Programmable credit continuity.

Depositor / lender

Supplies liquidity to the vault. Wants yield with bounded, visible loss exposure.

Loan broker

Underwrites, sets terms, posts first-loss cover (CoverAvailable ≥ DebtTotal × CoverRateMinimum), can impair or default a loan, earns a management fee on interest.

Borrower

Receives principal, owes the schedule; carries the borrower score.

Collateral provider

Locks the asset — the borrower or an approved third party.

Supporter

Injects rescue capital after delinquency for a claim with a capped premium and new-money priority; earns Rescue Points that vest on performance.

Orchestrator

Publishes state, checks eligibility, enforces the deadline, settles atomically, writes the receipt. On this platform: the Credit Machine service; in the target architecture: a native protocol extension (see the roadmap).

OFFER ──fund──▶ ACTIVE ──pay──▶ ACTIVE ──repay_full──▶ CLOSED
                  │
                 miss
                  ▼
                GRACE ──catch_up──▶ ACTIVE
                  │
         grace_expired ─ admissible ─▶ RESCUE_OPEN ──rescue_filled (cure)──▶ CURED ──resume──▶ ACTIVE
                  │                        │  └────── close (make-whole) ──▶ REFINANCED ──resume──▶ ACTIVE
                  │                        └── rescue_expired ──▶ DEFAULT                └──close──▶ CLOSED
                  └──────── not admissible ──────────────────────▶ DEFAULT ──recover──▶ RECOVERY ──liquidate──▶ LIQUIDATED

1. 01Requestamount · term · purpose
2. 02Underwriteidentity · score · limits
3. 03Securecollateral · haircut · reserve
4. 04FundXLS-65 vault → XLS-66 loan
5. 05Monitorhealth · payments · early warning
6. 06Rescuegrace · auction · cure or refinance
7. 07Closerepay · release · score update

XLS-66 mapping: ACTIVE and GRACE are the ledger's loan with its PaymentDueDate and GracePeriod; impairment is LoanManage(tfLoanImpair) and shows as the vault's LossUnrealized; DEFAULT is LoanManage(tfLoanDefault), which draws min(DebtTotal × CoverRateMinimum × CoverRateLiquidation, DefaultAmount) from the cover. The Rescue Window, CURED and REFINANCED are the LiquidLabs layer in between.

Amortization is a level-payment schedule whose last row absorbs the rounding so the balance closes at exactly zero; pro-rata allocation uses the largest-remainder method so Σ accepted equals R_t to the micro-unit. Money is Numeric(24,6) everywhere; floats never touch a receipt.

1. Lender's test. EV_rescue ≥ EV_liquidate + SafetyMargin (0).
2. Coverage after rescue. (C_eff + FLC) / Q_t ≥ C_min (30 %).
3. Supporter hurdle. E[NPV_s] ≥ m_s (0) under the disclosed seniority (new money priority) — someone rational will fund it.
4. Affordability. p ≥ p_min (0.50): the broker's estimate that the borrower can carry the restructured plan. If not, liquidate rather than refinance repeatedly.
5. Not ENFORCE. When collateral fully covers the debt and stress is extreme, enforcement makes the lender whole cheaply — the map says enforce, and the window stays shut.

The thresholds are governance parameters (an AdmissionPolicy), approved like every other decision on this platform, never hard-coded in a template.

The 256-byte Loan.Data descriptor: version (1) · jurisdiction (2) · collateral escrow index (32) · SHA-256 of the signed terms (32) · accepted credential type (32) · originator tag (4) · reserved. Any verifier can tie loan, collateral and contract together without asking us.

1. Detect the missed payment; confirm the XLS-66 grace condition is reached.
2. Freeze a deterministic snapshot: B_t, I_t, F_t, Q_t, C_t, h, FLC, eligibility, deadline.
3. Compute R_t for the mode; run the five-part admission test.
4. Publish the offer with every economic term visible before the first commitment. Nothing may change afterwards — re-issue instead.
5. Accept reservations until the deadline: funds are locked, never released to anyone.
6. If Σ S_i ≥ R_t: allocate exactly R_t (pro rata, largest remainder, concentration cap), refund the excess, settle the vault or payoff account — never the borrower's wallet.
7. Mint the recovery claims with their seniority; update the borrower's repayment waterfall.
8. If Σ S_i < R_t at the deadline: release every reservation; DEFAULT → RECOVERY → the loss waterfall.
9. Write the SettlementReceipt with the transaction hashes and the state transition.

