Challenges I Ran Into

The $0.90 problem that almost broke everything. Our first implementation used simple yield calculation: $1,000 at 6% for 6 months equals $30 of yield, so send $970 to Morpho and $30 to the position. But $970 earning 6% for 6 months returns $999.10, not $1,000. That $0.90 gap seems tiny until you scale it. On a $10 million deposit that is $9,000 missing. You cannot call something "principal protected" when thousands of dollars are unaccounted for. We spent hours debugging this before realizing TradFi solves it with the zero coupon present value formula: PV = FV / (1 + r)^t. This gives the EXACT amount that grows to EXACTLY the deposit. No approximation. No gap. Once we switched to this formula, the math became airtight at any deposit size, which our 12.5 million simulation backtest confirmed with 100.0000% principal protection.

Real estate liquidation scanning was way harder than crypto. For crypto assets, checking if a position got liquidated during a holding period is straightforward because you have hourly or minute level price data. Real estate indexes from Parcl Labs update once per day. A position could appear safe at the daily close but might have breached the liquidation level intraday if we had finer data. We solved this conservatively by checking every daily data point during the entire holding period, not just the entry and exit price. This means our backtest slightly overestimates liquidation rates for real estate (because a daily close below liquidation triggers a liq even if the intraday low was only briefly there), but it ensures our principal protection claim is honest and defensible.

Yield rates kept changing on us. We initially hardcoded yield rates at 5 to 6% based on what Morpho historically paid. Midway through building, we pulled live rates from DeFiLlama and discovered current rates were 2.46 to 3.63%. This completely changed the product economics. Exposure budgets shrank, returns got more modest, and strategies that worked at 6% didn't work at 2.5%. We had to rebuild our backtest engine to pull live APYs from DeFiLlama's API on every run and recalculate everything. The silver lining is this forced us to find strategies that work at TODAY's rates, not theoretical ones, making our backtest results genuinely honest.

Making 5 sponsor SDKs work together in one flow was a nightmare. The WhatsApp shield activation touches Twilio for messaging, OpenAI for intent parsing, HeyElsa x402 for wallet analysis (paying real micropayments on Base mainnet), BitGo SDK for custodial wallet creation on Base testnet, our smart contracts on Base Sepolia for the actual deposit, ENS NameWrapper on Ethereum Sepolia for subname registration, and Fileverse agents on Sepolia with Pinata storage for encrypted document creation. That is 7 different services across 3 different chains in a single user flow triggered by a WhatsApp message. Getting the async orchestration right, handling failures gracefully at each step, and making sure the user gets a clean confirmation message even when one service is slow took more debugging than building the actual PPN math.

CoinGecko rate limits during backtest runs. Running 12.5 million simulations requires fetching historical price data for multiple assets. CoinGecko Pro has a 500 calls per minute limit. Our first backtest run hit the limit, got throttled, and returned incomplete data that silently corrupted thousands of simulations. We solved this by building a local cache layer that fetches historical data once, saves it to disk, and reuses it across all subsequent runs. This also made iteration dramatically faster since we could rerun the full backtest in minutes instead of waiting for API calls.