> ## Documentation Index
> Fetch the complete documentation index at: https://docs.infercrane.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Compare the baseline and the winning setup of one campaign on cost, speed, and answer quality



## OpenAPI

````yaml /openapi.json get /api/v1/optimization/campaigns/{id}/report
openapi: 3.1.0
info:
  description: >-
    Durable, authenticated control-plane operations. Mutation responses may
    outlive the requesting client.
  title: InferCrane Control API
  version: 1.0.0
servers:
  - description: Local InferCrane API and gateway
    url: http://127.0.0.1:18000
security: []
tags:
  - name: Admission
  - name: Adoption
  - name: Agents
  - name: Alerts
  - name: Async inference
  - name: Audit
  - name: Autopilot
  - name: Benchmarks
  - name: Billing
  - name: Billing administration
  - name: Burst Guard
  - name: Capacity intelligence
  - name: Compute
  - name: Context Passport
  - name: Deployments
  - name: Diagnostics
  - name: Endpoints
  - name: External capacity
  - name: FinOps
  - name: Identity
  - name: Inference
  - name: Inference Lab
  - name: Inference decisions
  - name: Inference optimization
  - name: Infrastructure
  - name: Marketplace administration
  - name: Model API administration
  - name: Model APIs
  - name: Model artifacts
  - name: Model catalog
  - name: Monitoring
  - name: Notifications
  - name: Operations
  - name: Planning
  - name: Providers
  - name: Recipes
  - name: Release Guard
  - name: Release evidence
  - name: Replay
  - name: Revisions
  - name: Routing
  - name: Sandboxes
  - name: Scaling
  - name: Secrets
  - name: System
  - name: Targets
  - name: Training lineage
paths:
  /api/v1/optimization/campaigns/{id}/report:
    get:
      tags:
        - Inference optimization
      summary: >-
        Compare the baseline and the winning setup of one campaign on cost,
        speed, and answer quality
      operationId: getOptimizationCampaignReport
      parameters:
        - in: path
          name: id
          required: true
          schema:
            minLength: 1
            type: string
      responses:
        '200':
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/OptimizationCampaignReportEnvelope'
          description: Success
        default:
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorEnvelope'
          description: Typed API error
      security:
        - bearerAuth: []
components:
  schemas:
    OptimizationCampaignReportEnvelope:
      additionalProperties: false
      properties:
        report:
          $ref: '#/components/schemas/OptimizationCampaignReport'
      required:
        - report
      type: object
    ErrorEnvelope:
      properties:
        error:
          $ref: '#/components/schemas/Error'
      required:
        - error
      type: object
    OptimizationCampaignReport:
      additionalProperties: false
      properties:
        algorithm_version:
          type: string
        appendix:
          additionalProperties: false
          properties:
            baseline:
              oneOf:
                - additionalProperties: false
                  properties:
                    accelerator:
                      type: string
                    accelerator_class:
                      description: simulated_cpu numbers are never GPU results.
                      enum:
                        - gpu
                        - simulated_cpu
                        - unknown
                      type: string
                    accelerator_count:
                      type:
                        - integer
                        - 'null'
                    args:
                      items:
                        type: string
                      type: array
                    candidate_id:
                      type: string
                    compute_mode:
                      type: string
                    deployment_name:
                      type: string
                    engine:
                      type: string
                    engine_version:
                      type: string
                    env:
                      additionalProperties:
                        type: string
                      description: >-
                        Reviewed engine environment switches the setup ran with;
                        absent when none were set.
                      type: object
                    hourly_usd_per_replica:
                      description: >-
                        Sourced hourly price divided by the quoted maximum
                        replica count; null without a sourced price.
                      type:
                        - number
                        - 'null'
                    image:
                      type: string
                    label:
                      type: string
                    max_replicas:
                      type: integer
                    min_replicas:
                      type: integer
                    model:
                      type: string
                    price_source:
                      type: string
                    provider:
                      type: string
                    region:
                      type: string
                    revision_id:
                      type: string
                  required:
                    - candidate_id
                    - label
                    - model
                    - engine
                    - engine_version
                    - args
                    - accelerator
                    - accelerator_count
                    - accelerator_class
                    - provider
                    - min_replicas
                    - max_replicas
                    - hourly_usd_per_replica
                  type: object
                - type: 'null'
            evidence:
              items:
                additionalProperties: false
                properties:
                  digest:
                    type: string
                  id:
                    type: string
                  kind:
                    type: string
                  subject:
                    type: string
                required:
                  - kind
                  - digest
                type: object
              type: array
            winner:
              oneOf:
                - additionalProperties: false
                  properties:
                    accelerator:
                      type: string
                    accelerator_class:
                      description: simulated_cpu numbers are never GPU results.
