10 — GHZ Collapse¶
Category: Quantum (signed-GHZ stabilizer readout) | Difficulty: Hard
Flag: HTB{p4u1i_st4b1l1z3rs_unl0ck_th3_ghz_5ecr3t_291e599502e752180980a5584dae6fa8}
How I found the flag¶
Scenario: Halcyon Dynamics' STELLAR-4 relay shares a 5-qubit GHZ state across five satellite nodes; the authentication descriptor is baked into how the qubits are jointly entangled. The tamper cage never exposes the secret as a classical bitstring — the only way in is to measure the qubits in the right Pauli bases and rebuild the descriptor from the joint correlations. Recover the per-session secret across all 8 GHZ preparations, then /submit.
Technique: Each preparation is a signed GHZ stabilizer state |ψ⟩ = (|v⟩ + (-1)ᵖ|v̄⟩)/√2. Secret = 5 bits/state = 4 (s_vec) + 1 (p) → 8×5 = 40 bits = 10 hex.
- MEASURE ZZZZZ → outcome is
vorv̄(random global flip). The pairwise-XOR correlations are flip-invariant → recoverv(qubit 0 is the GHZ control,v₀=0) ⇒s_vec = v[1:5]. - MEASURE XXXXX →
X^⊗5eigenvalue =(-1)ᵖ⇒ parity of the 5 bits = p. YYYYY(odd # of Y) is a random decoy; only even-Y Pauli strings are stabilizers (YYXXXetc. are deterministic) — the "which bases give deterministic outcomes" puzzle.
Server traps (reversed from server.py):
- Per-shot measurement noise (that's why each state gets 6 rounds × up to 5 shots) → majority-vote;
vvia ML fit over all 10 pairwise correlations (error-correcting),pvia 20-shot parity majority. - One
/submitper session (first is free, thenRATE_LIMITED) ⇒ every one of the 40 bits must be perfect in a single shot. - Convention gotcha:
s_vec = v[1:5](GHZ control + X-flips on qubits 1–4), not the textbook adjacent-generator XORvⱼ⊕vⱼ₊₁— the adjacent reading was the first guess and failed.
Winning line: per state spend 2 rounds ZZZZZ + 4 rounds XXXXX (≤6 budget), ML-reconstruct v, majority-vote p, pack with the server's own _pack_secrets, single SUBMIT. Re-verified on a fresh session.