MB
MBTI Typing Skill Benchmark Replay Lab
Blind replay, reference reveal, repair loop
Blind Replay

Replay benchmark traps before seeing the answer.

Select a canonical benchmark case, copy the blind prompt, record the leading type, serious runner-up, and falsifier guess, then reveal the reference and export a Replay Receipt.

1. Select Case

16 cases from benchmark-cases.json.

local-first
Benchmark replay cases are synthetic regression traps, not psychometric ground truth, clinical advice, hiring advice, or deterministic personality truth.

2. Enter Guess

Score the replay attempt before looking at the reference.

0 / 100

3. Reference

Reveal after the guess is recorded.

source-synced

Replay Receipt

Machine-readable replay-lab/v1 result.

Repair Prompt

Copy this back into the agent when replay misses a gate.

Issue Seed

Turn repeated replay misses into a benchmark_replay_improvement.yml contribution.

Open Issue
Copied