What Flows?

TL;DR

Systems are pipes and reservoirs — find what moves, what waits, and what leaks.
Bottlenecks dominate throughput — optimize the constraint, not the average step.
Flows can be matter, energy, information, or attention — often all at once.

How to use this lens

Draw sources and sinks — where does input enter and output leave?
Trace the main carrier — packets, blood, money, messages, decisions.
Mark queues and buffers — anywhere things wait is a design lever.
Find the slowest step — that is your bottleneck (Theory of Constraints).
Check for leaks and rework — flow that circulates without exiting wastes capacity.
Quantify if possible — rate in, rate out, average wait.

Core concepts

Takeaway: Conservation laws constrain design.

What enters must exit, transform, or accumulate. Unaccounted flow becomes debt — technical, metabolic, or organizational. Steady state means inflow equals outflow; growth means reservoirs fill.

Takeaway: Queues appear whenever arrival exceeds service.

Randomness makes queues form even when average capacity looks sufficient. Queueing theory turns this intuition into design: how many workers, how much buffer, what tail latency to expect.

Takeaway: Information flow is often the real product.

In an office, electricity flows trivially compared to attention and decisions. In a datacenter, bits move — but coordination messages determine whether those bits mean anything.

Flow checklist

Question	Why it matters
What is the unit of flow?	Enables measurement
Where do items wait?	Latency lives in queues
What is the service rate?	Capacity planning
Is flow bursty or smooth?	Burstiness inflates queues
Can flow be dropped?	Lossy vs lossless design

Mini-examples

Sleep: metabolic waste clears from brain interstitial space; glucose and hormones cycle; neural activity patterns flow between wake and REM.
Office: email and Slack carry requests; calendars buffer attention; cash flow exits to payroll and vendors.
Blood: oxygen rides erythrocytes; capillaries are the last-mile network; the heart is the scheduler.

Curriculum bridge

Formalize flow with:

Queueing Theory — waits, utilization, tail latency
Streaming Algorithms — one-pass computation on flows too large to store

Try it now

Trace one request you made today (coffee order, bug fix, text reply). List every wait and handoff. Where was the bottleneck?

Going deeper

Takeaway: Parallel paths need synchronization.

Blood splits into parallel capillary beds; microservices fan out to shards. Joining partial results reintroduces a sync point — often the new bottleneck.

Takeaway: Backpressure is conversation.

When downstream is full, upstream must slow or drop. Protocols that ignore backpressure buffer until they explode. Healthy systems signal congestion early.

Takeaway: Attention is zero-sum flow.

In cognitive systems, focus allocated here is unavailable there. Multitasking is rapid switching with switching cost — not true parallelism.

Seeing flow turns vague “slowness” into a located queue you can redesign.