Each day, five prompts that all solve the same task — each at a different quality level. Reorder them worst→best. After you submit, each gap reveals the prompt engineering principle that separates the two rungs.
Rung 1 → 2: Clarity of task. A one-word prompt asks for zero context.
Rung 2 → 3: Role or convention. Tells the model the posture and output shape to adopt.
Rung 3 → 4: Constraints and scope. What's in-bounds, what's off-limits, what kind of output.
Rung 4 → 5: Examples, edge cases, and acceptance criteria. The model produces what you'd accept without follow-up.
Wordle-style daily prompt engineering puzzle. Guess the prompt that produced the output.
Eight questions that tell you which AI coding stack fits how you actually work.
18 documented cases of LLMs confidently hallucinating — with primary sources.
Stack prompt-token groups in a simulated context window. Learn tokenization by playing.
Each puzzle has a canonical order authored by the maker. Rungs escalate from "vague one-liner" to "role + constraints + examples + edge cases." The gaps are the teachable moment — every answer reveal includes the principle separating adjacent rungs.
Inversions from canonical order. A perfect sort is 0 inversions → 100 ladder points. Each swap away from canonical costs 10 points. Fully reversed is 10 inversions → 0 points. It's the same metric Kendall tau uses for rank correlation.
One puzzle per day, rolling over at 00:00 UTC. Use the Shuffle again button to re-randomize the starting order of today's rungs before you submit — but you only submit once per day. Shared URLs reveal other days' puzzles without counting toward your streak.
After submitting, the page shows a URL like ?day=5&order=dbcae where each letter is the canonical rung position (a=worst, e=best). So dbcae means "I placed canonical-rung-d in position 1, rung-b in position 2…" — your submitted order, bookmarkable and sharable. Visitors can replay the same puzzle.
In your browser's localStorage under promptLadderStreak. Nothing is uploaded or tracked. Clearing site data resets streak to 0.
Read our CLAUDE.md guide for the skeleton of a well-scoped prompt, grab the Claude Code Starter Kit for 20 power prompts, or try the CLAUDE.md Playground to see how scope inheritance works across levels.
Saved you time? Tip the maker in BTC — no account, no signup, just paste.
bc1qs04leape97ner4wqa98n94l9n0gv9aa84eg4ux