Stop Building Liquidity Layers That Avoid Repricing Risk

---

Liquidity layers that avoid repricing risk do not make markets safer. They postpone the moment risk is acknowledged and amplify the cost when it finally is.

This is not a moral claim about complexity. It is a mechanical claim about what happens when markets move out of the regime a design assumes.

Mechanism

AMMs already encode a decision: a pricing curve that converts order flow into inventory drift. The curve is the policy. It chooses what the pool will do when pressure arrives.

A “liquidity layer” is any add-on that changes the surface behavior of that curve without changing the underlying exposure. These layers typically optimize one of three things:

• Capital efficiency: make less liquidity look like more liquidity.
• Local stability: make small price deviations cheaper to arbitrage.
• Yield smoothing: bolt external cashflows onto swap fees.

The common move is substitution: explicit repricing is replaced with assumptions.

Concentrated liquidity assumes a price interval will hold often enough. Stable-swap assumes parity holds often enough. Boosted pools assume the lending leg remains liquid enough. IL protection assumes the backstop remains solvent enough.

The abstraction is not “adding safety.” The abstraction is making the risk look quieter until the moment it cannot.

A range-based liquidity position is the cleanest example because the assumption is explicit: the position only exists inside its band.

active_liquidity = (price >= lower && price <= upper)
fees_accrue      = active_liquidity
inventory_risk   = always_on

The layer improves behavior in the expected regime by narrowing where liquidity participates. It does not delete inventory risk. It changes when the risk becomes visible.

Repricing vs Suppression

Risk can be handled in two broad ways:

• Repricing risk

  • Losses become smaller because they happen earlier.
  • The system updates terms when conditions change.
  • The result is unpopular because it looks worse in dashboards.

• Suppressing risk

  • The system feels stable because the terms do not move.
  • The true cost is deferred.
  • The failure arrives discontinuously when assumptions break.

Suppression is not free. It is borrowing safety from normal conditions.

Failure Modes

Liquidity layers fail when the market leaves the regime they were designed to assume.

The failure modes are structurally consistent across designs:

• Depegs

  • “Stable” liquidity becomes one-way flow.
  • The pool’s inventory converges toward the failing asset.
  • Stability was purchased by assuming parity; the bill arrives at discontinuity.

• Range exits

  • Capital efficiency collapses into inactivity.
  • The position stops earning precisely when volatility is highest.
  • Rebalancing becomes mandatory rather than optional.

• Liquidity cliffs

  • Price impact changes abruptly rather than smoothly.
  • Routing becomes brittle: a small change in available liquidity causes a large change in execution quality.

• Protocol insolvency

  • Any design that relies on backstops, incentives, or external yield inherits solvency and liquidity risk from those systems.
  • When the backstop is questioned, the “safety layer” becomes a leverage point for runs.

None of these are edge cases. They are the places where risk actually matters.

Tradeoffs

Explicit repricing is not cosmetically attractive.

• It looks worse in dashboards.
• It reduces apparent yield.
• It surfaces losses sooner.

That is the cost of correctness.

The alternative is not “free yield.” The alternative is deferred accounting: a period of apparent stability followed by a regime where the system re-prices all at once.

Signal

If a liquidity design reads like a wrapper around every possible market condition, it is not a risk model. It is an escape hatch.

Signals that a protocol is avoiding repricing:

• If a system promises safety without repricing, it is borrowing credibility from assumptions.
• If risk disappears in normal conditions but reappears all at once, it was never removed.
• If yield is stable while volatility is not, something else is absorbing the variance.
• If “efficiency” increases while failure modes become sharper, the system is compressing exposure into tail events.

A layer that cannot state what it assumes is not a layer. It is a junk drawer of dependencies.