Reading Paths: Four Routes Through Arc's Documentation

Four curated routes through this site: domain understanding, system architecture, real-world scenarios, or interview prep. Pick the one that fits your goal.

Orientation · pick one

This site has about forty pages across four tabs. Reading it front to back works, but most people want one of these four routes.

---

The domain~50 minutes · no code

You want to understand how stablecoin cross-border payments work, well enough to evaluate a claim about them. This path never opens a source file.

1. Why corridors cost so much

   Correspondent banking, pre-funding, and where the 8.78% actually goes. Read →

2. How stablecoins work

   Issuance, redemption, reserves, and the difference between a peg and a promise. Read →

3. Settlement and finality

   Why “instant” is a marketing word and finality is a probability curve. Read →

4. The 2026 regulatory picture

   GENIUS Act rulemaking, MiCA’s transitional cliff, Kenya’s VASP Act, Nigeria’s ISA. Read →

5. The case against

   The BIS argument that stablecoins fail the three tests of sound money: taken seriously, not strawmanned. Read →

Finish at the bibliography if you want the primary sources.

The system~90 minutes · code and diagrams

You want to read the architecture. Order matters here: the ledger has to exist before money can move.

1. Architecture overview

   Six bounded contexts, one deployable, an event bus between them. Read →

2. Why money is never a float

   The single decision the rest of the system is built on. Read →

3. The ledger

   Chart of accounts, the balance rule, the worked EUR→KES example. The core. Read →

4. Contexts and events

   The outbox, at-least-once delivery, and how boundaries are mechanically enforced. Read →

5. The chain layer

   Chain-agnostic driver, deterministic simulator, reorg rollback. Read →

6. The settlement saga

   Five steps, five compensations, and why order matters. Read →

7. What the tests prove

   Property-based, mutation-checked, chaos-injected. The proof layer. Read →

Then walk a flow end to end to see it assembled.

IRL Scenarios~40 minutes · start anywhere

Seven narratives. Each opens on a symptom, walks the diagnosis, and lands on the specific piece of Arc that prevents it. They are self-contained and can be read in any order, but if you read only one:

The journal that balanced and lied

A mutation that broke the audit trail and passed all sixteen tests. The best argument on this site for why balance is necessary but not sufficient.

The rest, roughly by difficulty:

• The cent that vanished: rounding residuals
• The payout that paid twice: idempotency under timeout
• The cut-off that cost a day: batch rails and compounding delay
• The fastest chain that settled last: block time is not finality
• The quote that aged: stale rates as unhedged loss
• The boundary that wasn’t: an architecture rule that silently didn’t hold

Interview prep~2 hours · active recall

You have a payments, ledger, or infrastructure interview. Do this in order and answer out loud before reading the answer.

1. Read the ledger and the saga

   You need the ledger and the saga in your head first. Everything else builds on them.

2. Work the question banks

   Ledger · Distributed systems · Chain and settlement · Compliance

3. Self-check

   Ledger quiz · Settlement quiz

4. Steal the scenarios

   A concrete failure narrative beats an abstract answer in every interview. The IRL scenarios are there to be retold.

The strongest answers on this site are the ones about what didn’t work: the mutation that passed, the boundary rule that didn’t fire. Interviewers remember those.

---

Conventions used throughout

A teal block is a load-bearing claim the code enforces: there is a test or a database constraint behind it.

An amber block is a known gap: something modelled but not wired, or a limitation accepted on purpose. These are stated rather than hidden.

Account codes appear as structured strings, asset.float.bank.EUR, liability.customer.va_1.KES: because a journal should be legible in a log or a psql session without joining anything.

All monetary figures are integer minor units. 10000n is €100.00. There are no floats anywhere near money, and there is a lint rule that makes sure.