RescuePoints_i = Base × CapitalShare_i × Performance × Independence × Vesting

Independence is 1 only for capital whose beneficial owner is not the borrower or a related party; Vesting rises from 0 to 1 as the rescued loan performs (90 days or repayment). A score gain must represent real external value created.

The worked example is a 100,000 RLUSD loan with collateral. The volume case is the opposite: 50–5,000 RLUSD for a week to a few months, unsecured, decided in seconds. The request page prices it from three inputs and funds it into the same machine.

• Why a risk fee and not a higher rate. Microcredit losses are per loan, not per year: a 30-day loan at any lawful APR cannot earn a 5 % default rate back in interest. Hiding that in the rate would mean 60 % APR on a four-week loan; charging it as a named fee is the same money, disclosed.
• Why a ceiling. When covering the expected loss would cost more than 8 % of the amount, the product does not offer — a guarantor or a supporter who vouches lifts the score, a longer term spreads the cost. Lucía (score 52, 150 for 14 days) gets an offer at 6.7 %; at a score of 41 she does not.
• Why the same score. The tier that prices the loan and the recovery probability the admission test uses later come from one BorrowerScore, so a cheap loan is also one the Rescue Window will admit.
• Schedule. One bullet up to 30 days, weekly up to 90, monthly after — the XLS-66 PaymentInterval in seconds, the last installment closing at exactly zero.
• Rescue on a microcredit — what the machine says, honestly. Run Amara's 300 for 30 days to a missed bullet and let grace expire: the admission test refuses the window (the microcredit policy waives the coverage floor — there is no collateral — but the supporters' expected net is negative at the 15 % premium when a quarter of borrowers do not come back) and the loan goes default → recovery → waterfall: the broker's 5 % first-loss absorbs first, the vault the rest. That is the design: a microcredit is not rescued loan by loan, it is priced so that the portfolio absorbs its losses — the risk fee is exactly that expected loss, collected up front. A supporter-backed rescue becomes admissible only where the score is high enough for the capped 15 % premium to cover the gap (Amara's loan: a score of about 88, p ≈ 0.87), or where a guarantor stands behind the borrower — the social-collateral object the bootcamp roadmap lists.

1. The v4 state map liquidated every loan with coverage below 90 %. That is DeFi logic — enforce while collateral still covers the debt. XLS-66 loans are unsecured by design and LiquidLabs collateral is optional, so for a partially covered loan enforcing early guarantees the loss. v5 reserves ENFORCE for collateral that covers the debt under extreme stress and lets the expected-value test decide the rest. Under v4 the concept's own worked example (coverage 55 %) was not admissible.
2. Supporters lost money in expectation. A 1.08× claim with a 70 % chance is an expected loss of 24 %; no rational supporter funds that, so the market would never form. v5 gives rescue capital new-money priority on realised proceeds in a later default (the restructuring-law principle), which bounds the supporters' downside by the collateral and makes the worked example fundable at +8 %.
3. Admission was one-sided. v5 adds the supporter hurdle, the coverage-after-rescue floor and an affordability floor to the lender's EV test, so the stress case is refused for the right reasons (coverage 24 %, p = 35 %) and the over-collateralised case is enforced rather than rescued.
4. The worked example's figures were illustrative. The engine computes them exactly: installment 8,745.15; B_7 = 51,120.61; A_7 = 8,995.15; Q_7 = 51,754.02; claim pool 9,714.76 at 8 %.
5. Protocol units. Rates in 1/10 basis points, intervals in seconds, loan amounts as asset-agnostic Number strings, the management fee capped at 10 % — the builders refuse anything else.