Short answer: most cellular unpredictability is born at interfaces and gates. The bulk cytosol is a big diffuser; the sharp, thin, high-gain places are where noise gets amplified and turned into bursts, thresholds, and regime switches.

If you want a rough variance budget without pretending there’s a universal number (there isn’t), think in timescales:

• Milliseconds to seconds: boundary layers and transport gates dominate the randomness.
• Minutes to hours: chromatin/transcriptional bursting and network nonlinearities take over.
• Hours to days: environment/exposome and cell-cycle state set the floor and kick the system between regimes.

Below is a pragmatic, “how much and why” map by interface. Treat the percentages as order-of-magnitude slices you calibrate per system, not commandments.

Where the unpredictability actually comes from

Plasma membrane ↔ interstitial fluid/blood/exposome

• Why noisy: ligand arrival is discrete (shot noise), receptors cluster and uncluster, channels gate stochastically, endocytosis happens in bursts. Microdomains near the membrane are reaction-limited, so tiny copy-number fluctuations become macroscopic decisions.
• Upshot: this interface can contribute ~15–35% of variance in fast signaling outputs (seconds to minutes) and seed longer-term variability by triggering cascades. In vivo, exposome swings can push that higher.

ER–mitochondria contacts and IMM/OMM (TIM/TOM, MCU/NCX, ANT, ATP synthase)

• Why noisy: Ca²⁺ “hotspots” at MAMs are nanoscale and intermittent; proton motive force forms nanodomains along cristae; protein import/export is a gated queue; a few pores or channels make the difference between plenty of ATP and a local drought. Rare events (e.g., brief mPTP openings) are low-probability, high-impact tail risks.
• Upshot: contributes ~10–25% to variability in energetic and Ca²⁺-coupled signaling on subsecond to minute scales; effects echo into metabolism and gene regulation.

Nuclear envelope ↔ nucleoplasm (NUPs, karyopherins; transcriptional bursting)

• Why noisy: nuclear pores are finite, cargoes compete, and promoter states flicker. Transport and bursty initiation create heavy-tailed mRNA distributions; small changes in residence time produce big fold-changes downstream.
• Upshot: ~10–20% of variance in protein levels over minutes to hours is often traceable to this gate plus promoter kinetics; for some genes it’s the main act.

Cytosol (bulk)

• Why mostly tame: diffusion is fast; many metabolites are buffered.
• Where it matters: when copy numbers are small (TFs, kinases at microdomains) or reactions are diffusion-limited near surfaces.