                      enum:
                        - gpu
                        - simulated_cpu
                        - unknown
                      type: string
                    accelerator_count:
                      type:
                        - integer
                        - 'null'
                    args:
                      items:
                        type: string
                      type: array
                    candidate_id:
                      type: string
                    compute_mode:
                      type: string
                    deployment_name:
                      type: string
                    engine:
                      type: string
                    engine_version:
                      type: string
                    env:
                      additionalProperties:
                        type: string
                      description: >-
                        Reviewed engine environment switches the setup ran with;
                        absent when none were set.
                      type: object
                    hourly_usd_per_replica:
                      description: >-
                        Sourced hourly price divided by the quoted maximum
                        replica count; null without a sourced price.
                      type:
                        - number
                        - 'null'
                    image:
                      type: string
                    label:
                      type: string
                    max_replicas:
                      type: integer
                    min_replicas:
                      type: integer
                    model:
                      type: string
                    price_source:
                      type: string
                    provider:
                      type: string
                    region:
                      type: string
                    revision_id:
                      type: string
                  required:
                    - candidate_id
                    - label
                    - model
                    - engine
                    - engine_version
                    - args
                    - accelerator
                    - accelerator_count
                    - accelerator_class
                    - provider
                    - min_replicas
                    - max_replicas
                    - hourly_usd_per_replica
                  type: object
                - type: 'null'
          required:
            - winner
            - baseline
            - evidence
          type: object
        banner:
          description: >-
            Fixed text every renderer must show: present only for simulated or
            CPU runs.
          type: string
        bottleneck:
          additionalProperties: false
          properties:
            gpu_telemetry:
              type: string
            reason:
              type: string
            signals:
              items:
                additionalProperties: false
                properties:
                  baseline:
                    type:
                      - number
                      - 'null'
                  name:
                    type: string
                  stat:
                    type: string
                  unit:
                    type: string
                  winner:
                    type:
                      - number
                      - 'null'
                required:
                  - name
                  - stat
                  - baseline
                  - winner
                type: object
              type: array
            source:
              type: string
            state:
              enum:
                - measured
                - unknown
              type: string
          required:
            - state
            - source
            - gpu_telemetry
            - signals
          type: object
        campaign_id:
          type: string
        campaign_state:
          type: string
        candidates:
          items:
            additionalProperties: false
            properties:
              cost_per_million_output_tokens:
                type:
                  - number
                  - 'null'
              error_rate_percent:
                description: >-
                  Share of the setup's measured requests that failed, in
                  percent. Null when not recorded.
                type:
                  - number
                  - 'null'
              failure_code:
                type: string
              guards:
                description: >-
                  Why the setup cannot be chosen whatever it gained: it failed
                  more requests than the baseline, or answered more than 5%
                  shorter.
                items:
                  additionalProperties: false
                  properties:
                    baseline:
                      type: number
                    basis:
                      description: >-
                        same_window compares with the baseline runs measured in
                        turns with the setup; separate_runs with the baseline's
                        own runs.
                      enum:
                        - same_window
                        - separate_runs
                      type: string
                    code:
                      enum:
                        - measured_error_rate_regression
                        - measured_output_length_regression
                      type: string
                    metric:
                      enum:
                        - error_rate_percent
                        - output_tokens_per_request
                      type: string
                    reason:
                      type: string
                    value:
                      type: number
                  required:
                    - code
                    - metric
                    - baseline
                    - value
                    - basis
                    - reason
                  type: object
                type: array
              id:
                type: string
              label:
                type: string
              latency_p95_ms:
                type:
                  - number
                  - 'null'
              output_tokens_per_request:
                description: >-
                  Mean server-counted output tokens of a successful request in
                  the setup's own runs. Null when any run did not record it.
                type:
                  - number
                  - 'null'
              output_tokens_per_second:
                type:
                  - number
                  - 'null'
              reason:
                type: string
              regressions:
                description: >-
                  Headline outcomes (throughput, p95 time to first token, p95
                  latency) on which the setup measured worse than the baseline
                  in its own window.
                items:
                  additionalProperties: false
                  properties:
                    basis:
                      description: >-
                        Set on a regression: drifted_control_runs is read from
                        runs whose control drifted and can show a setup is
                        worse, never better.
