API Reference

The OCS scoring API lets you score divergent-thinking responses programmatically. Requests are application/x-www-form-urlencoded; responses are JSON. No authentication is required for moderate use; an X-API-KEY header lifts rate limits (email peter.organisciak@du.edu for one).

The same endpoint powers the website's scoring forms, so anything you can do on this site you can also do from Python, R, JavaScript, curl, or any other HTTP client.

POST`https://openscoring.du.edu/llm`

Score one or more divergent-thinking responses with an LLM-based model.

Try it

Input (CSV) *

Model

Task

Language

Prompt (default)

Question (default)

Prompt in input

Question in input

Elaboration scoring

Logprob (weighted) scoringrecommended

X-API-KEY (optional)

Sample code

curl -X POST 'https://openscoring.du.edu/llm' \
  --data-urlencode 'input=brick,door stop
brick,plant pot
brick,fly into space' \
  --data-urlencode 'model=ocsai2' \
  --data-urlencode 'task=uses' \
  --data-urlencode 'language=eng' \
  --data-urlencode 'elab_method=none' \
  --data-urlencode 'logprob_scoring=true'

Response

No request sent yet. Use the form on the left.

Parameters

Name	Type	Required	Default	Description
input	string (CSV)	yes	brick,door stop brick,plant pot …	CSV-encoded rows. By default the first column is the prompt and the second is the response. Newlines separate rows; commas separate cells. Standard CSV quoting applies.
model	string	yes	ocsai2	Which scoring model to use. See the live list below. Defaults to "ocsai2" (recommended).
task	string	no	uses	Divergent-thinking task type (e.g. "uses", "consequences", "instances", "metaphors", "completion"). Models advertise the tasks they were trained on; pass "CUSTOM" or a free-text value if the model allows it.
language	string	no	eng	ISO 639-3 code (e.g. "eng", "spa", "fre"). Models advertise supported languages; only ocsai2 is multilingual.
prompt	string	no	—	If all rows share one prompt, set it here and disable Prompt-in-input below. Otherwise leave blank and let the first CSV column carry the prompt.
question	string	no	—	Optional plain-text task question (e.g. 'What would happen if rain were soda?'). Provides extra context to the scorer. Only ocsai-1.5 and later.
prompt_in_input	boolean	no	true	When true (default), the first CSV column is the prompt. Set false if you supply a single `prompt` parameter above and your CSV has only responses.
question_in_input	boolean	no	false	When true, an extra CSV column carries the per-row task question (between prompt and response). Default false.
elab_method	"none" \| "whitespace"	no	none	How to compute elaboration alongside originality. 'whitespace' adds an `elaboration` field equal to the word count of the response.
logprob_scoring	boolean	no	true	When true, return a weighted-average score using the top OpenAI logprobs (more stable than the single top-token score), along with a confidence value. Recommended, and on by default (the API also defaults to true when omitted). Not supported by ocsai-1.5 — disabled and forced off for those models.
X-API-KEYheader	string (header)	no	—	Optional API key for higher rate limits. Sent as the X-API-KEY HTTP header. Email peter.organisciak@du.edu to request one.

Available models

Generated from config.yaml. Deprecated models are hidden.

Name	Description	Tasks	Languages
ocsai2recommended	Cross-lingual, multi-task, gpt-4.1-mini base	uses, completion, consequences, instances, metaphors, CUSTOM	ara, chi, dut, eng, fre, ger, heb, ita, pol, rus, spa, CUSTOM
ocsai2-paper	Paper model, gpt-4.1-mini base	uses, completion, consequences, instances, metaphors, CUSTOM	ara, chi, dut, eng, fre, ger, heb, ita, pol, rus, spa, CUSTOM
ocsai2-paper-xs	Paper model, gpt-4.1-nano base	uses, completion, consequences, instances, metaphors, CUSTOM	ara, chi, dut, eng, fre, ger, heb, ita, pol, rus, spa, CUSTOM
ocsai-1.6	Ocsai 1.5, gpt-4o-mini base	uses, completion, consequences, instances, metaphors, CUSTOM	ara, chi, dut, eng, fre, ger, heb, ita, pol, rus, spa, CUSTOM
ocsai1-4o	Ocsai 1, English-focused, gpt-4o-mini base	uses, completion, consequences, instances	—

Response shape

{
  "parameters": { "model": "...", "language": "...", "task": "...", ... },
  "scores": [
    {
      "prompt": "brick",
      "response": "door stop",
      "originality": 1.4,
      "confidence": null,
      "flags": null
    }
    // …one entry per input row, in order
  ],
  "summative": {
    "n_examples": 3,
    "originality": { "mean": 1.7, "max": 2.3, "min": 1.4 }
    // "elaboration": {...}     when elab_method=whitespace
    // "by_participant": [...]  when input had a participant_id column
  },
  "version": "...",
  "cite": [ "...", "..." ]
}

GET`https://openscoring.du.edu/about`

Service metadata (description, author, contact). No parameters.

curl 'https://openscoring.du.edu/about'

Notes

Requests are rate-limited by IP; an X-API-KEY lifts limits substantially. Bulk scoring jobs are best served via API key.
Identical requests are cached server-side keyed by (prompt, response, language, task, model), so re-scoring the same data is free.
Found a bug or want a code sample for a language not listed? Email me.