(Analysis by Nick Wu)

To solve the subtask where $N=2$, we can brute force all possible programs that Elsie could have written and see if any of those programs could have given the observed output.

#include <bits/stdc++.h>
 
using namespace std;
 
int prog(string s, int idx1, int v1, int v2, int r1, int r2, int r3) {
	if (s[idx1] == '0' + v1) return r1;
	else if (s[!idx1] == '0' + v2) return r2;
	else return r3;
}
 
void solve() {
	int n, m; cin >> n >> m;
	assert(n == 2);
 
	vector<string> inputs(m);
	vector<char> outputs(m);
 
	for (int i = 0; i < m; i++) {
		cin >> inputs[i] >> outputs[i];
	}
 
	for (int idx1 = 0; idx1 < 2; idx1++) {
		for (int v1 = 0; v1 < 2; v1++){
			for (int v2 = 0; v2 < 2; v2++) {
				for (int r1 = 0; r1 < 2; r1++){
					for (int r2 = 0; r2 < 2; r2++) {
						for (int r3 = 0; r3 < 2; r3++) {
							bool ok = true;
							for (int i = 0; i < m; i++) {
								if ('0' + prog(inputs[i], idx1, v1, v2, r1, r2, r3) != outputs[i]) {
									ok = false;
								}
							}
							if (ok) {
								cout << "OK" << endl;
								return;
							}
						}
					}
				}
			}
		}
	}
	
	cout << "LIE" << endl;
}
 
int main() {
	int t; cin >> t;
	while (t--) solve();
 
	return 0;
}

To solve the subtask where $M=2$, we can prove that the only time when Elsie must be lying is when the two inputs are equal and the two outputs are not equal. Obviously, in this situation Elsie must be lying. If the two inputs and outputs are equal, then this is equivalent to the case where $M=1$ and the answer when $M=1$ is always OK. Otherwise, pick some variable where the two inputs vary and return the correct result based on the value of that variable.

#include <bits/stdc++.h>
 
using namespace std;
 
void solve() {
	int n, m; cin >> n >> m;
	assert(m == 2);
 
	vector<string> inputs(m);
	vector<char> outputs(m);
 
	// Read in input
	for (int i = 0; i < m; i++) {
		cin >> inputs[i] >> outputs[i];
	}
 
	if (outputs[0] != outputs[1] && inputs[0] == inputs[1]) {
		cout << "LIE" << endl;
	} else {
		cout << "OK" << endl;
	}
}
 
int main() {
	int t; cin >> t;
	while (t--) solve();
 
	return 0;
}

To fully solve the problem, let us assume that Elsie is not lying, and consider the first if statement that Elsie has in her program. If we take that variable $v$ and desired value $x$ and look at all input/output pairs where $v$ takes on value $x$, those outputs must all be equal to the return value from the first if statement.

Therefore, let us construct Elsie's program from top to bottom as follows: identify some variable $v$ and value $x$ where all given inputs that match share the same output value. Delete all such input/output pairs. Repeat this process until we have at most one input/output pair, then the answer is OK. If we cannot find such a variable/value pair at any point, then the answer is LIE.

If this procedure outputs OK, then we have constructed a valid program and we have correctly printed out OK. Therefore, it remains to prove that if Elsie was lying, this procedure will properly detect it. The concern here is that perhaps the order in which we write the if statements matters, and if we can order them differently, we might be able to get a valid program.

The reason that this concern does not matter is that, the moment we can add an if statement of the form that variable $v$ takes on value $x$, we can always write this if statement in the future and it will not exclude additional inputs. Therefore, there is no benefit to rearranging if statements.

Nathan Wang's C++ code:

#include <bits/stdc++.h>
 
using namespace std;
 
void solve() {
	int n, m; cin >> n >> m;
 
	vector<string> inputs(m);
	vector<char> outputs(m);
 
	// if tcPassed[i] is true, then the program we have generated
	// so far gives the right answer for the i'th test case.
	// if tcPassed is true for every test case we're given,
	// then we know that Elsie isn't lying.
	vector<bool> tcPassed(m, false);
 
	// Read in input
	for (int i = 0; i < m; i++) {
		cin >> inputs[i] >> outputs[i];
	}
 
