Skip to content

Verification API · pre-launch

Your AI makes financial claims. Artizar makes them provable.

A verification API for teams shipping AI over financial data. Send a model-generated claim; get back a deterministic verdict, the evidence behind it, and a signed audit record you can keep.

Request early access
POST /v1/verificationsillustrative
claim
“Apple’s fiscal 2024 10-K reports total net sales of $391.0 billion.”
origin
ai_generated

200 · signed verification record

verdict
verified
rule_fired
R-11 · primary-source match
evidence[0]
10-K · AAPL · FY2024 · EDGAR 0000320193-24-000123“Total net sales… $391,035” (in millions)
signature
ed255199f2c44a1c7e30d5b88e1b7
record_id
rec_01J9ZC3AB8QY
logged
2026-07-01T14:32:07Z · append-only

Illustrative response. Figures are from Apple’s FY2024 Form 10-K as filed on EDGAR.

  • Deterministic rule table
  • EDGAR-traced evidence
  • Ed25519-signed records

The problem

Confidence is not evidence.

Language models answer financial questions fluently, and sometimes wrongly. When an agent quotes revenue that was never filed, or summarizes a risk factor that isn’t in the 10-K, the team that shipped it can’t point to a source, a rule, or a record.

The question that decides whether you can ship isn’t whether AI can be wrong. It’s whether you can show why an answer deserved trust.

How it works

One API call in. Three artifacts back.

  1. (01) EVIDENCE

    The claim meets the filing

    Artizar pulls the primary record behind the claim (the filing itself, from EDGAR or PACER) and extracts structured evidence that keeps its provenance at every step.

  2. (02) RULE TABLE

    Rules decide, not the model

    A fixed, versioned rule table maps that evidence to a verdict. The model gathers evidence; it never votes on the outcome. Same evidence in, same verdict out.

  3. (03) SIGNED RECORD

    The check becomes a record

    The whole check (claim, evidence, rule fired, verdict) is written to an append-only log and cryptographically signed. Anyone holding the record can replay it.

Deterministic verdicts

Run it twice. Same answer, byte for byte.

Verdicts come from rules over structured evidence, not from sampling a model. Verify the same claim tomorrow and the verdict, the rule that fired, and the output digest are identical. No LLM votes on truth anywhere in the path.

RUN 1 · 2026-07-01T14:32:07Z

claim_sha256
c41e88f2a3d97b60
verdict
verified
rule_fired
R-11 · primary-source match
output_sha256
7d194c3eb6e02a58

RUN 2 · 2026-07-02T09:05:41Z

claim_sha256
c41e88f2a3d97b60
verdict
verified
rule_fired
R-11 · primary-source match
output_sha256
7d194c3eb6e02a58

✓ digest match7d194c3eb6e02a58 ≡ 7d194c3eb6e02a58

What makes it defensible

An audit log you can hand to anyone

Every verification is appended to an immutable, cryptographically signed log. Anyone holding the record can replay the check and confirm nothing was edited after the fact: a defensible trail from claim to verdict, not a black box.

  1. CLAIM

    “Net sales of $391.0 billion in fiscal 2024.”

  2. EVIDENCE

    Form 10-K, AAPL, FY2024 · “Total net sales… $391,035” (in millions) · EDGAR 0000320193-24-000123

  3. RULE

    R-11 · primary-source match · rule_table v3

  4. VERDICT

    verified

  5. SIGNATURE

    ed25519 9f2c44a1c7e30d5b88e1b7

Evidence that ends at the filing, not at a website

Verdicts trace to primary sources through an unbroken provenance chain. Not a news story about the 10-K. The 10-K, as filed. When the evidence can’t reach a primary source, the verdict says so instead of pretending.

Tier-1 sourcesEDGAR (SEC filings) · PACER (federal court records)

Who it's for

Built for teams shipping AI over financial data

  • AI agents & copilots

    Verify before the user sees it

    Your agent quotes a revenue figure or cites a filing. Artizar checks the claim against the source and attaches the verdict and evidence before the answer leaves your pipeline.

  • Fintech apps embedding AI

    A verification layer between the model and your customers

    Ship AI features over financial data with a check in the path, and a signed record for every claim you surfaced, ready when someone asks how you knew.

  • Research & investment platforms

    Publish work your readers can check

    AI-assisted research where every figure carries its evidence chain back to the filing. Your analysts can verify the model's work; so can your readers.

Principles

Verifiability, not infallibility

We don’t claim AI is always right; nobody honestly can. We make every claim checkable and every check auditable.

The verification engine is built to ship as open source under Apache 2.0. A system that asks for your trust should let you read how it decides.

Don’t ask users to trust your AI. Give them the record.

Artizar is pre-launch and onboarding a small set of early partners.

Request early access