AIRVS — The Recommendation Provenance Standard  ·  v1.0.0  ·  DOI: 10.5281/zenodo.20391984

Recommendation Provenance

AI made an investment call. We score it, in public.

AIRVS — The Recommendation Provenance Standard.

Our POV

A world where any AI-generated investment opinion can be checked in 5 seconds — by anyone, with provenance.

Read our 5-year vision →

The Standard

What does AIRVS look at?

Four independent dimensions, never summed into one number: six process axes (Pass/Fail, evidence-gated), macro/micro coherence (3 tiers), outcome tracked at D+30/60/90/180, and a single verdict label.

1

Data Source

Sources real, primary, and tier-classified — each claim cites where it came from.

2

Reasoning Logic

The argument holds from premise to conclusion.

3

Counter Scenario

≥2 downside cases, each with a primary source and weighted probabilities.

4

Timing

Explicit entry window and a standard horizon — not an open-ended bet.

5

Accuracy / Hallucination

No non-existent tickers, fabricated figures, or invented facts.

6

Causal Chain

The cited sources actually support the stated conclusion.

TrustworthyAcceptableQuestionableUnreliableHallucinated
Read the full standard →

Live

This week's verification

Preview — sample data. Live verifications begin once real records are published.

Differentiation

Why AIRVS, not vendor self-checks?

01

Open by design

Methodology and data released under CC BY 4.0. Anyone can audit, reuse, and challenge it.

02

Cited by peers

Versioned with a DOI, archived on Zenodo, authored under ORCID — built to be referenced.

03

Multi-provider

Every AI is graded on the same bar. No vendor scores its own homework.

04

Public methodology

No black box. The rubric and each evaluator's decision rule are published before any verification.

“Standards belong to the standard, not the vendor.”

The problem

Who's accountable?

An AI told you to buy. When it's wrong, no one signed their name to it.

Re-runs don't match.

Ask the same model twice and the recommendation can change. We log the distribution.

Self-graded trust.

“Trust score” sites are graded by their own publishers. The information value is near zero.

How it works

The verification lifecycle

1

Publish

An AI investment recommendation is captured with its full context.

2

Process score

Six axes, Pass/Fail, each gated on cited evidence.

3

Coherence + outcome

Macro/micro coherence, then return vs benchmark at D+30/60/90/180.

4

Verdict

A five-tier label from a pre-published, version-locked decision rule.

5

Reply & disclosure

7-day right of reply; conflicts of interest disclosed.