> ## Documentation Index > Fetch the complete documentation index at: https://docs.errorbar.ai/llms.txt > Use this file to discover all available pages before exploring further. # Mid-run anchor check > Submit the current policy's outputs on your FROZEN anchor prompt set (same request ids every time). The server scores them with the session's reward AND the independent anchor judge (corrected for its measured error), appends the point to the session's history, and decides: reward up ≥ 10 points over the last three checks while the anchor moves ≤ 2, or the anchor ≥ 10 points below its best, HOLDS the session. Held sessions refuse scoring (409) until a human resumes them; `pinned_step` is the checkpoint to ship. Points with fewer than 20 gradable prompts hold nothing (`underpowered` says so). Hold, never kill. ## OpenAPI ````yaml /openapi.json post /v1/reward/sessions/{id}/anchor openapi: 3.1.0 info: title: errorbar Management API description: >- The management API behind the improvement loop: capture and setup, request logs and datasets, grades (labels), judges (criteria), evals and deploy gates, fine-tuning and reinforcement learning, dedicated GPU endpoints, and model aliases and versions. Authenticated with a workspace API key (sk_sovereign_...). The inference API (chat, embeddings, rerank, responses) is OpenAI-compatible and documented separately. Responses are snake_case, list endpoints on the loop products use the {"object": "list", "data": [...]} envelope, and refusals use the same nested error shape the gateway emits: {"error": {"message", "type", "code"}}. Request bodies on the loop products (logs, labels, criteria, evals, datasets, aliases) are snake_case; the training and infrastructure products (fine-tuning, GRPO, environment tools, dedicated, model-version adoption) validate camelCase bodies, and each schema below says which it is. Endpoints that spend money require a key minted by a workspace owner or admin and return 403 otherwise. version: 1.0.0 servers: - url: https://gateway.errorbar.ai description: Production - url: https://www.errorbar.ai/api description: Control plane (also served at this base URL) security: - bearerAuth: [] tags: - name: Certified reward description: >- The certified reward for your own trainer: sessions with judge certificates, scoring with a ledger-true kill switch, and the mid-run anchor that holds a run when the reward diverges from an independent judge. paths: /v1/reward/sessions/{id}/anchor: parameters: - name: id in: path required: true schema: type: string post: tags: - Certified reward summary: Mid-run anchor check description: >- Submit the current policy's outputs on your FROZEN anchor prompt set (same request ids every time). The server scores them with the session's reward AND the independent anchor judge (corrected for its measured error), appends the point to the session's history, and decides: reward up ≥ 10 points over the last three checks while the anchor moves ≤ 2, or the anchor ≥ 10 points below its best, HOLDS the session. Held sessions refuse scoring (409) until a human resumes them; `pinned_step` is the checkpoint to ship. Points with fewer than 20 gradable prompts hold nothing (`underpowered` says so). Hold, never kill. operationId: anchorCheck requestBody: required: true content: application/json: schema: type: object required: - step - items properties: step: type: string pattern: ^[A-Za-z0-9_.-]{1,40}$ items: type: array minItems: 1 maxItems: 64 items: $ref: '#/components/schemas/RewardItem' responses: '200': description: The verdict for this check. content: application/json: schema: type: object properties: session_id: type: string held: type: boolean reason: type: - string - 'null' pinned_step: type: - string - 'null' point: $ref: '#/components/schemas/AnchorPoint' underpowered: type: - string - 'null' history: type: array items: $ref: '#/components/schemas/AnchorPoint' spend_micros: type: string '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '402': $ref: '#/components/responses/InsufficientBalance' '404': $ref: '#/components/responses/NotFound' '409': description: 'Already held (`code: held`).' content: application/json: schema: $ref: '#/components/schemas/Error' '422': description: The session has no anchor, or its anchor judge no longer exists. content: application/json: schema: $ref: '#/components/schemas/Error' components: schemas: RewardItem: type: object required: - requestId - conversation properties: requestId: type: string maxLength: 200 description: >- Your id for the rollout; re-sending the same (step, requestId) settles into the same ledger rows. conversation: type: string description: The prompt/conversation the policy answered. response: type: string description: Single-turn completion (omit when sending steps). steps: type: array maxItems: 200 items: type: object required: - role - content properties: role: type: string enum: - assistant - tool content: type: string source: type: string enum: - code - recorded - declared - simulated - flagged toolName: type: string description: >- Agentic sessions: the whole rollout as structured steps; the server renders and sanitises it through the same instrument trace calibration uses. sessionId: type: string description: >- The episode's errorbar environment session id (exec verifiers + attestation). AnchorPoint: type: object properties: step: type: string at: type: string format: date-time 'n': type: integer description: Prompts both the reward and the anchor judge could grade. reward_rate: type: number description: >- Mean composite grade under the session's reward on the anchor outputs — the signal being optimised. anchor_rate: type: number description: >- The independent anchor judge's pass rate, corrected for its measured error (Lang–Reiczigel) when the criterion carries report counts. anchor_ci: type: array items: type: number minItems: 2 maxItems: 2 anchor_corrected: type: boolean masked: type: integer description: Items the anchor judge could not parse — counted in neither rate. Error: type: object properties: error: type: object properties: message: type: string type: type: string description: >- invalid_request_error, insufficient_quota, rate_limit_error, or api_error. code: type: string description: >- Machine-stable cause, e.g. invalid_api_key, not_found, insufficient_permissions, precondition_failed. required: - message - type - code description: >- Every refusal — gateway and management API alike — uses this one envelope. responses: BadRequest: description: Malformed request or invalid field. content: application/json: schema: $ref: '#/components/schemas/Error' Unauthorized: description: Missing, malformed, or revoked API key. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: Invalid API key type: invalid_request_error code: invalid_api_key InsufficientBalance: description: Your wallet can't cover the required prepaid runway. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: >- Insufficient balance: deploying this endpoint requires at least 1h of runway. Top up and try again. type: insufficient_quota code: insufficient_balance NotFound: description: The resource does not exist in your workspace. content: application/json: schema: $ref: '#/components/schemas/Error' securitySchemes: bearerAuth: type: http scheme: bearer description: >- Your workspace API key, e.g. `sk_sovereign_...`, sent as `Authorization: Bearer `. ````