alphafi on sui · may 2025 to sep 2026 · @kitty4DHD
alphafi wound down. here's what the chain says happened.
on sep 25 2026 at 23:09 utc alphafi tweeted that it was "winding down in an orderly way." i had a withdrawal ticket in its ALPHA vault at the time, so i went & read the chain. this section is everything i found: the 68 minutes before the tweet, the lock that didn't actually unlock everyone, the $17.1M of bad debt & the 4 positions behind it & the september sequence that paid it off. i'm not saying i know who anyone is. every claim links to a tx so you can check me.
the numbers
head start
68 min
on-chain wind down began before the tweet
bad debt repaid
$17,129,783
sep 20 2026, one transaction, nine loans
positions behind it
4
5.7M ALPHA posted, 57% of supply
freed early
86,672 ALPHA
october-stamped tickets, released by a sep 26 vault upgrade
deposits after the depositor's own borrow; ~2 expected by chance
rescue USDC from coinbase
~$18.9M
all the USDC the bailout & the sep 8 to 9 repays used
the short version
the shutdown started on chain at 22:01:01, 68 minutes before anyone was told. emergency mode on 7 slush & loop pools, then at 22:16:48 the ALPHA vault's 28 day lock went to 0, then at 22:47:59 borrowing was cut on all 36 alphalend markets. the tweet came at 23:09:29. the 68 minutes.
exactly one wallet claimed ALPHA from the vault b/w the lock change & the tweet: 0x59faca57…4725, 11,566 ALPHA at 22:35:54, sold by 23:04. a027's 214,001 in the same batch sat until 23:21:04; the other 24 receipts (23,729 ALPHA) didn't move.
the shutdown-night lock change didn't unlock everyone. a ticket gets its unlock date stamped at the weekly settle & the contract enforced the stamp. 51 requests settled aug 30 to sep 20 (86,672 ALPHA) were stamped sep 27 to oct 18, so they watched ALPHA go from ~$0.93 to ~$0.02 from inside the lock. then a second upgrade the next morning wired lock=0 to skip the stamp & let them claim early. the queue is empty & the lock is 0, but ~3.4M ALPHA is still staked there (1,228 positions on oct 15), free to leave. the vault lock.
that one early wallet is part of a cluster of at least five linked by funding & timing. four of them pulled all their alphalend collateral b/w 18:02 & 18:15, five hours before the tweet, on an otherwise boring afternoon. the cluster has done business w alphafi's treasury wallets since nov 2024: 38 BTC relayed for conversion, 50k ALPHA & 100k USDT in one evening, OTC ALPHA buys. who moved early.
idear.sui, the other big seller, also left alphalend early (44,318 SUI at 17:29) & had been paid 25,404 USDC by alphafi wallets over the prior year. purpose unknown. i have guesses.
the ~$17.1M of bad debt was paid on sep 20 at 22:50 by a wallet created on sep 14. so the problem was known at least 11 days before the announcement. the four positions were opened may to dec 2025 by four wallets that alphafi's fee wallet gasses, funds, or receives borrowed money from. the protocol booked their ALPHA at $2.77, $3.65 & $14.16 while it traded around $1. the bad debt.
the fee wallet held the ownership caps of all four, from dec 25 2025. it kept them in a holder object built from a package the vault admin published two hours after P4 opened & borrowed $12.47M on P4 itself, every dollar sent on to operator hubs within minutes. the borrowed money.
alphafi's own sdk changed ALPHA's price feed in step w four of the five remaps, one of them 4m37s before the chain switched it to AVAX's feed. raj's account synced every vault-admin upgrade within minutes for six weeks & raj says, in his own words, that he raised a market's liquidation threshold that only the lending admin ever signed. the code.
about $21.5M went one-way into 16 exchange accounts & 52 of 201 deposits landed within an hour of the depositor's own alphalend borrow (about 2 expected by chance), 37 of them for the same amount. the exits.
september reads as a workout, not a heist: the fee wallet spent its treasury on repayments, concentrated the leftover debt on one ALPHA-only position, handed its real collateral to the rescue side, which paid about $27.6M, sent about $12M of security into kraken deposit addresses & released the users' ALPHA. the fee wallet held about $2,300 on sep 27. september.
the rescue's USDC came out of coinbase & its gas out of kraken. the wallet that analysts call the sui foundation's gave alphafi's validator its 30M SUI joining stake & pulled it out on oct 6 2026. both sides use the same two exchanges; no wallet connects them. who paid.
none of this proves who these people are, or whether the early movers were tipped vs being the team itself. both readings are bad for alphafi. the chain shows what moved & when. it does not show intent. method & corrections has everything i got wrong along the way, cuz there was some of that.
