Controls
Presets
Parameters
Quality
Camera
Model checks
The science
A race of two clocks. A credential leaks at t = 0 (for example a key committed to a repository). Attackers who scrape for secrets find it after a random time T; with a constant discovery rate λ this is exponential, P(T < t) = 1 - e-λt. The defender must detect the leak and rotate the secret. A scanner that runs every s hours finds it after a wait U that is uniform on (0, s) (the leak lands at a random point between scans), triage takes c, rotation and redeployment take R: the exposure window is W = U + c + R.
Chance of exploitation. The attacker wins when T < W. Averaging over the scan phase: P = 1 - e-λ(c+R) · (1 - e-λs)/(λs). The expected time the attacker holds a working secret is E[(W - T)+] = E[W] - P/λ with E[W] = c + R + s/2. For a very low discovery rate this reduces to P ≈ λE[W]: every hour of window adds a fixed small risk.
What to improve. The chart of P against scan interval for three rotation speeds shows that shortening only the scan interval is useless when rotation takes days, and the reverse. When bots find public leaks within minutes (large λ) the only winning move is prevention (push-time blocking), because even a perfect response is slower than the discovery.
Simplifications. Constant discovery rate (real scraping is front-loaded in the first minutes); one attacker process; a deterministic triage and rotation time; no use of the secret before it is found; the scanner finds the leak with certainty at its next run (no misses). Defensive teaching only. Checks: Monte Carlo of the race and of the access time, two quadratures over the scan phase, and the rare-discovery limit.