AIRVS v1.2.0 — §9 rebuttal simplified (MINOR)

MINOR amendment. §9-1 D-7 pre-publication notice removed; rebuttal restructured to post-publication rights (§9-R1–R4: guaranteed unedited rebuttal addendum, 7-day correction duty, public contact channel, anti-defamation principles unchanged). No change to axes, Pass model, verdict vocabulary, or decision-rule requirements — decision-rule v1.0.0 remains frozen and valid. Not retroactive: prior evaluations keep their version lock. Spec: v1.2/airvs-v1.2.md

AIRVS v1.1.0

AIRVS v1.1.0 (MINOR, backward-compatible, not retroactive)

• Continuous-strategy outcome mode (outcome_mode: continuous) for AI-managed funds/indexes.
• Annex A-F: AI-managed-fund variant of Annex A (prompt-based items N/A; adds holdings/methodology/cost transparency).
• Two type-specific result sheets: Template L (LLM single recommendation) and Template F (AI-managed fund).
• Clarified version-lock and re-evaluation-addendum policy. No peer review required (MINOR).
  v1.0.0 evaluations remain locked at v1.0.0.

AIRVS v1.0.0 — first public release

AI Recommendation Verification Standard (AIRVS) — v1.0.0

First stable public release of the AIRVS standard.

AIRVS is an open, version-controlled, peer-reviewable standard for evaluating AI-generated investment recommendations across four independent dimensions:

1. Six Process axes with mandatory evidence (Data Source, Reasoning Logic, Counter-Scenarios, Timing, Hallucination, Causal Chain).
2. Macro / Micro Coherence in three tiers (Sufficient / Partial / Missing).
3. Outcome time-series at four time points (D+30, D+60, D+90, D+180) with drawdown trajectory.
4. Five-tier Verdict label vocabulary (🟢 Trustworthy / 🔵 Acceptable / 🟡 Questionable / 🟠 Unreliable / 🔴 Hallucinated). The label vocabulary is standard; the decision rule that maps measurements to a label is implementer-defined and must be pre-published, version-locked, and disclosed per core.md §6-2.

Scope

• L1 (external recommendation review) only. Self-grading is structurally excluded.
• Applies to any asset class in any market. Framework rows for US, other developed markets, and emerging markets.

Documents in this release

• README.md — overview
• WHY.md — author motivation
• CHANGELOG.md — release log and known limitations
• CONTRIBUTING.md — how to submit external peer review
• LICENSE — CC BY 4.0
• v1.0.0/core.md — main standard text (§0–§15)
• v1.0.0/annex-a-ai.md — AI-specific assessment items (11)
• v1.0.0/tier-rulebook.md — source tier classification rulebook
• PEER-REVIEWS/internal-3-persona-review.md — author's adversarial pre-review

Reference implementation

A reference implementation of the §6-2 decision rule by the maintainer (MC AI Labs) is published in a separate sibling repository: mc-ai-labs-airvs-implementation. It is not part of the AIRVS standard.

Citation

Kim, Mincheol (2026). AI Recommendation Verification Standard (AIRVS), v1.0.0. MC AI Labs. CC BY 4.0.

Versioning policy

• PATCH (v1.0.1, …): wording, edge cases, indicator examples. No peer review required.
• MINOR (v1.1.0, …): backwards-compatible additions. No peer review required.
• MAJOR (v2.0.0, …): behavior-changing modifications. External peer review required per core.md §11-1.

Known limitations

See CHANGELOG.md "Known limitations (v1.0.x patch candidates)" for the catalogued patch backlog and PEER-REVIEWS/internal-3-persona-review.md for the author's own adversarial pre-review.

External peer reviews are actively invited. See CONTRIBUTING.md.