Credit / lending pack: fixed-rate loan pools with CPR prepayments, CDR defaults, loss severity and a recovery lag. Benchmarked in benchmarks/credit/ against independent month-by-month reference implementations.

Contract types

credit.pool_level_pay

Homogeneous level-pay (fully amortizing) pool. The engine's expression dialect has no loops, so streams use the exact closed form for a homogeneous pool under constant SMM/MDR (the standard pool-factor decomposition) — see the convention block at the top of lowering/rules.toml.

Terms:

Streams (all suffixed by contract instance): credit.pool.interest, credit.pool.sched_principal, credit.pool.prepay, credit.pool.recoveries, credit.pool.servicing (outflow), credit.pool.penalty.

servicing_fee and prepay_penalty_rate are available on every pool type. The penalty is a flat rate on prepaid balance (simplified yield maintenance); a discounted make-whole needs an engine primitive and stays on the parity worklist.

The contract term must span term_months + recovery_lag_months schedule periods so the recovery tail has periods to land in; the expressions gate themselves, so a longer term is harmless.

credit.pool_io_bullet

Interest-only pool with a principal bullet at maturity. Balance declines only through prepayment and default; the final period pays no SMM prepayment — the whole surviving balance pays as the bullet. Same terms as pool_level_pay (rate may be anything, including 0). Adds stream credit.pool.bullet.

credit.pool_float_io_bullet

Floating-rate IO/bullet pool. The coupon indexes off a model-declared curve statement:

curve sofr {
  2026-01: 0.048
  2026-07: 0.045
}

coupon = clamp(curve_value(index_curve, date) + margin, rate_floor, rate_cap). Balance dynamics are identical to pool_io_bullet (rate-independent), so the closed form stays exact. Extra terms on top of the common credit terms:

Floating level-pay pools are not supported: the balance path depends on the rate path, which has no closed form under the pack's loop-free expressions.

credit.purchase

Acquisition price paid at term_start (price term), stream credit.purchase.price. Discount/premium purchases are just a price different from face (e.g. 99.0 = 0.99 * balance); the level_pay_pool benchmark purchases at a 1-point discount.

Conventions

• Defaults leave the pool at the start of the period and earn no interest that period.
• cpr/cdr are annualized; converted with cpr_to_smm(x) = 1 - (1-x)^(1/12).
• Prepayment applies SMM to the performing balance net of scheduled principal.
• Recoveries return (1 - severity) of defaulted face, recovery_lag_months later. Defaulted-balance write-offs are not cash and emit no stream.

Metrics

domain.credit.interest, .principal (scheduled + prepay + bullet), .recoveries, .penalties, .servicing (omitted when zero), .collections (interest + principal + recoveries + penalties), .purchase, .collections_multiple (collections / purchase), and .wal_years — principal-weighted average life over principal returned (scheduled + prepay + bullet + recoveries; interest/penalties excluded).

Not in v0.1

• Floating-rate loans — needs the curve input concept and mean-reverting rate paths (stochastic roadmap item 4).
• Zero note rate on pool_level_pay (closed form divides by r).
• Delinquency states, servicer advances, loan-level heterogeneity.

Quick start

A $25mm level-pay pool with prepayments, defaults, a servicing strip, and a prepayment penalty:

version 0.1
model "my-pool"
use pack "credit" version "0.1.0"
time calendar monthly from 2026-01 for 126

entity fund buyer

contract credit.pool_level_pay.auto_a on entity fund.buyer {
  term 2026-01..2036-06
  terms {
    balance = 25000000
    rate = 0.065
    term_months = 120
    cpr = 0.08
    cdr = 0.02
    severity = 0.35
    recovery_lag_months = 6
    servicing_fee = 0.005
    prepay_penalty_rate = 0.01
  }
}

contract credit.purchase.auto_a on entity fund.buyer {
  term 2026-01..2026-01
  terms { price = 24750000 }
}

credit.purchase takes the dollar purchase amount — a 1-point discount (99.0) on the $25mm balance here.

Let the contract term span term_months + recovery_lag_months so lagged recoveries have periods to land in.

Run it

cfdl compile my-pool --packs packs --out my-pool/ir.json
cfdl run my-pool/ir.json --packs packs --pack credit --out my-pool/results.json --rate 0.08

Domain metrics include interest/principal/recoveries/collections, the collections multiple, servicing, penalties, and principal WAL (domain.credit.wal_years).

Recipes

Floating-rate IO bullet off a rate curve (margin/floor/cap against a model-declared curve):

curve sofr linear {
  2026-01: 0.050
  2027-01: 0.038
}

contract credit.pool_float_io_bullet.bridge on entity fund.buyer {
  term 2026-01..2028-12
  terms {
    balance = 15000000
    index_curve = "sofr"
    margin = 0.0275
    rate_floor = 0.065
    rate_cap = 0.095
    term_months = 36
    cpr = 0.05
    cdr = 0.01
    severity = 0.30
    recovery_lag_months = 3
  }
}

IO pool with bullet maturity: credit.pool_io_bullet — same loss vocabulary, interest-only until the balloon.

Full worked models: benchmarks/credit/level_pay_pool/, io_bullet_loan/, float_bridge_pool/, and the loan-pool notebook in examples/notebooks/.

Metrics reference

Computed automatically whenever a model runs with the credit pack, alongside the core metrics (NPV, IRR, MOIC, payback, WAL). Enumerated from the pack definition, so this list is always complete.

Validations reference

Checked at compile time. Each is a stable diagnostic code that is never renamed or reused; see diagnostics.

Worked example models

Benchmark cases are validated period-by-period against an independent reference implementation.

• Credit: floating-rate bridge pool
• Credit: IO/bullet bridge loan
• Credit: level-pay auto pool