                      enum:
                        - interleaved_control
                        - drifted_control_runs
                      type: string
                    change_percent:
                      type: number
                    decision:
                      enum:
                        - improved
                        - regressed
                        - no_measurable_difference
                        - unchanged
                        - unknown
                      type: string
                    metric:
                      type: string
                    noise_floor_percent:
                      type: number
                  required:
                    - metric
                    - change_percent
                    - noise_floor_percent
                    - decision
                  type: object
                type: array
              requests_per_second:
                type:
                  - number
                  - 'null'
              role:
                enum:
                  - baseline
                  - baseline_kept
                  - winner
                  - candidate
                type: string
              state:
                type: string
              tpot_p95_ms:
                type:
                  - number
                  - 'null'
              ttft_p95_ms:
                type:
                  - number
                  - 'null'
              versus_baseline:
                oneOf:
                  - additionalProperties: false
                    properties:
                      basis:
                        description: >-
                          Set on a regression: drifted_control_runs is read from
                          runs whose control drifted and can show a setup is
                          worse, never better.
                        enum:
                          - interleaved_control
                          - drifted_control_runs
                        type: string
                      change_percent:
                        type: number
                      decision:
                        enum:
                          - improved
                          - regressed
                          - no_measurable_difference
                          - unchanged
                          - unknown
                        type: string
                      metric:
                        type: string
                      noise_floor_percent:
                        type: number
                    required:
                      - metric
                      - change_percent
                      - noise_floor_percent
                      - decision
                    type: object
                  - type: 'null'
            required:
              - id
              - label
              - role
              - state
              - output_tokens_per_second
              - requests_per_second
              - ttft_p95_ms
              - tpot_p95_ms
              - latency_p95_ms
              - cost_per_million_output_tokens
              - error_rate_percent
              - output_tokens_per_request
              - versus_baseline
            type: object
          type: array
        finished:
          type: boolean
        intent:
          type: string
        metrics:
          items:
            additionalProperties: false
            properties:
              baseline:
                type:
                  - number
                  - 'null'
              basis:
                enum:
                  - interleaved_control
                  - separate_runs
                  - baseline_kept
                  - unknown
                type: string
              change_percent:
                description: Winner relative to baseline, in percent.
                type:
                  - number
                  - 'null'
              decision:
                enum:
                  - improved
                  - regressed
                  - no_measurable_difference
                  - unchanged
                  - unknown
                type: string
              higher_is_better:
                type: boolean
              key:
                enum:
                  - cost_per_million_output_tokens
                  - ttft_p95_ms
                  - output_tokens_per_second_per_user
                  - output_tokens_per_second_per_accelerator
                  - requests_per_second
                type: string
              label:
                type: string
              noise_floor_percent:
                description: >-
                  Noise floor from the interleaved control runs; null without
                  them.
                type:
                  - number
                  - 'null'
              reason:
                type: string
              unit:
                type: string
              winner:
                type:
                  - number
                  - 'null'
            required:
              - key
              - label
              - unit
              - higher_is_better
              - baseline
              - winner
              - change_percent
              - noise_floor_percent
              - decision
              - basis
            type: object
          type: array
        model_identity:
          type: string
        monthly:
          additionalProperties: false
          properties:
            assumption:
              type: string
            baseline_usd:
              type:
                - number
                - 'null'
            basis:
              enum:
                - replayed_traffic
                - unknown
              type: string
            capture_cut_off:
              description: >-
                True when the capture reached its request limit and kept only
                its newest requests. Null for a capture that did not record it;
                such a projection may be too low.
              type:
                - boolean
                - 'null'
            capture_window_seconds:
              description: The whole capture window.
              type:
                - number
                - 'null'
            change_percent:
              type:
                - number
                - 'null'
            change_usd:
              type:
                - number
                - 'null'
            decision:
              enum:
                - improved
                - regressed
                - no_measurable_difference
                - unchanged
                - unknown
              type: string
            hours_per_month:
              type: number
            monthly_output_tokens:
              type:
                - number
                - 'null'
            output_tokens_in_window:
              type:
                - integer
                - 'null'
            reason:
              type: string
            requests_in_window:
              type:
                - integer
                - 'null'
            trace_digest:
              type: string
            window_seconds:
              description: >-
                Time the replayed volume is divided by: the capture window, or,
                when the capture was cut off at its request limit, the span from
                the first to the last kept request.
