integer',
+            minimum: 0,
+            description: "Flow cases that stopped at a write the replay refused: no output to score, so they're left out of the metrics.",
+        },
+        reads: { type: 'string', enum: ['recorded', 'live'] },
+        sampling: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['models'],
+            properties: {
+                models: {
+                    type: 'array',
+                    description: "The models that answered the candidate's replays, and how many replays each.",
+                    items: {
+                        type: 'object',
+                        additionalProperties: false,
+                        required: ['providerId', 'model', 'runs'],
+                        properties: {
+                            providerId: { type: 'string' },
+                            model: { type: 'string' },
+                            runs: { type: 'integer', minimum: 0 },
+                        },
+                    },
+                },
+            },
+        },
+        repetitions: { type: 'integer', minimum: 1 },
+        metrics: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['weightedYesShare', 'judgedCoverage', 'weightedPrecisionAtK'],
+            properties: {
+                weightedYesShare: { $ref: '#/components/schemas/ComparisonMetric' },
+                judgedCoverage: { $ref: '#/components/schemas/ComparisonMetric' },
+                weightedPrecisionAtK: { $ref: '#/components/schemas/ComparisonMetric' },
+            },
+        },
+    },
+};
+const outputScore = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['yesWeight', 'totalWeight', 'items', 'judgedItems', 'topK'],
+    description: "An output's score: Σ yesWeight and Σ totalWeight over its judged items, and over those among the first `k` ranked items.",
+    properties: {
+        yesWeight: { type: 'number' },
+        totalWeight: { type: 'number' },
+        items: { type: 'integer', minimum: 0 },
+        judgedItems: { type: 'integer', minimum: 0 },
+        topK: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['yesWeight', 'totalWeight'],
+            properties: { yesWeight: { type: 'number' }, totalWeight: { type: 'number' } },
+        },
+    },
+};
+export const ComparisonCaseResultSchema = {
+    description: "One case of a comparison: its replay runs, the scores, the items kept, dropped and new, the tool calls, and why it didn't run when it didn't.",
+    type: 'object',
+    additionalProperties: false,
+    required: [
+        'caseId',
+        'runIds',
+        'baseline',
+        'candidate',
+        'diverged',
+        'refusedWrites',
+        'noContext',
+        'approvalSkipped',
+    ],
+    properties: {
+        caseId: { type: 'string' },
+        runIds: {
+            type: 'array',
+            items: { type: 'string' },
+            description: "The candidate's replay runs, one per repetition.",
+        },
+        baseline: outputScore,
+        candidate: { type: 'array', items: outputScore, description: 'One per repetition that ran.' },
+        changes: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['kept', 'dropped', 'new'],
+            description: "The first repetition's items against the judged ones.",
+            properties: {
+                kept: {
+                    type: 'array',
+                    items: {
+                        type: 'object',
+                        additionalProperties: false,
+                        required: ['key'],
+                        properties: {
+                            key: { type: 'string' },
+                            rankBefore: { type: 'integer' },
+                            rank: { type: 'integer' },
+                        },
+                    },
+                },
+                dropped: {
+                    type: 'array',
+                    items: {
+                        type: 'object',
+                        additionalProperties: false,
+                        required: ['key'],
