Harvest Finance

Summary

Harvest Finance launched on Ethereum on September 1, 2020, in the same yield-farming season that produced Yearn, as an aggregator for people who wanted farming returns without farming's overhead. Depositors put a single asset into a vault; the vault pooled deposits, deployed them into a strategy across external protocols, and periodically harvested and reinvested the rewards, so compounding happened at shared gas cost and nobody had to watch the fields themselves. The FARM token was a deliberate fair launch — no premine, no presale, no investor allocation, and a hard cap of 690,420 tokens emitted weekly over four years, split 70% to depositors, 20% to the development team, and 10% to a treasury. A fee taken at each harvest funded a profit-share pool that paid FARM stakers, tying the token to protocol activity rather than pure emission. Eight weeks after launch, the protocol suffered the flash-loan price-manipulation attack documented separately in this Atlas as the Harvest Finance Flash Loan Exploit; the vault-and-emissions design itself worked as specified, but TVL and prominence never recovered to their pre-exploit peak.

Design (Mechanism)

Each vault accepts one asset and mints fTokens (fUSDC, fDAI, and so on) representing a proportional claim: share price is the vault's underlying value divided by fToken supply, so value accrues by the share price rising rather than by rebasing balances. Strategies are separate contracts the vault delegates to — farm a reward token on an external protocol, sell it for more of the vault asset, redeposit — and the "do hard work" harvest call executes this loop for everyone at once, amortising gas that would make small positions uneconomic to compound individually.

The token design is the second mechanism. FARM's hard cap and weekly emission schedule made dilution legible in advance, and the 70% depositor share meant the majority of supply was earned by using the product. The profit-share pool closed the loop: a percentage fee on every harvest was used to buy FARM and distribute it to stakers, so staking yield tracked real vault activity — a revenue claim, not a pure emissions subsidy, at a time when most farm tokens were only the latter.

Outcome

Harvest scaled quickly through September and October 2020 as one of the era's largest aggregators, with TVL near a billion dollars at peak. On October 26, 2020, the flash-loan attack on its stablecoin vaults' share-price calculation drained an estimated $24–34 million and triggered a mass withdrawal; the team reimbursed affected depositors pro-rata from treasury and revenue, and the protocol kept operating — the vault mechanics, emissions schedule, and profit-share pool all continued functioning as designed, and the product later expanded to Polygon and Arbitrum. But the aggregator category consolidated around Yearn, the exploit permanently marked the brand, and Harvest settled into a long tail of modest TVL. The mechanism worked; the market position did not survive the trust shock.

Why it worked

• Pooled harvesting solved a real cost problem: compounding a small farming position individually could cost more in gas than it earned, and sharing one harvest transaction across a vault made compounding rational at any size.
• The fair launch with a hard cap and a published four-year schedule gave FARM credibility with the exact audience yield farming attracted, and the 70% depositor share made users and owners largely the same people.
• Routing a harvest fee into a profit-share pool gave the token a cash-flow link most 2020 farm tokens lacked, which is why the design outlived the season that produced it.

Where the design broke

• The vault share price read manipulable spot values, which is what the October 2020 exploit attacked; the aggregator's convenience concentrated depositors behind a single pricing assumption, so one flaw taxed everyone at once. The incident is documented in full in the companion exploit entry.
• Strategy risk was inherited, not created: an aggregator compounds whatever its strategies touch, so every underlying protocol's failure modes flowed through to vault depositors, while the aggregator's brand absorbed the blame.
• A 20% continuous team emission from a fair launch was defensible as compensation but sat awkwardly with the no-insiders framing, and became a persistent sell pressure critique as price declined.

Lessons

• Amortising gas across depositors is a genuine mechanism contribution, and it survives independently of any one protocol: shared-cost compounding became a standard pattern across DeFi.
• An aggregator is a risk concentrator wearing a convenience costume; its security budget and oracle assumptions must be sized for the sum of deposits, not the sophistication of the median depositor.
• Tying token yield to protocol revenue rather than emissions is what let FARM remain meaningful after the farming season ended; tokens without that link from the same cohort did not.

Redesign (EDITORIAL)

EDITORIAL — hypothesis, not fact. A version of Harvest built to survive its own success might (1) price vault shares from manipulation-resistant sources — time-weighted or multi-block — from day one, treating the share price as the protocol's most attackable number; (2) cap any single strategy's share of a vault and any single external dependency's share of the protocol, so one integration cannot tax all depositors; (3) fund an explicit, pre-sized depositor backstop from the harvest fee before paying profit share, converting the reimbursement it improvised after the exploit into a standing mechanism; and (4) vest the team emission against shipped milestones rather than the calendar, aligning the one non-user allocation with the product's continued development.