the pages
roughly in the order things happened. the oracle timeline is the older piece that started this & the rest builds on it.
short names i use everywhere. wallet links open suiscan, git accounts open github & x accounts open x, all in a new tab. "treasury" & "rescue side" are my labels, based on what the wallets do, not on anything anyone signed. a git account isn't a wallet: nothing public ties one to the other.
alphafi operated
vault admin0xe25b5d16ca31ddfdc31a7219c90f88bdfc56b606c13df6619aef22515580e293changed the lock, ran the settles, paused deposits, activated emergency mode on the pools, did the vault package upgrades. holds the ALPHA vault's UpgradeCap; published the holder package the caps were kept in.
f71e (treasury ops)0xf71e0bc45d84bba3e1ab09c8ae636735da6f2b8d8bb2481fb5b437b537b0e65bfunds alphalend rewards, pays the keeper, receives strategy withdrawal batches. its "payroll-shaped" payouts turned out to be gate & kucoin deposits. its alphalend position became the security package. on 2025-09-03 it bought the nft raj set as his x avatar 3 minutes later (the code).
the alphafi validator0xa0920b0776bf13ee51b009c97b85dbf48100ae1510623b9a767450e4a481a1e2its 30M joining stake came from 0x341fa71e, which withdrew it on oct 6 2026; the validator itself quit on oct 12 (voluntary exit) & its wallet sent its last 14,682 SUI to 0x862d2bfd the same morning; its commission went to the operator wallets; holds six team position caps.
0x5f81 (holds the caps)0x5f81363f48573dd002e09f34217ed00976afa748731bce61ea03e155f918d475alphafi-made, rescue-used. the fee wallet gassed it fresh on sep 7 & gave it all five PositionCaps; from sep 8 the rescue's USDC (via 0xf2f1) repaid P4 & P2 through it. who controls it isn't on chain.
exchange deposit accountsokx · gate · binance · kucoinwhere ~$21.5M of stablecoins went one-way: okx $11.57M (all from P1's creator), gate ~$5.25M (13 operator wallets, the last two deposits on oct 15 2026), binance $1.93M, kucoin ~$1.85M (incl the validator's commission), kraken ~$0.92M. the exits.
rajalphafigithub.com/rajalphafi↗raj, commits as raj@alphafi.xyz. 8 commits in alphalend's sdk, ever: merged the oracle v3 switch, wrote the v4 one. synced all 41 of the vault admin's packages in may & jun 2025. the code.
11felixgithub.com/11felix↗wrote every ALPHA price-feed change in the sdk (#69, #114, #120, #121) & the oracle v3 & v5 switches. top committer.
jangidgithub.com/jangid↗merged the feed PRs & 123 of the sdk's 175. profile company "AlphaFi". the AI auditor repo raj launched as alphafi's moved into his account on sep 24 2026, the day before the shutdown.
Zorag44github.com/Zorag44↗his copy is where the old vault sdk survives (the org's repo is gone). 100 alphafi-sdk-js commits since nov 2025.
rishikesh-chavangithub.com/rishikesh-chavan↗wrote #57, the lending package switch raj merged 8 min after the publish.
harshaalphafigithub.com/harshaalphafi↗merged the multi-position withdrawal code as PR #103 on sep 27 2026, 34.5 h after the tweet. built the betterstack incident bot.
sid-alphafigithub.com/sid-alphafi↗commits as "sid" w no account attached, but this account opened the PRs carrying them (17 across the two sdks, incl #103). wrote that withdrawal code on sep 22 & 23 2026, two days before anyone was told.
rg-alphagithub.com/rg-alpha↗28 of the 50 commits in alphalend-contracts-interfaces, the repo whose function bodies are mostly abort 0 stubs.
@0xrajendrax.com/0xrajendra↗raj, alphafi's ceo. posts on open-sourcing, audits & the IKA threshold; the last one i can find is his sep 26 wind-down reply↗. account private since oct 6 2026. his avatar nft was bought by the treasury ops wallet 3 minutes before he set it. the code.
@SuiNetworkx.com/SuiNetwork↗sui's statement↗ (archived) six minutes after alphafi's: the foundation chose to "provide assistance". "No user funds are impacted."
position labels P1 to P4 follow the oracle timeline page (P1 = 0x9e9d, P2 = 0x74d3, P3 = 0xd439, P4 = 0xf0a4). a few of the receipts cards use different letters; the receipts page says which. times are utc. i'm one person w a public rpc endpoint, not an auditor & the write-up behind this has been through six revisions of me finding my own mistakes.