Where growth becomes instability

Amplification is easy to want and easy to build. A user gets value, shows evidence of it, someone else sees the evidence and tries the system. Each turn of that loop makes the next turn more likely. Growth of this kind does not feel like a risk while it is happening. It feels like the thing working.

But a loop does not discriminate. It carries whatever is travelling through it. The same mechanism that spreads a genuine result spreads a fabricated one; the same referral structure that finds real users finds farms. Nothing in the loop can tell the difference, because the loop is not a judgement — it is a multiplier.

So you add resistance

Identity checks. Proof review. Trust thresholds. Withdrawal checks. Usage limits. Reputation. Each of these is negative feedback: it makes the next turn of the loop less likely, and it makes the system survivable.

And each one also slows the thing you built the loop for. This is the part that does not resolve. You are not choosing between growth and control; you are choosing the relationship between them, and that relationship has to change as the system gets bigger.

The threshold

The question I keep arriving at is: how far can positive feedback amplify before the stabilising mechanism has to become stronger?

There is a point in most systems where behaviour changes in kind rather than degree. Below it, abuse is noise — irritating, absorbable. Above it, abuse becomes the dominant behaviour, because the returns to abusing the system have exceeded the returns to using it properly. That crossing is usually visible in hindsight and rarely announced in advance.

I do not think there is a general formula. I think there is a general question, and most systems ask it later than they should.

What Gamerz360 taught me about the location of friction

Requiring strong verification at registration reduces abuse. It also reduces legitimate participation, and it reduces it disproportionately among exactly the people the system was built to reach — the ones least likely to hold the documents being asked for.

Frictionless entry does the opposite in both directions.

Both statements are true, which means the argument about whether restriction is good or bad cannot be settled. It is the wrong argument. Restriction is not a quantity to be tuned up or down. It has a location.

Put it at the risky action — withdrawal, elevated privilege, anything irreversible — and the front door stays open while the expensive edge stays controlled. Trust then has somewhere to accumulate toward, which turns verification from a toll into a reward.

I am fairly confident about this for systems where value leaves at an identifiable moment. I am much less sure it generalises to systems where the damage is diffuse.