	// Repeatedly add one more "if" statement to our program
	// until every test case gives the right answer
	while (true) {
		bool foundIfStatement = false;
 
		for (int bit = 0; bit < n; bit++) {
			if (foundIfStatement) break;
			for (int val = 0; val <= 1; val++) {
				if (foundIfStatement) break;
				for (int output = 0; output <= 1; output++) {
					if (foundIfStatement) break;
					// try the following if statement:
					// if (input[bit] == val) return output;
 
					bool consistent = true;
					bool atLeastOneInput = false;
					for (int tc = 0; tc < m; tc++) {
						// ignore test cases that are covered by
						// previous if statements
						if (tcPassed[tc]) continue;
 
						// check if inputs[tc] satisfies the if statement
						if (inputs[tc][bit] == '0' + val) {
							atLeastOneInput = true;
							// if the expected output for inputs[tc]
							// is not the same as `output`, then
							// the if statement we generated won't work
							if (outputs[tc] != '0' + output) {
								consistent = false;
							}
						}
					}
 
					// if this if statement works, mark all the inputs
					// that the if statement covers as passed
					if (consistent && atLeastOneInput) {
						foundIfStatement = true;
						for (int tc = 0; tc < m; tc++) {
							if (tcPassed[tc]) continue;
							if (inputs[tc][bit] == '0' + val) {
								tcPassed[tc] = true;
							}
						}
					}
				}
			}
		}
 
		if (!foundIfStatement) break;
	}
 
	// Check if every test case is passed
	bool ok = true;
	for (int i = 0; i < m; i++) {
		if (tcPassed[i] == false) {
			ok = false;
		}
	}
	if (ok) {
		cout << "OK" << endl;
	} else {
		cout << "LIE" << endl;
	}
}
 
int main() {
	int t; cin >> t;
	while (t--) solve();
 
	return 0;
}

Danny Mittal's Java code:

import java.util.ArrayList;
import java.util.HashSet;
import java.util.List;
import java.util.Scanner;
 
public class ReverseEngineering2 {
 
    public static void main(String[] args) {
        Scanner in = new Scanner(System.in);
        for (int t = in.nextInt(); t > 0; t--) {
            int n = in.nextInt();
            int m = in.nextInt();
            in.nextLine();
            List<String> instances = new ArrayList<>();
            for (; m > 0; m--) {
                instances.add(in.nextLine());
            }
            while (true) {
                int oldSize = instances.size();
                for (int k = 0; k < n; k++) {
                    for (char bit = '0'; bit <= '1'; bit++) {
                        int pos = 0;
                        for (String instance : instances) {
                            if (instance.charAt(k) == bit) {
                                pos |= 1 << (instance.charAt(n + 1) - '0');
                            }
                        }
                        if (pos != 3) {
                            int finalK = k;
                            char finalBit = bit;
                            instances.removeIf(instance -> instance.charAt(finalK) == finalBit);
                        }
                    }
                }
                int newSize = instances.size();
                if (newSize == oldSize) {
                    break;
                }
            }
            if (new HashSet<>(instances).isEmpty()) {
                System.out.println("OK");
            } else {
                System.out.println("LIE");
            }
        }
    }
}

My Python code:

def solve():
    nvars, ninputs = (int(x) for x in input().split())
    inputs = []
    results = []
    for _ in range(ninputs):
        data = input().split()
        inputs.append(data[0])
        results.append(data[1])
    programs = [i for i in range(ninputs)]
    while True:
        updated = False
        for i in range(nvars):
            if updated:
                break
            offvals = set()
            offinputs = []
            onvals = set()
            oninputs = []
            for j in programs:
                if inputs[j][i] == '1':
                    onvals.add(results[j])
                    oninputs.append(j)
                else:
                    offvals.add(results[j])
                    offinputs.append(j)
            if len(offvals) <= 1 and len(onvals) <= 1:
                print("OK")
                return
            if len(offvals) == 0 or len(onvals) == 0:
                continue
            if len(offvals) == 2 and len(onvals) == 2:
                continue
            if len(offvals) == 1:
                updated = True
                programs = oninputs
            elif len(onvals) == 1:
                updated = True
                programs = offinputs
            else:
                assert False
        if not updated:
            print("LIE")
            return
 
t = int(input())
for _ in range(t):
    input()
    solve()

(Lecture notes on which this problem was based)