physics-clock — Temporal Inference from Physics

No RTC. No NTP. No GPS. The physics IS the clock.

Infers elapsed time from physical observations using Bayesian temporal inference across multiple independent physics models. Each model provides an independent clock signal. The joint probability gives precise time.

How It Works

1. Observe physical quantities (temperature, sound speed, signal propagation, etc.)
2. Each physics model gives P(observation | t)
3. Bayesian fusion: P(t | all observations) ∝ Π P(obs_i | t) × P(t)
4. Peak of posterior = best time estimate

Physics Clock Models

Quick Start

from physics_clock import PhysicsClock, ThermalClock, SoundSpeedClock

clock = PhysicsClock([
    ThermalClock(),
    SoundSpeedClock(),
])

result = clock.infer_time({
    "temperature": 40.0,
    "initial_temp": 60.0,
    "ambient_temp": 25.0,
    "thermal_tau": 120.0,
    "measured_speed": 1495.0,
    "descent_rate": 0.5,
    "surface_temp": 20.0,
    "salinity": 35.0,
}, t_min=0, t_max=300)

print(f"Elapsed: {result.timestamp:.1f}s ± {result.uncertainty:.1f}s")
print(f"Precision: {result.precision}")
print(f"Clocks used: {result.n_clocks}")

Use Cases

• Underwater vehicles: infer time from sound speed + absorption + thermocline
• Robot fleets: infer time from silicon timing + thermal + constraint load
• IoT networks: infer time without NTP (physics is the sync protocol)
• Security: verify device honesty through reality parity (can't fake physics)

Security: Reality Parity

The timing of constraint evaluation IS attestation:

from physics_clock import RealityParity

parity = RealityParity()
result = parity.check({
    "eval_ns": 15000,        # reported eval time
    "n_constraints": 100,     # constraints evaluated
    "temperature": 30.0,      # reported die temperature
    "voltage_mv": 3300.0,     # supply voltage
})

if not result.honest:
    print(f"Device flagged: {result.reason}")
    print(f"Expected {result.expected_eval_ns:.0f}ns, got {result.actual_eval_ns:.0f}ns")
    print(f"Deviation: {result.deviation_sigma:.1f}σ")

• eval_time = f(complexity, temperature, voltage)
• If reported timing doesn't match reported temperature → device is lying
• No cryptographic keys needed. The physics can't be spoofed.

Composable With

• cocapn-schemas: temporal fingerprint tile format
• fleet-constraint-kernel: evaluation timing as temporal signal
• fleet-proto: Rust version of physics clock types
• temporal-auth: full authentication using physics-clock

Install

pip install physics-clock

License

Apache 2.0