Yield backtest

How much would a Tako pool have paid over the last year of TON mainnet, compared with what Tonstakers and Hipo holders actually got? This page answers with on-chain data and a script anyone can rerun.

Result

Over 12 months, 2025-10-29 to 2026-10-27, 479 validation rounds:

	APY	Share of the network’s reward that reached holders
Network: every reward kept, all Gram at work	11.87%	100%
Tako, testnet parameters, modeled	9.41%	79.3%
Tonstakers, realized	8.68%	73.2%
Hipo, realized	10.37%	87.3%

Window	Network	Tako	Tonstakers	Hipo
12 months	11.87%	9.41%	8.68%	10.37%
6 months	18.03%	14.21%	13.32%	16.06%
3 months	17.24%	13.60%	12.38%	17.05%
1 month	17.25%	13.61%	12.52%	16.96%

The jump between the 12-month and 6-month numbers is the network itself: in April 2026 the reward per round rose about fourfold.

What it shows

• Tako stays ahead of Tonstakers in every window.
• With the testnet parameters, Hipo pays more. The difference is Tako’s own choice: 10% of the pool’s take to the insurance fund and 5% to keepers.
• Hipo is close to the ceiling in the last months, and no Tako setting beats it there.

Where the reward goes

Tonstakers lends at an interest rate its manager sets by hand, now 0.0564% per round. Over the year that rate was on average 88% of what the network paid per round; validators kept the rest. Tonstakers then keeps a 16% fee. 88% × 84% ≈ 74%, close to the 73% its holders got. The rate does follow the network: it was raised the same month the reward jumped.

Tako has no rate to set. Validators bid, and the pool keeps at least 97.25% of what each loan earned. What it then gives up is the insurance and keeper shares.

Insurance and keeper shares

Insurance / keepers	12 months	6 months	3 months	1 month
10% / 5% (testnet)	9.41%	14.21%	13.60%	13.61%
5% / 2%	10.34%	15.65%	14.98%	14.98%
2% / 1%	10.81%	16.38%	15.67%	15.67%
0% / 0%	11.16%	16.92%	16.19%	16.20%
Tonstakers, realized	8.68%	13.32%	12.38%	12.52%
Hipo, realized	10.37%	16.06%	17.05%	16.96%

The mainnet shares are chosen with this table in mind.

Method

• Network reward. For every election, the Elector’s past_elections entry is read at the last archival block before its stakes unfreeze, when the reward is final. bonuses / total_stake is what one staked Gram earned that round.
• Every other round. A stake stays frozen through the round after it validated, so each round half of the working Gram earns that round’s reward. The pool compounds.
• Tako applies the holders’ share and 97% of the pool at work to the same rounds. Lending is assumed to clear: validators take every loan at the testnet split.
• Tonstakers and Hipo are their share rates read at the same blocks every four rounds: total_balance / supply from get_pool_full_data and total_coins / total_tokens from get_treasury_state. That is what their holders really got, after every fee and idle Gram.

Limits

• Tako is a model; the others are realized. The biggest assumption is loan demand at the testnet split. If validators need a bigger share, Tako’s yield falls: at 90% kept by the pool and 97% at work, the 12-month figure is 8.68%, equal to Tonstakers.
• No slashing or failed rounds are modeled for Tako. They are rare, and the insurance fund takes the first loss.
• The backtest does not explain why Hipo comes so close to the ceiling; its share rate is simply what its holders got.

Reproduce

The script, its tests and the collected data live in the protocol repository under research/backtest:

npx tsx research/backtest/fetch.ts 365   # about half an hour on toncenter's free tier; resumes from the cache
npx tsx research/backtest/report.ts
npx jest tests/backtest.spec.ts