When the answer has to be right.

Certainize turns multi-engine AI + human expert verification into something you can build on: a live Certainty Score, independent model certification, and — new — human-verified ground-truth data to train & verify specialized models — for any team fine-tuning or evaluating a vertical model, frontier labs included. One API. Powered by ReallySolved™.

# Quick start — fetch a brand's Resolution Score
curl -H "Authorization: Bearer YOUR_API_KEY" \
  https://api.certainize.ai/v1/brands/samsung/score

{
  "brand": "samsung",
  "resolution_score": 73,
  "open_truths": 3,
  "last_updated": "2026-05-24T14:22:00Z",
  "resolver_count": 22,
  "status": "active"
}

Everything you can do with Certainize

One capability — multi-engine AI answers, settled by human experts — powers 4 products. Enterprises verify and certify; support teams catch errors before customers do; AI labs buy the verified data to train on. Pick your path:

The Certainty Score

One number, 0–100, that tells a person — or a system — how far to trust an AI answer. It's built from 5 weighted sub-scores, so you can act on AI output with a known confidence level instead of a guess.

Who we serve

Why verification pays for itself

Authentication

All API requests require an Authorization header with a Bearer token. API keys are issued when you subscribe via reallysolved.com/enterprise.

Authorization: Bearer ck_live_XXXXXXXXXXXX

curl

curl -H "Authorization: Bearer ck_live_XXXXXXXXXXXX" \
  https://api.certainize.ai/v1/brands/openai/score

Python

import requests

headers = {"Authorization": "Bearer ck_live_XXXXXXXXXXXX"}
r = requests.get(
    "https://api.certainize.ai/v1/brands/openai/score",
    headers=headers
)
data = r.json()
print(data["resolution_score"])

Node.js

const res = await fetch("https://api.certainize.ai/v1/brands/openai/score", {
  headers: { "Authorization": "Bearer ck_live_XXXXXXXXXXXX" }
});
const data = await res.json();
console.log(data.resolution_score);

Endpoints reference

Base URL: https://api.certainize.ai

GET /v1/brands/{brand}

Returns the current Resolution Score and summary metadata for a brand. The brand path parameter is the brand's lowercase slug (e.g. samsung, openai, binance).

{
  "brand": "samsung",
  "display_name": "Samsung",
  "resolution_score": 73,
  "score_band": "moderate",
  "open_truths": 3,
  "resolved_truths": 41,
  "resolver_count": 22,
  "last_updated": "2026-05-24T14:22:00Z",
  "status": "active",
  "embed_badge_url": "https://reallysolved.com/api/embed/brand/samsung.svg"
}

Score bands

GET /v1/brands/{brand}/truths

Returns the list of active Truths filed against a brand. Each Truth represents a specific factual claim that expert Expert Solvers are voting on.

{
  "brand": "openai",
  "truths": [
    {
      "id": "truth_8kxp3n",
      "claim": "OpenAI's GPT-4 safety evals were rushed before launch",
      "status": "open",
      "resolver_votes": 14,
      "consensus_pct": 71,
      "filed_at": "2026-05-10T09:14:00Z"
    }
  ],
  "next_cursor": "eyJpZCI6Ijg...",
  "total": 7
}

GET /v1/score-feed

Returns a chronological list of recent score change events across all brands. Use this endpoint to build monitoring dashboards or trigger alerts on score movements.

{
  "events": [
    {
      "brand": "binance",
      "score_before": 58,
      "score_after": 51,
      "delta": -7,
      "trigger": "truth_resolved",
      "truth_id": "truth_4mzq9w",
      "timestamp": "2026-05-24T13:05:00Z"
    }
  ],
  "total": 1
}

GET /v1/resolvers/{handle}

Returns public profile data for an Expert Solver — a human expert who votes on Truths. The handle is the Expert Solver's public username on reallysolved.com.

{
  "handle": "mkbhd",
  "tier": "Senior Solver",
  "riq": 8420,
  "specializations": ["consumer-tech", "ev"],
  "truths_resolved": 312,
  "accuracy_rate": 0.94,
  "badge_url": "https://reallysolved.com/api/embed/resolver/mkbhd.svg",
  "profile_url": "https://reallysolved.com/r/mkbhd"
}

POST /v1/embed/badge

Generates a badge configuration for programmatic embedding of Resolution Scores into your own UI.