+                        properties: { key: { type: 'string' }, rankBefore: { type: 'integer' } },
+                    },
+                },
+                new: {
+                    type: 'array',
+                    items: {
+                        type: 'object',
+                        additionalProperties: false,
+                        required: ['key', 'pointer'],
+                        properties: {
+                            key: { type: 'string' },
+                            pointer: { type: 'string' },
+                            rank: { type: 'integer' },
+                        },
+                    },
+                },
+            },
+        },
+        tools: {
+            type: 'array',
+            description: "The first repetition's tool calls, and what happened to each.",
+            items: {
+                type: 'object',
+                additionalProperties: false,
+                required: ['step', 'callId', 'toolId', 'toolVersion', 'arguments', 'source'],
+                properties: {
+                    step: { type: 'integer', minimum: 0 },
+                    callId: { type: 'string' },
+                    toolId: { type: 'string' },
+                    toolVersion: { type: 'string' },
+                    arguments: {},
+                    source: { type: 'string', enum: ['live', 'recorded', 'refused'] },
+                    reason: { type: 'string' },
+                },
+            },
+        },
+        diverged: { type: 'boolean' },
+        refusedWrites: { type: 'integer', minimum: 0 },
+        noContext: { type: 'boolean' },
+        approvalSkipped: { type: 'boolean' },
+        error: { type: 'string' },
+        stopped: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['toolId', 'arguments'],
+            description: 'Set when the replay stopped at a refused write: what it would have done.',
+            properties: { toolId: { type: 'string' }, arguments: {}, reason: { type: 'string' } },
+        },
+    },
+};
+export const JudgedComparisonResultSchema = {
+    description: "A comparison's `result` (a `judged` eval run's): the summary and each case. `EvalRun.result` stays an open object, since each kind has its own; the clients read it as this (TS `comparisonOf(run)`, Python `comparison_of(run)`).",
+    type: 'object',
+    additionalProperties: false,
+    required: ['summary', 'perCase'],
+    properties: {
+        summary: { $ref: '#/components/schemas/JudgedComparisonSummary' },
+        perCase: { type: 'array', items: { $ref: '#/components/schemas/ComparisonCaseResult' } },
+    },
+};
 export const EvalRunSchema = {
     type: 'object',
     additionalProperties: false,
@@ -3603,10 +4687,11 @@ export const EvalRunSchema = {
         result: {
             type: 'object',
             additionalProperties: true,
-            description: 'Kind-specific opaque JSON. For `accuracy`, contains `{ passCount, totalCount, meanScore, perCase[] }`. Other kinds define their own shapes as their dispatchers ship.',
+            description: 'Kind-specific opaque JSON. For `accuracy`, contains `{ passCount, totalCount, meanScore, perCase[] }`. For `judged` (a comparison), `{ summary, perCase[] }`: the summary has the baseline (the versions behind the recorded runs) and the candidate (`{ kind: "agent", agentId, version }` or `{ kind: "flow", flowId, version, versions? }`), the case counts (`cases`, `diverged`, `refusedWrites`, `errors`, and `stopped`: flow cases that stopped at a write the replay refused, left out of the metrics), the models that answered, and `metrics` (`weightedYesShare`, `judgedCoverage`, `weightedPrecisionAtK`, each `{ baseline, candidate, delta, n, weight, baselineN, baselineWeight, direction, k?, spread? }`); each case has its replay runs, the scores, the items kept, dropped and new, the tool calls with what happened to each, and `stopped` (what it would have done) when it stopped. Other kinds define their own shapes as their dispatchers ship.',
         },
         error: { type: 'string' },
