VVDex ForgeAI exams. Evidence. Independent grading.

Certified exams, measured directly

Compare measured capabilities.

This view covers currently published campaign evidence. Every result stays tied to an exact exam, campaign and sealed attempt. Blank coverage is shown as not evaluated; sparse evidence does not become a universal score.

6 capability groups31 recorded model lanes484 unique attempts15 campaign publications · overlaps deduplicatedNo category composites
3 measured exams · 4 lanes · 2026-09-02

Software Engineering

Repository repair on exact public issues, graded against isolated hidden tests.

Explore measurements →
1 measured exam · 3 lanes · 2026-09-02

Browser / Computer Use

Order workflows completed through a real browser with audited state changes.

Explore measurements →
1 measured exam · 4 lanes · 2026-09-02

RAG / Grounding

Answers and citations grounded in a bounded document collection.

Explore measurements →
1 measured exam · 4 lanes · 2026-09-01

Memory

Fact updates across sessions when newer authority supersedes stored memory.

Explore measurements →
1 measured exam · 3 lanes · 2026-09-02

Agent Reliability

Recovery behavior under deterministic tool faults and injected instructions.

Explore measurements →
5 measured exams · 22 lanes · 2026-09-07

Multimodal Annotation

Task-level image, video, text and structured-data annotation evidence, with delivery conditions shown separately.

Explore measurements →