Cognitive API · v1

Score thinking quality with one API call

The same six-dimension engine that powers NavikLab hiring, exposed as a usage-based API. Score any AI session — in your IDE, your eval pipeline, your product. This is the cognitive layer for all AI work.

Quickstart

Try it right now with the public demo key nvk_demo_key:

curl https://naviklab.com/api/v1/score \
  -H "Authorization: Bearer nvk_demo_key" \
  -H "Content-Type: application/json" \
  -d '{"transcript": "You: build a rate limiter\nAssistant: ...\nYou: no, that has a race condition, use a token bucket instead"}'

Response

{
  "version": "v1",
  "engine": "llm",
  "overall": 74,
  "dimensions": {
    "analytical_rigor":      { "score": 70, "level": "proficient", "signals": ["Identifies constraints"] },
    "creative_range":        { "score": 65, "level": "proficient", "signals": [] },
    "metacognitive_depth":   { "score": 80, "level": "advanced",   "signals": ["Corrects AI mistakes"] },
    "intellectual_humility": { "score": 85, "level": "advanced",   "signals": ["Pushes back on AI"] },
    "synthesis_ability":     { "score": 60, "level": "proficient", "signals": [] },
    "persistence":           { "score": 72, "level": "proficient", "signals": ["Recovers from dead ends"] }
  },
  "meta": { "iterations": 3, "decisions": 1, "branches": 1 }
}

Endpoint

POST /api/v1/score

Body: { transcript } (raw .jsonl or You:/Assistant: text) or a pre-parsed { trace }. Optional mode: auto (default, LLM judge with heuristic fallback), heuristic, or llm. The engine field reports which scored it.

Auth & pricing

Bearer key. Usage-based.

Demo key is for testing only. Create metered production keys (quota-enforced) in Dashboard → Cognitive API.

Get a production API key

Building evals, IDE tooling, or a product that needs to measure thinking? Let's talk.

Request access