Risk
How your funds are protected, what was tested, and what each result shows.
How your funds are protected
Your funds are protected by 4 layers of backstops before a loss touches your principal.
Each pool stands alone. A loss in Earn's vault reaches neither Borrow's pool nor the company's operating cash. Borrow has its own four layers: the borrower's collateral, the primary and tail reserve sleeves, and the pilot backstop during its pilot.
Protection does not start at zero. The pilot backstop, a small portion of the capital Sacred raises from investors, protects the pilot phase of both Earn and Borrow while primary and tail protection build.
Four controls sit around that order:
| Wallet control | You hold your own keys, through an embedded wallet you can export or a wallet you connect |
| Contract security | New contracts go through independent review before launch, followed by a small bounded pilot |
| Collateral controls | Pledged assets are never lent, staked or reused. Coverage limits, price checks and depth caps govern every sale |
| Withdrawal conditions | Withdrawals rank ahead of new financing. Earn pays from idle cash, then a daily queue. Repayment and exit stay open during pauses |
The layers absorb trading losses. Failures of a contract, an oracle, USDC or the chain sit outside them, and are listed under known risks.
Reviews and audits
| Smart-contract audit | Not yet |
| Shariah review | Not yet |
| Risk research and test results | Below |
Reading the numbers
Three kinds of number appear below and they are different things:
| Kind | What it is | Example |
|---|---|---|
| Historical loss | What a replay of past prices produced | No principal loss at 2× in the Binance normal replay |
| Planning allowance | A budget line for expected loss | 5 bps a year on financing |
| Policy cushion | Cash chosen to absorb losses | 1% of approved financing |
None of them is a statistically certified lifetime guarantee. Nothing here was run on Arc. The results are price history from centralized exchanges.
Earn and Trade
The final tiers
| Leverage | Trigger / restore | Initial price room | Protection stock | Status |
|---|---|---|---|---|
| 2× | 160% / 190% | 19.7% | 1.0% | Opens first |
| 2.5× | 145% / 160% | 12.6% | 1.5% | Opens gradually |
| 3× | 130% / 145% | 12.9% | 9.0% | Opens gradually |
| 3.5× | 125% / 135% | 10.2% | 10.0% | Opens gradually |
| 4× | 120% / 130% | 9.4% | 14.0% | Opens gradually |
| 4.5× | 116% / 125% | 9.1% | 17.0% | Opens gradually |
| 5× | 115% / 123% | 7.2% | 17.0% | Opens gradually |
Stocks are chosen from 1.10 × the larger one or three-month-block four-hour annual ES99, rounded up, with floors of 1%, 1.5% and 9% at 2×, 2.5× and 3×. Nested financing caps limit the share above 2× (ladder). There is no diversification credit.
The full hourly history
Binance BTC and ETH, January 2018 to September 2026, about 76,000 overlapping 14-day cohorts per asset and configuration. Missing timestamps are kept as outages, so adverse windows are not deleted.
| Scenario, 2× | Result |
|---|---|
| Normal, 1 hour | No principal loss |
| Four-hour delay, 3% friction, Binance | No principal loss |
| Four-hour delay, 3% friction, OKX ETH | Six losing daily cohorts, worst individual loss 3.42%. Maximum annual book loss 0.835% of nominal ETH capacity, about 0.418% on an equal BTC and ETH book |
| Bybit, July 2021 onward | No normal-case principal loss |
| 24-hour outage, 5% friction, Binance | Worst individual loss 21.71% BTC, 27.99% ETH. Worst rolling year 14.45% of peak financing |
The 24-hour case is a separate emergency. The 1% routine stock does not cover it.
Partial sales are ordinary. In replay, 9.4%, 11.2% and 14.3% of 2× positions entered in bull, sideways and bear conditions saw a forced sale.
Normal annual principal-loss rates are zero at 2× and 2.5×, and about 0.107%, 0.68% and 1.18% of actual financing at 3×, 4× and 5×.
The realistic-delay panel
224 re-verified minute archives: 105 systematic monthly dates plus seven stress dates, for each asset. Each observation starts hypothetically exactly at the trigger and takes the worst fall over the detection minute plus five more, with 1% exit friction. It is a gap test, not a continuous replay or a liquidation frequency.
| Maximum shortfall on stress dates | BTC | ETH |
|---|---|---|
| 3× at 130% | 0% | 0% |
| 4× at 120% | 1.07% | 4.80% |
| 5× at 115% | 5.19% | 8.77% |
The largest stress-date gaps were 16.71% on BTC and 19.85% on ETH. Every tier had zero hypothetical shortfall on the systematic dates. Windows are not independent, and these frequencies cannot be annualized.
A saturated mix
At the most concentrated mix the caps allow, with correlated losses summed before aggregation, worst rolling annual loss is 0.186% in the one-hour case and 0.856% in the four-hour case, against total approved capacity. ES99 is 0.336% and 1.489%. The 2.6% stock exceeds those modeled cases. The 24-hour emergency still loses 16.53% in the worst historical rolling year.
