Bond Lock Model

Bond Model Progression

Stage 1: Empirical Baseline Bond Model

IRIS begins with an empirical baseline bond model built from historical walk-forward shortfall outcomes. The purpose of Stage 1 is to establish a bond floor that is grounded in realized loss behavior, easy to audit, and simple enough to enforce safely. In this stage, bond sizing is driven primarily by loan duration and a chosen risk target, such as keeping re-auction frequency at or below a specified threshold. Rather than assuming one closed-form theoretical curve, the model uses observed historical outcomes to determine what bond level would have been required for loans of different tenors.

For borrower requests that do not fall exactly on a research tenor knot, Stage 1 uses an exact-duration baseline curve rather than a small set of hard buckets. In practice, that means the model starts from empirical tenor points and interpolates between them to produce a bond fraction for arbitrary durations. The result is a smooth duration-based baseline that can support real borrower requests without forcing the protocol into only a few fixed maturities.

Stage 1 is intentionally a baseline model. It captures the strongest and most stable first-order driver of bond need, which is time under exposure. It does not yet attempt to fully price every live market feature or every quote-specific nuance. That is a deliberate choice: the first production model should be reliable, interpretable, and governable before it becomes highly dynamic.

The core Stage 1 formula is:

requiredBond = notional × baselineBondFrac(duration) × guardBand

Where:

• baselineBondFrac(duration) is the baseline bond fraction for that exact tenor

• guardBand is a fixed safety margin above the raw empirical curve

The baseline bond fraction is not limited to a few coarse buckets. It is produced from empirical tenor knots and then converted into an exact-duration curve. The recommended implementation is:

• use empirical bond fractions at known tenor points

• interpolate on the log of bond fraction

• exponentiate back to get the bond fraction for any requested duration

So operationally:

log(baselineBondFrac(D)) = linear interpolation across tenor knots baselineBondFrac(D) = exp(interpolated value)

A better later refinement of Stage 1 would be to replace log-linear interpolation with a smooth monotone tenor curve with a smooth function of duration that never decreases as duration increases.

Why Stage 1 starts with an empirical baseline
The key design choice in Stage 1 is to anchor bond sizing to realized historical shortfall distributions rather than to a purely parametric formula. This is especially important for DeFi lending markets, where rate behavior is path-dependent, utilization-kinked, venue-specific, and prone to clustered stress events. A historical baseline directly measures the thing IRIS is trying to protect against: cumulative shortfall over the life of a loan. That makes it a better starting point than imposing a smooth theoretical term structure and hoping it matches actual protocol behavior.

Stage 1 should be read as a duration-first baseline, not as a claim that duration is the only thing that matters. It is the layer that defines the minimum credible bond schedule from which richer context-aware adjustments can later be made.

Reference pricing profile
Stage 1 is calibrated against a reference launch pricing profile and risk target, rather than trying to solve every possible quote configuration at once. This is intentional. A baseline needs a stable operating reference point. That reference profile should be chosen to represent the center of the intended launch range, not an extreme borrower-friendly or solver-friendly quote. In that sense, the baseline is not “the universal true bond for all spreads”; it is the base schedule around which later stages add quote- and context-specific adjustments.

What Stage 1 gives the protocol
Stage 1 gives IRIS a bond schedule that is:

• empirically grounded

• duration-aware

• exact-duration compatible

• simple enough to govern

• simple enough to enforce onchain

• conservative enough to act as a credible floor

That is the right place to start. It replaces static one-size-fits-all intuition with a real bond policy while keeping the system operationally legible.

Stage 2: Quote-Aware Baseline

The second stage extends the baseline so that bond no longer depends only on duration. It adds explicit sensitivity to quote aggressiveness and product profile. In practice, this means the bond floor begins to distinguish between loans that are priced tightly and loans that are priced with more spread cushion. The goal is still to remain schedule-based and auditable, but with a better mapping between quoted economics and required protection. Stage 2 should still feel like a governed policy table, not a black-box prediction engine.

Stage 3: Context-Aware Risk Overlay

The third stage adds market-state awareness on top of the quote-aware baseline. This is where live inputs begin to matter more directly: volatility, utilization or kink proximity, venue stress, and cross-venue dispersion can all push required bond above the base schedule. The core idea is that the protocol keeps its stable baseline, then overlays a context multiplier when the current state is riskier than normal. At this stage, the bond system becomes responsive to conditions without losing its baseline discipline.

Stage 4: Versioned Offchain Risk Policy

The fourth stage formalizes the operational update process. Instead of treating the bond policy as static, IRIS begins publishing versioned risk-policy updates on a regular cadence. Baseline curves, quote-aware adjustments, and context overlays are recalibrated from fresh historical windows and validated against holdout periods. This makes the bond model a governed policy system rather than a one-time research artifact. The important principle at this stage is that updates remain deliberate and reviewable rather than changing continuously in opaque ways.

Stage 5: Signed Per-Quote Dynamic Bond Enforcement

The fifth stage introduces a richer per-quote bond requirement computed offchain but enforced onchain. At this point, the offchain model can use live quote-time state to produce a bond requirement specific to that RFQ, while the contract still retains an onchain floor schedule as a minimum bound. This gives IRIS the precision of a live risk model without giving up enforceability or safety. The bond model is no longer only a static schedule; it becomes a constrained dynamic system where richer context can require more bond than the baseline, but never less than the governed floor.