              type:
                - number
                - 'null'
            winner_usd:
              type:
                - number
                - 'null'
          required:
            - basis
            - window_seconds
            - capture_window_seconds
            - capture_cut_off
            - requests_in_window
            - output_tokens_in_window
            - monthly_output_tokens
            - hours_per_month
            - baseline_usd
            - winner_usd
            - change_usd
            - change_percent
            - decision
          type: object
        noise:
          additionalProperties: false
          properties:
            basis:
              enum:
                - interleaved_control
                - unknown
              type: string
            candidate_id:
              type: string
            candidate_runs:
              type: integer
            control_drift_percent:
              type:
                - number
                - 'null'
            control_runs:
              type: integer
            drift_limit_percent:
              type:
                - number
                - 'null'
            objective_change_percent:
              type:
                - number
                - 'null'
            objective_decision:
              enum:
                - improved
                - regressed
                - no_measurable_difference
                - unchanged
                - unknown
              type: string
            objective_metric:
              type: string
            objective_noise_floor_percent:
              type:
                - number
                - 'null'
            reason:
              type: string
          required:
            - basis
            - control_runs
            - candidate_runs
            - control_drift_percent
            - drift_limit_percent
            - objective_change_percent
            - objective_noise_floor_percent
            - objective_decision
          type: object
        objective:
          type: string
        outcome:
          additionalProperties: false
          properties:
            activated:
              type: boolean
            baseline_candidate_id:
              type: string
            baseline_label:
              type: string
            headline:
              type: string
            kind:
              enum:
                - winner_selected
                - baseline_kept
                - no_winner
                - in_progress
                - cancelled
              type: string
            winner_candidate_id:
              type: string
            winner_label:
              type: string
          required:
            - kind
            - headline
            - baseline_candidate_id
            - baseline_label
            - winner_candidate_id
            - winner_label
            - activated
          type: object
        quality:
          additionalProperties: false
          properties:
            baseline_score:
              type:
                - number
                - 'null'
            basis:
              enum:
                - release_guard
                - campaign_quality_gate
                - builtin_answer_check
                - baseline_kept
                - unknown
              type: string
            block_threshold_percent:
              type:
                - number
                - 'null'
            delta:
              type:
                - number
                - 'null'
            eval:
              type: string
            evals:
              items:
                additionalProperties: false
                properties:
                  baseline_score:
                    type:
                      - number
                      - 'null'
                  delta:
                    type:
                      - number
                      - 'null'
                  evaluator:
                    type: string
                  follow_up:
                    type: string
                  message:
                    type: string
                  regression_percent:
                    type:
                      - number
                      - 'null'
                  suite:
                    type: string
                  verdict:
                    enum:
                      - PASS
                      - WATCH
                      - BLOCK
                      - UNKNOWN
                    type: string
                  winner_score:
                    type: number
                required:
                  - suite
                  - evaluator
                  - verdict
                  - baseline_score
                  - winner_score
                  - delta
                  - regression_percent
                  - message
                type: object
              type: array
            reason:
              type: string
            regression_percent:
              type:
                - number
                - 'null'
            release_guard_evaluation_id:
              type: string
            source:
              description: Who produced the winner's deciding evidence.
              enum:
                - infercrane
                - customer
              type: string
            verdict:
              enum:
                - PASS
                - WATCH
                - BLOCK
                - UNKNOWN
              type: string
            watch_threshold_percent:
              type:
                - number
                - 'null'
            winner_score:
              type:
                - number
                - 'null'
          required:
            - verdict
            - basis
            - baseline_score
            - winner_score
            - delta
            - regression_percent
            - watch_threshold_percent
            - block_threshold_percent
            - evals
          type: object
        report_digest:
          pattern: ^sha256:[a-f0-9]{64}$
          type: string
        schema_version:
          const: infercrane.optimization-report/v1
          type: string
        simulated_accelerator:
          description: >-
            True when any setup ran on a simulated accelerator or a CPU. Its
            numbers prove the pipeline, not GPU speed.
          type: boolean
        unknown:
          description: Paths of report fields whose input is missing or incomparable.
          items:
            type: string
          type: array
        warnings:
          description: >-
            One line per setup that failed more requests or answered shorter
            than the baseline. A speed or cost gain may come from that.
          items:
            type: string
          type: array
        workload:
          additionalProperties: false
          properties:
            concurrency:
              type:
                - integer
                - 'null'
            method:
              type: string
            profile:
              type: string
            requests_per_run:
              type:
                - integer
                - 'null'
            source:
              type: string
          required:
            - source
            - concurrency
            - requests_per_run
          type: object
      required:
        - schema_version
        - algorithm_version
        - campaign_id
        - model_identity
        - objective
        - intent
        - campaign_state
        - finished
        - simulated_accelerator
        - outcome
        - metrics
        - quality
        - noise
        - monthly
        - workload
        - candidates
        - warnings
        - bottleneck
        - appendix
        - unknown
        - report_digest
      type: object
    Error:
      properties:
        category:
          type: string
        code:
          type: string
        message:
          type: string
        remediation:
          type: string
        request_id:
          type: string
        retryable:
          type: boolean
      required:
        - code
        - message
      type: object
  securitySchemes:
    bearerAuth:
      scheme: bearer
      type: http

````

This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.