Formalize, then check.
Three puzzles, three candidate formalizations each, and a real solver. Pick a candidate and the lab enumerates every assignment, returns the models, and checks what follows. When the formalization is wrong, the proof is still valid.
The formalize-and-check lab
Preset 01 · one key, three signs
The three boxes
Preset loaded. Choose a candidate and press Check.
Which box contains the key?
Trace
Three steps, one verdict.
The formula, not the text
The parser reads the prefix DSL and prints the tree. A well-formed formula is the only thing the solver ever sees.
Every assignment
With n atoms there are 2n assignments. The highlighted rows are the models; the formula is satisfiable when at least one exists.
All models, or a countermodel
A query is entailed only if it holds in every model. One countermodel is enough to refute it; zero models make every query vacuously true.
It shows a genuine brute-force model enumeration over propositional formulas, an entailment check against the full model set, and the gap between a valid proof and the intended reading. Preset 1 is the deliberate trap: all three candidates parse, all three solve, and only one matches the text. The lab does not show first-order logic, SMT theories, or a production solver. The puzzles are small on purpose: every assignment is visible.
The solver is sound with respect to the formula. When a wrong candidate produces a different answer, the solver is still sound — the formalization is what failed. That is why the lecture calls fidelity the unverified part.