> ## 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. # Create a reward session > Register the certified reward for YOUR trainer. The reward is validated first (calibrated, non-drift-flagged, population-bound judges — refused with the reason otherwise), the budget is held in the wallet, and the response carries every judge's signed certificate so the trainer holds the grader's error rate before the first rollout is scored. Optional `anchor`: an independent judge the run is checked against mid-run (POST …/anchor). Requires an owner/admin key with platform:write. errorbar never runs the training — that is what makes the certificate independent. ## OpenAPI ````yaml /openapi.json post /v1/reward/sessions 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: post: tags: - Certified reward summary: Create a reward session description: >- Register the certified reward for YOUR trainer. The reward is validated first (calibrated, non-drift-flagged, population-bound judges — refused with the reason otherwise), the budget is held in the wallet, and the response carries every judge's signed certificate so the trainer holds the grader's error rate before the first rollout is scored. Optional `anchor`: an independent judge the run is checked against mid-run (POST …/anchor). Requires an owner/admin key with platform:write. errorbar never runs the training — that is what makes the certificate independent. operationId: createRewardSession requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/CreateRewardSession' responses: '201': description: Created, with certificates. content: application/json: schema: $ref: '#/components/schemas/RewardSession' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '402': $ref: '#/components/responses/InsufficientBalance' '403': $ref: '#/components/responses/Forbidden' components: schemas: CreateRewardSession: type: object required: - reward - rewardBudgetUsd properties: reward: $ref: '#/components/schemas/RewardSpec' agentic: type: boolean default: false description: >- Whole-trajectory rewards need trace-unit judges; single-turn rewards need request-unit ones. rewardBudgetUsd: type: number description: >- Held in the wallet at creation; released on stop, expiry or exhaustion. Max 10000. ttlHours: type: number default: 72 description: >- Session expiry (max 720). Scoring past it is refused and the hold released. label: type: string maxLength: 120 anchor: type: object properties: criterionId: type: string description: >- An INDEPENDENT calibrated judge: a different criterion whose model or prompt differs from every reward judge, aligned ≥ 0.9, not drift-flagged, same unit. required: - criterionId tasks: type: array items: $ref: '#/components/schemas/RewardTask' description: >- Optional taskset (≤ 2,000). Stored on the session so `GET /v1/reward/sessions/{id}/environment?format=tasks` hands back tasks.jsonl. RewardSession: type: object properties: id: type: string status: type: string enum: - ACTIVE - COMPLETED - OVERBUDGET - STOPPED - FAILED description: >- OVERBUDGET: the reward budget's kill switch fired. STOPPED: stopped by you or expired. agentic: type: boolean reward: $ref: '#/components/schemas/RewardSpec' label: type: - string - 'null' budget_usd: type: number spent_usd: type: number description: Ledger-true judge spend so far. expires_at: type: string format: date-time created_at: type: string format: date-time stopped_at: type: - string - 'null' format: date-time certificates: type: array items: type: object description: >- One signed judge certificate per reward criterion (see GET /v1/criteria/{id}/certificate): sensitivity, specificity, κ with intervals, trust, what voids it. anchor: oneOf: - $ref: '#/components/schemas/RewardSessionAnchor' - type: 'null' task_count: type: integer description: Tasks stored on the session (0 when the trainer keeps its own). RewardSpec: oneOf: - type: object required: - mode - criterionId properties: mode: type: string enum: - single criterionId: type: string description: A calibrated judge (TPR and TNR ≥ 0.9, not drift-flagged). - type: object required: - mode - criterionIds properties: mode: type: string enum: - compositional criterionIds: type: array items: type: string minItems: 2 description: >- Two or more calibrated judges (each ≥ 0.75); the composite grade is their mean with any assertions. assertions: type: array items: type: object description: >- Deterministic components (json_schema, regex_match, tool_called, exec …), 0/1 each. RewardTask: type: object required: - goal properties: goal: type: array items: type: object properties: role: type: string content: type: string description: The task as chat messages (1..50). image: type: string description: Optional sandbox image. verifier: type: array items: type: object properties: command: type: string timeout_sec: type: integer description: >- Per-task exec verifiers (≤ 8), run server-side as deterministic reward components. RewardSessionAnchor: type: object properties: criterion_id: type: string criterion_name: type: string held: type: boolean held_at: type: - string - 'null' format: date-time held_reason: type: - string - 'null' pinned_step: type: - string - 'null' description: The step of the best anchor point — the checkpoint to ship. checks: type: integer history: type: array items: $ref: '#/components/schemas/AnchorPoint' description: The last 20 checks. 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. 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. 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 Forbidden: description: The key lacks the required owner/admin permission. 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 `. ````