Download run

Endpoint: GET /runs/{run_id}/download

Description Returns a CSV file of evaluation rows for the run. Column headers are dynamic: only columns that have at least one non-empty value are included. On trace runs, evaluations have no linked dataset record — record columns (external_id, prompt, input, context, output, golden_answer) are omitted when every row would be empty.

Possible columns (in this order when present):

Column	Description
`alias`	Run alias, when set
`external_id`	Caller-supplied record identifier, when set
`prompt`	Record prompt
`input`	Record input
`context`	Record context
`output`	Record output
`golden_answer`	Record golden answer
`explanation`	Evaluation explanation
`model_slug`	Model used for this evaluation
`is_success`	Whether the evaluation completed successfully
`evaluation_cost`	USD amount charged to your balance for this evaluation
`is_gte_threshold`	Whether the score met the threshold
`threshold`	Threshold for this evaluation
`result`	Evaluation score
`eval_metadata`	Metric-specific JSON metadata
`shortname`	Metric shortname (for example `ans_corr`)
`name`	Metric display name

Parse the file by header name, not by column position: which columns appear depends on the run's data.

Value formats

Booleans (is_success, is_gte_threshold) are written as True / False.
result is a decimal number, for example 100.0 or 87.5.
threshold is an integer, for example 80.
evaluation_cost is written with 15 decimal places, for example 0.000412000000000.
eval_metadata and any structured input / context / output / golden_answer values are JSON strings.
An empty cell means the value is not set for that row (for example result on an evaluation that did not complete).
Rows end with \r\n. Cells containing commas, quotes, or line breaks are wrapped in double quotes, with inner quotes doubled.

Parameters

Path: run_id — integer run id.

Error responses

401 — Authentication failed.
404 — Run not found or not accessible to your API key (including not assigned to the key's project), or no evaluation rows for that run (CSV would be empty).
500 — Error building the CSV.

Responses

200 — Content-Type: text/csv, file name like run_<run_id>_report.csv.

Example response (200)

alias,external_id,prompt,output,golden_answer,explanation,model_slug,is_success,evaluation_cost,is_gte_threshold,threshold,result,eval_metadata,shortname,name
qa-regression,row-42,What is 2+2?,4,4,The output matches the golden answer.,o4-mini,True,0.000412000000000,True,80,100.0,"{""statements_not_covered_in_golden_answer"": [], ""statements_not_covered_in_output"": []}",ans_corr,Answer Correctness
qa-regression,row-43,Capital of Germany?,Munich,Berlin,"The output names Munich, but the golden answer is Berlin.",o4-mini,True,0.000398000000000,False,80,0.0,"{""statements_not_covered_in_golden_answer"": [{""statement"": ""The capital of Germany is Munich."", ""reason"": ""Contradicts the golden answer.""}], ""statements_not_covered_in_output"": [{""statement"": ""The capital of Germany is Berlin."", ""reason"": ""Not stated in the output.""}]}",ans_corr,Answer Correctness

In this run no row has input or context, so those columns are omitted.

curl

curl -o "run_1_report.csv" \
  "https://api.aegisevals.ai/api/v1/runs/1/download" \
  -H "Authorization: Bearer sk_00000000000000000000000000000000"