{"id":415,"date":"2026-06-01T14:01:51","date_gmt":"2026-06-01T14:01:51","guid":{"rendered":"https:\/\/sites.clarkson.edu\/ai-institute\/workbench\/tools\/bp-002\/"},"modified":"2026-06-27T02:25:18","modified_gmt":"2026-06-27T02:25:18","slug":"bp-002","status":"publish","type":"page","link":"https:\/\/sites.clarkson.edu\/ai-institute\/workbench\/tools\/bp-002\/","title":{"rendered":"BP-002: Think Zebras, Not Horses"},"content":{"rendered":"\n\u2190 Workbench<\/a>\n\n\n\n\n
\n
\n\n

Blueprint<\/p>

BP-002<\/p><\/div>\n

Think Zebras, Not Horses<\/p>

Presentation Bias<\/a> Stress Test<\/p><\/div>\n

Rung 2 · Diagnostic<\/p>

Close blueprint ▴<\/p><\/div>\n<\/summary>\n

\n
\n

Clarkson AI Institute — Rung 2 Blueprint<\/p>\n

BP-002: Think Zebras, Not Horses<\/h2>\n

Presentation Bias Stress Test<\/p>\n

Rung 2<\/span>·<\/span>Diagnostic<\/span>·<\/span>Output: Diagnostic Stable Card<\/span><\/div>\n<\/div>\n
\n
\n

🐎 Diagnostic Stable<\/p><\/div>\n

\n

🐴 HORSE<\/p>\n

Task is what it appears.<\/p>\n

p ≈ 0.85  ██████████<\/p>\n

🦓 ZEBRA<\/p>\n

Needs a different response.<\/p>\n

p ≈ 0.10  ██<\/p>\n

🞫 DONKEY<\/p>\n

Constraint blocks completion.<\/p>\n

p ≈ 0.03  ▌<\/p>\n

🐴🦓 MULE<\/p>\n

Mixed. Requires tradeoffs.<\/p>\n

p ≈ 0.015 ▏<\/p>\n

🦓🞫 ZEDONK<\/p>\n

Looks familiar. Structurally unstable. Fools the prior.<\/p>\n

p ≈ rare   ·<\/p>\n<\/div>\n<\/div>\n

\n
\n\n
\n

1<\/p>\n

Self-Diagnose<\/p>\n

Use the stable. What animal is your problem? Do this before asking anything.<\/p>\n<\/div>\n

→<\/p>\n\n

\n

2<\/p>\n

Signal<\/p>\n

Ask your question. Note what category your signals imply.<\/p>\n<\/div>\n

→<\/p>\n\n

\n

3<\/p>\n

Model Responds<\/p>\n

It answers what you asked. Preserve the response.<\/p>\n<\/div>\n

→<\/p>\n\n

\n

4<\/p>\n

Compare<\/p>\n

What did the model respond to vs. what did you actually have?<\/p>\n<\/div>\n

→<\/p>\n\n

\n

5<\/p>\n

Introduce Taxonomy<\/p>\n

Give the model the stable. Ask it to reclassify your original prompt<\/a>.<\/p>\n<\/div>\n

→<\/p>\n\n

\n

6<\/p>\n

Audit Your Signals<\/p>\n

Were your signals accurate? Did you know what you had before you asked?<\/p>\n<\/div>\n

→<\/p>\n\n

\n

7<\/p>\n

Fill the Card<\/p>\n

Document what you sent, what you had, and what differed.<\/p>\n<\/div>\n\n<\/div>\n

▲ User self-diagnoses first                                    Signal accuracy<\/a> under review →<\/p>\n

\n

🧑‍💻 Human Control Point<\/p>\n

Before asking the model anything, classify your own problem. If you cannot place it in the stable, that is the finding. The model can only work with the signals you send.<\/p>\n<\/div>\n<\/div>\n

\n

⚠ Failure Signals<\/p><\/div>\n