---
title: "Get bandit results"
method: GET
path: "/api/v2/experiments/bandits/{bandit_id}/results"
tags: ["Experiments"]
---

# Get bandit results

`GET /api/v2/experiments/bandits/{bandit_id}/results`

## Request

- Base URL: `https://api.datadoghq.com`
- URL: `https://api.datadoghq.com/api/v2/experiments/bandits/{bandit_id}/results`
- Auth: one of:
  - API key in header `DD-API-KEY` and API key in header `DD-APPLICATION-KEY`
  - OAuth 2
    - Authorization code: `/oauth2/v1/authorize`
    - Token URL: `/oauth2/v1/token` scopes: `product_analytics_experiments_read`

Get what a bandit's refreshes have learned: each arm's estimate and probability of being best from the latest saved refresh, and each arm's point estimate over time. `filter[from]` and `filter[to]` narrow only `point_estimates`, not the latest estimates. For observation counts, use Get bandit traffic summary. Bandits must be enabled for the organization.

## Path parameters

- `bandit_id` string, uuid, required: The UUID of the bandit.

## Query parameters

- `filter[from]` string, date-time: Return only point estimates from refreshes saved at or after this RFC3339 timestamp. Inclusive.
- `filter[to]` string, date-time: Return only point estimates from refreshes saved before this RFC3339 timestamp. Exclusive.

## Response `200`

OK

- ExperimentsBanditResultsV2DTO: Response containing a bandit's results.
  - `data` ExperimentsBanditResultsV2DTOData, required: JSON:API resource containing a bandit's results.
    - `attributes` ExperimentsBanditResultsV2DTODataAttributes, required: What the bandit's refreshes have learned.
      - `confidence_level` number, double, required: The probability mass of each confidence interval.
      - `config_version` integer, nullable, required: The configuration version the latest saved refresh used. It can be older than the bandit's, because changes such as a new end date don't invalidate results.
      - `data_as_of` string, date-time, nullable, required: The timestamp of the most recent warehouse data that the latest saved refresh read.
      - `desired_change` 'METRIC_INCREASES' | 'METRIC_DECREASES', nullable, required: The direction the bandit optimizes the metric in. It's fixed when the bandit starts. `null` before the first saved refresh.
      - `next_refresh_at` string, date-time, nullable, required: An estimate of when the next scheduled refresh starts. `null` when the bandit isn't running or the time can't be estimated. Only daily and hourly schedules get an estimate.
      - `optimal_variant_key` string, nullable, required: The key of the arm most likely to be best, when one arm leads and the bandit is outside warmup. `null` otherwise.
      - `optimal_variant_unavailable_reason` 'NO_RESULTS_YET' | 'WARMUP' | 'NOT_ENOUGH_DATA' | 'NO_DIFFERENCE' | 'STATS_UNAVAILABLE': Why no optimal variant is reported. Absent when one is, or when the cause isn't known.
      - `point_estimates` ExperimentsBanditPointEstimateV2[], required: Each arm's point estimate from up to the latest 400 saved refreshes in the requested range, oldest first.
        - `confidence_interval` ExperimentsBanditConfidenceIntervalV2, required: A central interval around point_estimate at confidence_level.
          - `lower` number, double, required: Lower bound of the interval.
          - `upper` number, double, required: Upper bound of the interval.
        - `data_as_of` string, date-time, nullable, required: The timestamp of the most recent warehouse data that the refresh read.
        - `feature_flag_variant_id` string, uuid, required: The arm's variant ID on the flag.
        - `point_estimate` number, double, required: The arm's estimated mean of the metric, in the metric's units.
        - `refresh_id` string, uuid, required: The refresh the estimate comes from.
        - `results_last_updated` string, date-time, required: When the refresh saved its results.
        - `variant_key` string, nullable, required: The arm's variant key.
      - `refresh_id` string, uuid, nullable, required: The latest refresh that saved results. The estimates come from it.
      - `results_last_updated` string, date-time, nullable, required: When the latest saved refresh saved its results.
      - `state` 'WAITING_FOR_FIRST_REFRESH' | 'WARMUP' | 'OPTIMIZING' | 'STATS_UNAVAILABLE' | 'ENDED', required: `WAITING_FOR_FIRST_REFRESH` until a refresh saves results. `STATS_UNAVAILABLE` when the latest saved refresh has no complete probabilities of being optimal. `ENDED` after the bandit is stopped or completed, and its last estimates are kept. Otherwise, `WARMUP` or `OPTIMIZING` indicates which policy the bandit serves. A paused bandit keeps its state.
      - `statistics_status` 'calculated' | 'failed': Whether the latest saved refresh calculated its statistics. Absent for older refreshes.
      - `variants` ExperimentsBanditVariantResultV2[], required: One row per arm, in arm order.
        - `confidence_interval` ExperimentsBanditNullableConfidenceIntervalV2, nullable, required: A central interval around point_estimate at confidence_level. `null` when the estimate is unavailable.
          - `lower` number, double, required: Lower bound of the interval.
          - `upper` number, double, required: Upper bound of the interval.
        - `data_status` 'usable' | 'insufficient_data' | 'zero_variance' | 'pooled_within_variance' | 'pooled_total_variance': Where the arm's reward variance came from, or why it's unavailable.
        - `feature_flag_variant_id` string, uuid, required: The arm's variant ID on the flag.
        - `mature_observations` integer, required: The arm's valid observations whose aggregation window has closed. The estimate uses these.
        - `point_estimate` number, double, nullable, required: The arm's estimated mean of the metric, in the metric's units. It isn't a lift.
        - `posterior_unavailable_reason` 'no_data' | 'insufficient_data' | 'zero_variance': Why the arm's estimate and interval are unavailable. Absent when they are available.
        - `probability_of_being_optimal` number, double, nullable, required: The probability that this arm is the best.
        - `variant_key` string, nullable, required: The arm's variant key.
    - `id` string, uuid, required: The bandit ID.
    - `type` 'bandit-results', required: Bandit results resource type.

## Other responses

- `400`: Malformed bandit ID (not a valid UUID), or filter[from] or filter[to] isn't an RFC3339 timestamp, or filter[from] isn't before filter[to].
- `401`: Missing or invalid authentication (dd-api-key + dd-application-key headers, or a valid user session).
- `403`: Bandits are not enabled for the organization, or the caller lacks the product_analytics_experiments_read permission.
- `404`: No bandit with this ID exists for the organization.
- `429`: Too many requests

## Changes

- **2026-11-17** `633e476d0d57`: 1 info
  - endpoint added

[Operation history](https://skmtc.dev/datadog/apis/datadog-api-v2-collection/changes/api/v2/experiments/bandits/:bandit_id/results/get.md)

---

[API](https://skmtc.dev/datadog/apis/datadog-api-v2-collection.md) - [All operations](https://skmtc.dev/datadog/apis/datadog-api-v2-collection/llms.txt) - [OpenAPI document](https://skmtc.dev/datadog/apis/datadog-api-v2-collection/revisions/8d1a5e5973da?raw)