{
  "brand": "samsung",
  "embed_url": "https://reallysolved.com/api/embed/brand/samsung.svg?style=flat&theme=dark",
  "resolution_score": 73,
  "expires_at": "2026-05-24T15:22:00Z"
}

Badge embed endpoints (hosted on reallysolved.com)

HTML embed example

<img src="https://reallysolved.com/api/embed/resolver/mkbhd.svg"
     alt="Expert Solver Badge: mkbhd · Senior Solver">

Model Eval Data — your model's own accuracy, verified

A continuously-updated, human-verified QA feed built on factual ground truth — not preference votes or "vibes." Every time your model's answer is flagged as an outlier on a real user query, our human Expert Solvers verify what's actually correct, leaving a running record of your model's real-world accuracy on the facts.

Public preference-leaderboards rank models on which answer feels better, compared openly against every other lab. This is different by design: a private, single-provider feed about your own model only — a live integrity signal for teams that ship in domains where a confident hallucination is a liability, not a vibe.

If your team evaluates or QA's a model — vertical SLM or frontier — this is real-world signal on where it's actually right or wrong: generated organically by real usage, pre-filtered to genuine disagreement cases, and verified by domain experts — at a fraction of what human-labeling vendors charge. Every verified record is attributed to the specific Expert Solver who checked it (handle + a domain-matched confidence score — never a legal name) and carries its supporting citation, so you can weigh it yourself rather than take our word for it. In your stack it shows up as two tiers:

• Outlier flags — your model's answer diverged from the group on a real query. No verdict on who was right — a triage signal for your own review, generated without ever exposing what any other provider's model answered.
• Verified corrections — your model's answer was checked against a cited, human-verified ground truth and found incorrect, with the correction, resolver track record, and source attached.

Every record is scoped to your model alone. We do not sell or disclose comparative rankings, another provider's outputs, or cross-model benchmark standings — to you or to anyone else.

GET /v1/diff-feed Beta

A chronological stream of events about your model only: a prompt, your model's answer, whether it was an outlier, and — when available — the human-verified resolution with its resolver and citation. Content is anonymized; any named-person facts appear only as short attributed quotes. Available under a direct licensing agreement scoped to your own model — we do not include or disclose any other provider's answers.

{
  "events": [
    {
      "diff_id": "diff_9f2c7a",
      "prompt": "Is creatine monohydrate safe for adolescent athletes?",
      "vertical": "health",
      "your_model_answer": { "stance": "insufficient evidence" },
      "outlier": true,
      "resolution": {
        "verdict": "Safe at standard doses for most adolescents; advise physician consult.",
        "resolver": "@handle_xyz",
        "resolver_confidence": { "score": 940, "domain": "health" },
        "citations": 2
      },
      "resolved_at": "2026-05-30T17:11:00Z"
    }
  ],
  "total": 1
}

Vertical ground-truth corpora Beta

License a domain-specific corpus of human-verified resolutions — the fuel for fine-tuning or evaluating a specialized model in a field where accuracy and provenance matter (health, law, finance). Specialized models live or die on trustworthiness; this is how you train one to be reliable & prove it — low hallucination, verifiable. Every record carries its verification trail and citations.

2 ways to work with us:

• License the corpus — a de-identified, continuously-updated, provenance-tracked dataset for your vertical.
• Co-build the model — we supply and keep refreshing the ground-truth; you bring the training. Talk to us about exclusivity.

Webhooks

Subscribe to real-time events by registering a webhook URL in your account settings at reallysolved.com/enterprise. Events are sent as HTTP POST requests with a JSON body.

{
  "event": "silence_clock.fired",
  "timestamp": "2026-05-24T16:00:00Z",
  "data": {
    "brand": "binance",
    "grace_period_started": "2026-05-17T16:00:00Z",
    "open_truths": 5,
    "resolution_score": 44,
    "action_required": true
  }
}

{
  "event": "score.changed",
  "timestamp": "2026-05-24T13:05:00Z",
  "data": {
    "brand": "binance",
    "score_before": 58,
    "score_after": 51,
    "delta": -7,
    "trigger": "truth_resolved",
    "truth_id": "truth_4mzq9w"
  }
}

Rate limits

Rate limits are enforced per API key. Exceeding the limit returns HTTP 429 Too Many Requests with a Retry-After header.

For the full rate card and custom volume pricing, see reallysolved.com/enterprise.

Content scope

API responses include brands, products, services, and AI system claims. Expert Solver votes and Truth filings cover commercially and technically verifiable claims.

Changelog

Status

Status page coming soon.

To subscribe to status updates and incident notifications, email support@reallysolved.com.