         correlationId: { type: 'string' },
+        comparison: { $ref: '#/components/schemas/EvalComparison' },
     },
 };
 export const EvalRunCollectionPageSchema = {
@@ -3629,8 +4714,13 @@ export const StartEvalRunBodySchema = {
         flowRef: { $ref: '#/components/schemas/EvalRunFlowRef' },
         dryRun: { type: 'boolean' },
         correlationId: { type: 'string' },
+        baseline: { $ref: '#/components/schemas/EvalBaseline' },
+        reads: { type: 'string', enum: ['recorded', 'live'] },
+        repetitions: { type: 'integer', minimum: 1, maximum: 10 },
+        k: { type: 'integer', minimum: 1, maximum: 100 },
+        versions: { $ref: '#/components/schemas/FlowVersionOverrides' },
     },
-    description: 'Exactly one of `agentRef` or `flowRef` MUST be supplied. `dryRun: true` returns a plan preview without invoking the subject.',
+    description: "Exactly one of `agentRef` or `flowRef` MUST be supplied. `dryRun: true` returns a plan preview without invoking the subject. For a `judged` suite (a test set), the run is a comparison: `agentRef` or `flowRef` with its `version` is the candidate, replayed on each case without doing anything the past run didn't (a flow stops at a write the replay refuses); `baseline` (default `'recorded'`), `reads` (default `recorded`), `repetitions` (default 1) and `k` (default 10) set how. With `flowRef`, `versions` runs the flow with some of its agents or tools at other versions; an id the flow doesn't use, or a version that isn't published, is refused (`400 validation-failed`, each under `details.issues`).",
 };
 export const StartEvalRunResultSchema = {
     type: 'object',
@@ -3883,6 +4973,45 @@ const DeployedPrimitiveSchema = {
         version: { type: 'string', description: 'Absent for guardrails, which have no version.' },
     },
 };
+const DeployedVersionSchema = {
+    type: 'object',
+    additionalProperties: false,
+    required: ['id', 'version'],
+    properties: {
+        id: { type: 'string' },
+        version: { type: 'string', description: 'The version the agent is registered as.' },
+        authoredVersion: {
+            type: 'string',
+            description: "The version the agent's or flow's definition names, present when it differs from `version`: that version was registered already with other pins or content, and versions never change, so the deploy registered the next free version in its line (or an earlier deploy did).",
+        },
+        reason: {
+            type: 'string',
+            enum: ['pins-changed', 'unpinned', 'version-taken'],
+            description: 'Why `version` differs from `authoredVersion`: `pins-changed` (a tool or agent it uses has a new version), `unpinned` (`authoredVersion` was published before pins existed), `version-taken` (`authoredVersion` is registered with other content).',
+        },
+        newVersion: {
+            type: 'boolean',
+            description: '`true`: this deploy registered `version`; `false`: an earlier deploy did.',
+        },
+        pinChanges: {
+            type: 'array',
+            description: "For `pins-changed`: the pins that differ from `authoredVersion`'s.",
+            items: { $ref: '#/components/schemas/PinChange' },
+        },
+    },
+};
+export const PinChangeSchema = {
+    description: 'One pin that differs between two versions of an agent or a flow.',
+    type: 'object',
+    additionalProperties: false,
+    required: ['kind', 'id'],
+    properties: {
+        kind: { type: 'string', enum: ['tool', 'prompt', 'setting', 'agent'] },
+        id: { type: 'string' },
+        from: { type: 'string', description: "The earlier version's pin; absent when it had none." },
+        to: { type: 'string', description: "The later version's pin; absent when it has none." },
+    },
+};
 /**
  * Exactly what a deployment shipped. Versions are immutable, so this
  * says which code the deployment made live.
@@ -3894,8 +5023,8 @@ export const DeploymentContentsSchema = {
     properties: {
         tools: { type: 'array', items: DeployedPrimitiveSchema },