Month resampling produced 30,000 one-year paths from 103 complete months. Those are not 30,000 independent years.
What SIFT adds, and what it does not
Account-level limits sit on top of the position-level ones, and they answer different questions.
| Coverage | Account drawdown | |
|---|---|---|
| Measures | One position's asset against its debt | One account's unit NAV against its high-water mark |
| Trigger | 160% at 2× | 10% pauses growth, 20% halves capacity, 30% freezes it |
| Effect | Partial sale of the asset | Less or no new financing. Existing sales unchanged |
- Common shocks. Many wallets buying BTC are one correlated liquidation. A transparent score cannot replace aggregate asset and route limits. Production limits must come from execution evidence, and 1%-of-financing protection is not a bound on an arbitrary crash.
- Sticky limits. Risk ceilings and freezes have no automatic recovery in v1.
- Identity. Wallet splitting is limited by no double pledge, one newcomer budget and aggregate caps, not by knowing who is who.
- Oracles. Canonical daily prices and rounds are versioned settings still to be chosen. Stale or missing data fails closed for new financing.
- Selection can backfire. Aggressive selection can leave capital idle and reduce income.
Borrow
Borrow's policy is 45% entry, 62.5% liquidation and 50% restoration, the last two measured against full contractual debt. The 62.5% and 50% figures are controls chosen by judgment. The tests below describe how they behaved against history.
| Test, 2018 to 2026, 30-day sales at 8% | BTC annual loss on financing | ETH annual loss on financing | Equal book |
|---|---|---|---|
| Binance one hour, 0.1% friction | 0 | 0 | No shortfall |
| Binance four-hour path, 3% | 0 | 0 | No shortfall |
| OKX four-hour path, 3% | 0 | 0.02462% | Worst year 0.101%, ES99 0.162% |
| Binance 24-hour path, 5% | 1.4886% | 1.5044% | Worst year 12.24%, ES99 18.55% |
Sources are Binance, OKX and Bybit, with sampled minute gaps. Kraken lacks comparable history and no replay is claimed for it. The Bybit window excludes March 2020. At the 62.5% trigger, gross collateral is 1.6 times full debt, leaving 37.5% of further price fall before debt exceeds collateral, less slippage and costs.
What the 1% reserve covers. It is about 6.2 times the four-hour ES99, which is a judgment margin. The day-long emergency would need about 19% of financing, so it remains uncovered. Investors bear it.
Sale frequency. In healthy-repayment replay, 8.99% of BTC and 19.31% of ETH daily cohorts saw a price liquidation.
Route controls. A route's fixed-debt book cannot exceed 25% of independently measured USDC bid depth within five minutes at 1% slippage. Common bridge, oracle or custodian exposure counts together.
Maturity concentration. At most 5% of approved financing due in one day, and 30% in seven.
Price controls. New-risk pricing must be 60 seconds fresh or better, with two independently operated cross-checks. Over 1% disagreement, or USDC outside 0.98 to 1.02, freezes new risk. Debt health prices USDC at max(1, validated USDC/USD), so a depeg cannot create borrowing power. Repayment and top-ups stay open. Enforcement needs valid prices and bounded quotes.
Governance. Risk increases carry a 35-day timelock, apply prospectively and stay within the mandate. Reductions and pauses can be immediate.
Known risks across all three products
| Risk | What could happen | What limits it |
|---|---|---|
| Contract bug | Loss of deposits or pledged assets | Audits, caps and a bounded pilot. No audit has happened |
| Stale or wrong price | Bad sales, or new risk at a wrong price | Freshness and cross-checks. Enforcement needs valid prices |
| Thin Arc markets | Sales fail or fill badly | Depth caps and route limits. Arc depth is unmeasured, and a route is at zero capacity until proven |
| Keeper delay | Sales arrive late | A 60-second target, unmeasured. Keepers are permissionless |
| Wrapper or bridge failure | cirBTC or WETH loses value or redemption | Independent wrapper checks and route caps. BTC and ETH prices cannot prove backing |
| USDC freeze or depeg | Circle can block an address. A depeg hits every pool | New-risk bands. Nothing inside the design fixes a freeze |
| Arc outage | No sales or exits during a halt | Nothing inside the design |
| Withdrawal wait | Earn depends on settlement. Borrow investors are committed for 12 months | Withdrawal priority, idle floors and maturity limits |
| Admin keys | A captured key changes future settings | Timelocks, bounded limits, and pauses that protect exits |
| Regulation | Eligibility differs by country. Leverage is restricted in some, including Türkiye | Legal review before each market |
| Shariah | A board declines a key structure | The design changes. See Shariah |
Unknown or untested
- Live depth, spreads, MEV, keeper and oracle behavior on Arc.
- Real borrower repayment and real trader behavior.
- Depositor willingness to accept conservation terms.
- Anything a professional audit would find.
- A crash worse than the history above. The past does not bound the future.