Oracles

Analytic oracles: exact functions with synthetic cost models.

Useful for unit tests and for controlled driver-benchmarking experiments where the ground truth and the cost model are both known exactly.

class cadaques.oracles.analytic.AnalyticOracle(fn, price_fn=<function AnalyticOracle.<lambda>>, name='analytic')

Wrap fn(params) -> float as a metered Oracle.

price_fn maps a query to its declared cost (default: a flat one-second tariff), letting tests exercise heterogeneous-cost scenarios deterministically. The settled cost equals the declared cost plus the (negligible) measured evaluation wall time, keeping the declared/settled distinction alive even in toy settings.

Parameters:

• fn (Callable[[Mapping[str, float]], float])

• price_fn (Callable[[Query], Cost])

• name (str)

cadaques.oracles.analytic.quadratic_bowl(center)

Negative squared distance to center (maximum 0 at the center).

Parameters:

center (Mapping[str, float])

Return type:

Callable[[Mapping[str, float]], float]

Canonical illustrative oracle: the two-dimensional Ising model.

The campaign it supports is a textbook discovery task with an exact analytic answer: locate the critical temperature of the 2D Ising model by maximizing the magnetic susceptibility. Onsager (1944):

T_c = 2 / ln(1 + sqrt(2)) ≈ 2.269185 (J = k_B = 1)

The oracle is a checkerboard-Metropolis Monte Carlo simulation whose fidelity knobs — lattice size L and number of measurement sweeps — carry genuine, heterogeneous cost: accuracy is bought with compute. This makes it a faithful miniature of real discovery oracles, where every answer has a price and better answers cost more.

cadaques.oracles.ising.T_C_EXACT: float = np.float64(2.269185314213022)

Onsager’s exact critical temperature (J = k_B = 1).

class cadaques.oracles.ising.Ising2DOracle(default_L=24, default_sweeps=400, default_equilibration=200, seconds_per_spin_sweep=2.5e-08, seed=None)

Metered Monte Carlo oracle for the 2D Ising model.

Parameters:

• default_L (int) – Fidelity defaults used when a query does not specify them.

• default_sweeps (int) – Fidelity defaults used when a query does not specify them.

• default_equilibration (int) – Fidelity defaults used when a query does not specify them.

• seconds_per_spin_sweep (float) – Ex-ante cost model: declared price is proportional to L**2 * (equilibration + sweeps). The settled cost inside each Result is the measured wall time, so declared/settled discrepancies are visible in the ledger by construction.

• seed (int | None) – Base seed; each evaluation derives an independent stream.