Before Parallar sells protection on an asset, three questions need answers:
how risky is this borrower, what is fair protection worth, and
what does the market think?
Each answer is computed by a certified engine, and every number comes with a
cryptographic receipt showing exactly how it was made.
An engine reads the borrower's financial statements the way a seasoned underwriter would: debt, earnings, cash flow, payment history. From these it scores the chance of default. A built-in safety rule stops it from ever calling anyone "too safe to fail."
A second engine turns that risk score into a fair annual price for protection, priced the same way the professional credit-insurance market does it. Safer borrower, cheaper protection; riskier borrower, dearer.
Where a real market quote exists for the same name, a third engine reads it as a second opinion and blends it in. The recipe for the blend is published, never improvised.
The engines pass their answers down a line, and every hand-off carries a tamper-evident seal: change even one byte of the data anywhere, and the chain of receipts visibly breaks. Nothing depends on trusting the operator. The seals do the trusting.
Reads the borrower's own numbers and scores default risk, anchored to decades of real-world default history by credit grade. A good story can tilt the score but never rewrite the base rate.
Turns the risk score into the fair price of protection, using the same two-sided arithmetic the global credit-default-swap market runs on. The asset's terms are inputs; the method never bends per deal.
Reads a real, timestamped market quote and converts it into the market's own implied view of default risk. Kept clearly separate from the model's view: a second opinion, honestly labeled as one.
Combines the model's view and the market's view into one number, with the mixing recipe (70% model / 30% market) published and fixed. A name with no market quote keeps its model score. Nothing is invented.
The engine already running behind the live testnet: it turns the whole covered book into a premium, an expected loss, and a worst-case reserve. These are the numbers a settlement actually pays on.
A model looking only at financial statements can talk itself into absurd confidence: odds of default of 1-in-a-million for a spotless borrower. The anchor forbids that: no borrower scores better than the safest credit grade's real historical base rate (about 2 in 10,000). Ordinary and troubled borrowers are barely moved. Each row is one example borrower; the scale stretches so tiny and huge risks fit together.
| archetype | v1 PD | v2 PD |
|---|---|---|
| investment grade | 0.012 bp | 2.02 bp |
| strong | 0.13 bp | 5.14 bp |
| mid | 1.370% | 1.359% |
| stressed | 99.99% | 99.98% |
| near-default | 99.998% | 99.996% |
Protection on a safe borrower costs almost nothing per year. On a troubled one, the annual price explodes, and the expected years of premium collapse (right column), because a name close to default won't be paying premiums for long. That is exactly how the professional market behaves, which is the point: the engine reproduces it, checkably.
| name | fair spread | RPV01 |
|---|---|---|
| investment grade / strong | < 1 bp | 4.63 y |
| mid | 75 bp | 4.47 y |
| stressed | 40,832 bp | 0.15 y |
| near-default | 43,856 bp | 0.14 y |
Every settlement boils down to a small sealed receipt, 208 bytes plus a proof, that any chain can check for itself. Stellar checks it and pays out. An Ethereum-style chain checks the same receipt and publishes the verified facts, so lending markets there can rely on them without trusting anyone's word, including ours.
No bridge. No wrapped assets. No oracle committee. Read-only verification mirroring: the mirror is a fact oracle whose facts are proofs.
journal.v1 · 208 B frozen layout · sha256-keyed on both chains · 5-input Groth16 layout shared verbatim
For practitioners: PD/LGD scorecard with a grade × sector through-the-cycle cohort anchor · ISDA-lineage two-leg CDS pricing on a flat hazard · credit-triangle market-implied PD (risk-neutral, labeled) · pinned log-odds blend with a disclosed Q→P haircut · bit-exact fixed-point↔float parity gates · full specifications in the documentation.