---
title: "Get a run result"
method: GET
path: "/v1/runs/{id}/result"
tags: ["runs"]
---

# Get a run result

`GET /v1/runs/{id}/result`

## Request

- Base URL: `https://api.hyperresearch.ai`
- URL: `https://api.hyperresearch.ai/v1/runs/{id}/result`
- Auth: one of:
  - HTTP bearer
  - API key in cookie `hr_session`

Artifact URLs are short-lived signed Worker download URLs valid for one hour.

## Path parameters

- `id` string, required

## Response `200`

Result

- RunResult
  - `report_markdown` string, nullable, required
  - `report_url` string, nullable, required
  - `sources` unknown[], required
    - unknown
  - `skipped` unknown[], required: Every page this run tried and did not keep: `{ url, domain, outcome, reason, detail, rung, at }`. `reason` is a closed enum and is always set — `unknown` with the raw text in `detail` when the ladder recorded nothing to classify (E2E8-12). `sources` is what the run kept; this is the other half, so a reloaded run still explains the gap between pages considered and sources kept (E2E8-02).
    - unknown
  - `skipped_summary` string, nullable, required: The skipped list in one sentence — “30 pages considered, 21 kept; 9 were skipped: …” — written at finish so it survives a reload. Null for a run that finished before the artifact existed.
  - `claims` unknown[], required
    - unknown
  - `verification` unknown
  - `escalations` unknown[], required
    - unknown
  - `artifacts` object, required
  - `telemetry` unknown
  - `seat_models` object, nullable: Premium seats only: the model that really answered each promoted seat (planner, section writers, editor), with call counts and every fallback, and the receipt sentence ("Report planned, written and edited by Claude Opus 5.5."). Null for a run without premium seats.
    - `version` 1, required
    - `writer` 'phased' | 'phased-fell-back' | 'legacy' | 'light', required
    - `policy` 'base' | 'fail', required
    - `summary` string, required
    - `seats` object[], required
      - `seat` string, required
      - `label` string, required
      - `premium_model` string, required
      - `premium_name` string, required
      - `premium_calls` integer, required
      - `fallback_calls` integer, required
      - `models` object[], required
        - `model` string, required
        - `name` string, required
        - `calls` integer, required
        - `premium` boolean, required
    - `credit` object, nullable: The model credit the report may show ("Powered by Claude Opus 5.5"), set only when the accepted output of that seat really came from the model: the Light final edit that shipped, or the Deep plan that was used. Null on a fallback, a skipped or rejected edit, or a plan from the standard model. Absent on runs that finished before it existed.
      - `model` string, required
      - `role` 'edit' | 'plan', required
      - `label` string, required
      - `detail` string, required
    - `light_edit` 'accepted' | 'rejected' | 'skipped': Light only: how the final edit ended. `rejected` and `skipped` ship the checked report unchanged.
  - `report_language` string, nullable, required: The language the report was written in, as a BCP-47 primary subtag (`zh`, `en`, `ja`, …): the run’s `report_language` lever when one was set, otherwise the language of the question — the same rule, over the same frozen inputs, the run itself used. Null when it cannot be known (a run admitted before launch packets were stored). Clients label the report (source-list headings, dates) in this language rather than guessing from its text.
  - `attention` AttentionItem[], required: Findings a reader should act on before relying on the report, in the order to show them. Empty when there are none. Every other finding stays in `verification` as information.
    - `kind` 'unsupported-statements' | 'incomplete-chapters' | 'uncovered-question' | 'retracted-source' | 'unchecked-report', required
    - `check` string, required: The receipt check this finding came from, e.g. `citation-spot-check`.
    - `count` integer, required: How many: cited statements shipped uncorrected, chapters, parts of the question, retracted citations.
    - `of` integer, nullable, required: What `count` is out of (statements checked, parts of the question), or null when not stated.
    - `unit` 'figures' | 'statements': For `unsupported-statements`: whether the check counted cited figures or cited statements.
    - `names` string[], required: Chapter titles, question parts or the title of the retracted source. Empty when the receipt does not name them.
  - `flagged_statements` FlaggedStatement[], nullable, required: The cited statements the checks flagged that ship as written, at most 200. Null when the run kept no per-sentence record of its checks.
    - `text` string, required: The sentence as it ships in the report, at most 400 characters.
    - `verdict` 'wrong' | 'unsupported' | 'partly' | 'minor', required: What the check found: the source says otherwise, does not state it, states part of it, or differs in a detail.
    - `source_ids` string[], required: The ids of the sources the sentence cites.

## Other responses

- `304`: The result has not changed since the `ETag` the client holds. Sent for settled runs only.
- `400`: Invalid request
- `401`: Authentication failed
- `403`: Insufficient scope or plan
- `404`: Not found
- `429`: Rate limited or over quota
- `500`: Internal error

---

[API](https://skmtc.dev/hyperresearch/apis/hyperresearch-cloud-api.md) · [All operations](https://skmtc.dev/hyperresearch/apis/hyperresearch-cloud-api/llms.txt) · [OpenAPI document](https://skmtc.dev/hyperresearch/apis/hyperresearch-cloud-api/revisions/726ef831ebf4?raw)