Independent benchmark · No rates · No leads

Do AI assistants get annuities right? We check.

AnnuityRates.AI is preparing a recurring, reproducible benchmark of answers about annuity taxes, safety, surrender, and income mechanics. The same fixed prompts will be graded against disclosed sources under a published rubric.

Benchmark run #1 has not started. API spend requires Galen’s explicit approval and is outside this foundation build.

Quarterly benchmark

The scoreboard

Published results will show computed scores only after two independent grading passes and logged adjudication.

Awaiting benchmark run #1

No assistant answers have been collected, graded, or scored. Results will appear only after an approved run completes the published review protocol.

Awaiting run #1. No sample assistants, fabricated scores, or implied findings are shown.

Error library

A taxonomy for evaluating answer failures

These are definitions for future grading, not claims about any assistant and not examples from a completed run.

Methodology

Built to be reproduced, not believed

The protocol fixes inputs before results exist and publishes the evidence needed to inspect every grade.

  1. Fixed questions

    A versioned set is frozen before each run.

  2. Identical prompts

    Each assistant receives the same prompt in a fresh session.

  3. Disclosed sources

    Expected claims are checked against listed primary sources.

  4. Two grading passes

    Independent passes precede logged adjudication.

  5. Public transcripts

    Prompts, answers, grades, and notes ship with a report.