How Phantom built the attribution engine behind $100M in $CASH

Meet Phantom

Phantom is the leading self-custody wallet on Solana founded by Brandon Millman, Chris Kalani, and Francesco Agosti. Over 20M users rely on it for trading, staking, swaps, and now stablecoin payments – all through a single app. Backed by a16z, Paradigm, Sequoia, and Variant, it's become the default entry point for millions of people using crypto for the first time.

In September 2025, they launched CASH: a USD-backed stablecoin built on Stripe's Open Issuance platform, featuring fee-free P2P transfers, instant bank funding, and Visa debit card integration. The goal: make stablecoins work for everyday payments.

But CASH isn't just for Phantom users. It's an open loop stablecoin with multiple distribution partners who each earn on the reserves for supply they originate.. That means Phantom needs to know exactly how much CASH each partner net minted, down to the token level, to calculate monthly reward distributions.

The attribution problem

Stablecoin attribution sounds simple until you try to build it. CASH lives on Solana, where transactions happen fast and the account-based model means tokens quickly become intermingled. The complications stack up fast:

• Ownership changes constantly: A token minted by Partner A might be transferred ten times before it's burned. Who gets credit?
• Logic isn't obvious: Should attribution follow first-in-first-out? Last-in-first-out? The answer affects millions in payouts.
• Solana moves fast: Millions of on-chain events need to be processed and mapped to the correct partner; eventually this will need to happen in real time.
• Auditability is non-negotiable: Partners need to trust the numbers. Every attribution decision has to be traceable back to onchain data.

Any missed event or misattributed token means incorrect payouts.

Finding the right fit

Phantom needed more than a generic indexer. They needed a system flexible enough to handle custom attribution logic and fast enough to iterate when that logic changes. Turbo let Phantom's team build exactly what they needed:

• Custom from the ground up. The ingestion engine was designed specifically for CASH: tracking mints, transfers, and burns with the exact schema Phantom's data team specified. No middleware or transformation jobs.
• Iteration without rebuilds. When the team tested different attribution approaches – FIFO, LIFO, hybrid models – changes shipped in minutes rather than days or weeks. Full historical data replay from genesis meant the cost of iterating and trying a new model was nearly zero.
• Snowflake-native delivery. Data flows directly into Phantom's internal warehouse, where their analysts run payout calculations alongside other business data from one source of truth.
• Real-time, even when they don't need it (yet). Monthly payouts are the current cadence, but Turbo's architecture future proofs ability to expedite reward payments in the future. Being able to expand to complex logic and other tokens/chains is also trivial.

Results that compound

Speed to production

• <2 weeks from project kickoff to live production
• 6 rounds of attribution logic iteration with zero infrastructure rebuilds

Operational efficiency

• 40+ hours of manual reconciliation eliminated monthly
• Zero ETL jobs – data flows directly into Snowflake with no transformation layer

Scale and reliability

• 10M+ onchain events processed and attributed to the correct issuance partner
• 99.99%+ uptime sustained since launch

Within two months of launch, CASH crossed $100M in circulating supply, and Turbo tracked every token along the way. The pipeline has processed over 10 million onchain events, attributing each mint, transfer, and burn to the correct issuance partner. Partners now receive accurate monthly reward distributions based on this data.

The speed to production was just as striking. Phantom went from project kickoff to a live pipeline in under two weeks, then iterated through half a dozen rounds of attribution logic without rebuilding any infrastructure. When Phantom decides to expand the program (monthly to instant payouts, a handful of partners to a broader referral program, etc.), the pipeline won't need to change.

The operational wins are equally concrete. Data flows directly into Snowflake with zero transformation layer needed; no ETL jobs, no middleware, no internal work at all. The team estimates 40+ hours of manual reconciliation eliminated monthly.

“Turbo has saved us several days of engineering work involved in maintaining an algorithm for CASH partner mint/burn attributions. This means we can focus that time instead on efforts to expand CASH to other chains, defi opportunities, partners, and in-app utility rather than spend resources on maintaining this data indexing and attribution pipeline. The data is easily auditable and we can rely on the Goldsky team to quickly support us on any data quality ask.” — António Mendes, Senior Data Scientist @ Phantom

What’s next

$CASH is still early. As the stablecoin grows, both on Phantom and in the ecosystem, the attribution engine will need to handle more volume, more partners, more complexity, and entirely new use cases.

Phantom has always built for the future, and with Turbo powering the stablecoin launch, infrastructure won’t be holding them back.

Unblock your data team and make money move

👋🏻 Hey, we're Goldsky. We help teams like Phantom build attribution engines, payout systems, and financial infrastructure that make stablecoins work at scale. If you're launching a stablecoin, tracking distributions, or building anything where accurate onchain accounting matters, we'd love to chat to help you ship faster. Sign up for free or schedule a call to see how we can help.