         guardrails: { type: 'array', items: DeployedPrimitiveSchema },
-        agents: { type: 'array', items: DeployedPrimitiveSchema },
-        flows: { type: 'array', items: DeployedPrimitiveSchema },
+        agents: { type: 'array', items: DeployedVersionSchema },
+        flows: { type: 'array', items: DeployedVersionSchema },
     },
 };
 /**
@@ -5708,6 +6837,27 @@ export const COMPONENT_SCHEMAS = [
     ['ObservationStatus', ObservationStatusSchema],
     ['Observation', ObservationSchema],
     ['ObservationCollectionPage', ObservationCollectionPageSchema],
+    ['JudgeClassScope', JudgeClassScopeSchema],
+    ['JudgeClass', JudgeClassSchema],
+    ['JudgeClassCollectionPage', JudgeClassCollectionPageSchema],
+    ['CreateJudgeClassBody', CreateJudgeClassBodySchema],
+    ['UpdateJudgeClassBody', UpdateJudgeClassBodySchema],
+    ['UnregisterJudgeClassResult', UnregisterJudgeClassResultSchema],
+    ['JudgedSubject', JudgedSubjectSchema],
+    ['JudgedItem', JudgedItemSchema],
+    ['JudgmentAssertedBy', JudgmentAssertedBySchema],
+    ['Judgment', JudgmentSchema],
+    ['JudgedRunContext', JudgedRunContextSchema],
+    ['JudgedRunCopy', JudgedRunCopySchema],
+    ['JudgmentWithCopies', JudgmentWithCopiesSchema],
+    ['JudgmentCollectionPage', JudgmentCollectionPageSchema],
+    ['CreateJudgmentBody', CreateJudgmentBodySchema],
+    ['UnregisterJudgmentResult', UnregisterJudgmentResultSchema],
+    ['JudgedItemSummary', JudgedItemSummarySchema],
+    ['JudgedEvalCase', JudgedEvalCaseSchema],
+    ['JudgedEvalCaseCollectionPage', JudgedEvalCaseCollectionPageSchema],
+    ['BuildJudgedSuiteBody', BuildJudgedSuiteBodySchema],
+    ['BuildJudgedSuiteResult', BuildJudgedSuiteResultSchema],
     ['PromptParameter', PromptParameterSchema],
     ['RetrievalIntent', RetrievalIntentSchema],
     ['ConversationPolicy', ConversationPolicySchema],
@@ -5717,6 +6867,15 @@ export const COMPONENT_SCHEMAS = [
     ['AgentOutputSpec', AgentOutputSpecSchema],
     ['ToolErrorsSpec', ToolErrorsSpecSchema],
     ['Agent', AgentSchema],
+    ['AgentPins', AgentPinsSchema],
+    ['PromptRef', PromptRefSchema],
+    ['BlockRef', BlockRefSchema],
+    ['PinChange', PinChangeSchema],
+    ['VersionDerivation', VersionDerivationSchema],
+    ['DeriveAgentVersionBody', DeriveAgentVersionBodySchema],
+    ['AgentPinSwaps', AgentPinSwapsSchema],
+    ['FlowPins', FlowPinsSchema],
+    ['FlowVersionOverrides', FlowVersionOverridesSchema],
     ['PublishAgentBody', PublishAgentBodySchema],
     ['PublishAgentResult', PublishAgentResultSchema],
     ['UnregisterAgentResult', UnregisterAgentResultSchema],
@@ -5866,9 +7025,25 @@ export const COMPONENT_SCHEMAS = [
     ['PublishEvalSuiteResult', PublishEvalSuiteResultSchema],
     ['UnregisterEvalSuiteResult', UnregisterEvalSuiteResultSchema],
     ['ReinstateEvalSuiteVersionResult', ReinstateEvalSuiteVersionResultSchema],
+    ['BlockKind', BlockKindSchema],
+    ['PromptBlockContent', PromptBlockContentSchema],
+    ['SettingsBlockContent', SettingsBlockContentSchema],
+    ['Block', BlockSchema],
+    ['BlockCollectionPage', BlockCollectionPageSchema],
+    ['PublishBlockBody', PublishBlockBodySchema],
+    ['PublishBlockResult', PublishBlockResultSchema],
+    ['UnregisterBlockResult', UnregisterBlockResultSchema],
+    ['ReinstateBlockResult', ReinstateBlockResultSchema],
     ['EvalRunStatus', EvalRunStatusSchema],
     ['EvalRunAgentRef', EvalRunAgentRefSchema],
     ['EvalRunFlowRef', EvalRunFlowRefSchema],
+    ['EvalBaseline', EvalBaselineSchema],
+    ['EvalComparison', EvalComparisonSchema],
+    ['ComparisonMetric', ComparisonMetricSchema],
+    ['ComparisonCandidate', ComparisonCandidateSchema],
+    ['JudgedComparisonSummary', JudgedComparisonSummarySchema],
+    ['ComparisonCaseResult', ComparisonCaseResultSchema],
+    ['JudgedComparisonResult', JudgedComparisonResultSchema],
     ['EvalRun', EvalRunSchema],
     ['EvalRunCollectionPage', EvalRunCollectionPageSchema],
     ['StartEvalRunBody', StartEvalRunBodySchema],