the code moved w the oracle, minutes ahead of it

ALPHA's price feed, in the sdk & on chain

the sdk hardcodes ALPHA's pyth feed (an override in src/core/client.ts). every on-chain change below was signed by the lending admin 0xa1eb94d1…. feed names are pyth's own catalog. "gap" is sdk time minus chain time, so negative means the sdk got there first.

every one of those sdk feed changes swaps in a whole different 64-character feed id (SUI/USD 23d73151…, STSUI/USD 0b3eae8c…, APT/USD 03ae4db2…, AVAX/USD 93da3352…), not a typo of a nearby value & the branches are still in the repo under their own names: feature/alpha-sui-price, feature/alpha-stsui-price, feature/alpha-apt-price. the XAUm (gold) mapping on nov 22 2025 & the dec 9 flip-flop got no sdk change.

& through all of it the sdk's ALPHA override kept reporting ALPHA at its coingecko market price (pythPrice: alphaCoin.coingeckoPrice), while the protocol's own oracle booked it at SUI's, stSUI's, APT's & then AVAX's. whether alphafi.xyz's screens used that field i didn't check, so i'm saying "the sdk", not "the app".

raj's commits, all of them

raj is alphafi's ceo (that's how he's introduced on stage). on github he's rajalphafi↗ & commits as raj@alphafi.xyz. pulled from every branch of alphalend-sdk-js by login & by email, plus every PR's merged_by. no commit by him exists there before 2025-04-29.

so: 8 commits in that repo, ever (archived list), out of 694 on main. six are PR merges from two days in april 2025 & two of those merges put a package the lending admin had just published into the sdk, 8 & 24 minutes after it went on chain. after P1 opened (2025-05-02) he committed there exactly twice, both on dec 9 2025, both the oracle v4 bump. he opened no PRs. jangid merged 123 of the 175.

& 27 more, in the old vault sdk

his only other public commits anywhere since april 2025 are 27 in alphafi's old vaults sdk, may 4 to jun 12 2025. that repo (AlphaFiTech/alphafi-sdk) is gone from the org now, deleted or made private on a date github doesn't show, but the commits survive in Zorag44's copy, Zorag44/alphafi-sdk↗ (his commits there, archived). 21 of raj's 27 switch the sdk to new package ids, 41 packages in total & every one of the 41 was published by the vault admin 0xe25b5d16…. the commit always comes after the publish, never before: 19 of 21 within 20 minutes of the batch's last upgrade, median 9.

the other 6 of the 27: five "Merge branch 'main'" pulls (they bring in other people's ids, incl. three god-wallet packages; not his) & one that adds a bluefin pool object. on 2025-05-23, the day after the cetus exploit, the vault admin upgraded ten packages in 47 minutes & raj's commit landed 14 minutes after the tenth.

the other direction is the stronger half. every package the vault admin & the god wallet published from 2025-04-30 to 2025-12-09, against the first commit in either public sdk that adds it:

for six weeks, every upgrade the vault admin made reached the sdk through raj's account within minutes (median 18 after each package's own publish) & none of the god wallet's did. when he stopped, others picked the job up w lags of weeks (11felix's median is about 15 days). the vault admin is the wallet that ran emergency mode on seven pools at 22:01 on shutdown night, zeroed the ALPHA vault's lock at 22:16:48, published the sep 26 v3 upgrade that let the locked tickets out (the vault lock) & published the Holder package the fee wallet kept the bad-debt caps in (the borrowed money). it doesn't prove who holds the vault admin's key. it's as close as public data gets to "the same person, or someone sitting right next to them".

raj, in his own words: "I raised the liquidation threshold"

dec 12 2025, about alphalend's IKA market, from raj's own account: "I raised the liquidation threshold when price was bottoming so users wouldn't get liquidated" ↗.

here's what that market's settings did on chain. IKA is alphalend market 15. its settings change through a MarketCap object, 0xe1e2db42…, whose client_address is the lending admin & every update_market_collateral_ratios call on it was sent by the lending admin:

the only threshold raises before his post are oct 29 & nov 19 2025, both signed by the lending admin. that's 0xa1eb94d1…, the same wallet that published every oracle version & re-pointed ALPHA's feed at AVAX. raj says the decision was his. the chain says who signed it.

"flow limits cap the blast radius at $250K per market"

jul 14 2026, from raj's account: "Flow limits cap the blast radius at $250K per market"↗ (screenshot). a flow limit caps how much can leave a market (borrows & withdrawals) in a rolling window. each market's is set w update_flow_limiter through its MarketCap. USDC is market 6, its cap is 0x9b8879a0… & every call on it was sent by the lending admin. here's the USDC limit around that post & the fee wallet's sep 6 borrows:

a limit of 0 doesn't mean nothing can leave. alphalend's own flow limiter code skips the check entirely when the limit is 0 (flow_limiter::check_and_update returns before it counts anything), so 0 is off. it was off for an hour, set to 300,000 at 18:03, 4,900,000 at 18:28 & 250,000 the next morning. it also got switched off for the ALPHA market on sep 25 at 11:38:51 & for all 36 markets at 17:30:52, hours before the shutdown post (september).

who can sign what

read from the objects each wallet owns today (sui graphql, 2026-10-05).

• the lending admin 0xa1eb94d1… holds the oracle's UpgradeCap 0x003f74ba… (now at v5) & the UpgradeCap of alphalend's lending package itself. it published every oracle version: v1 2025-04-27, v2 04-28, v3 04-30, v4 12-09, v5 2026-07-31. every alphalend MarketCap i read (the 21 markets whose ratios changed through feb 2026) names this wallet as its client_address. it signed every ALPHA remap & it's the lending protocol's protocol_fee_address, where alphalend's fees go.
• the vault admin 0xe25b5d16… holds the ALPHA vault's UpgradeCap 0x7ab23af8… & signed vault v1 to v3 & holds the UpgradeCap of the Holder package the bad-debt caps were kept in.
• the god wallet 0xa511088c… published ALPHA & holds 112 other UpgradeCaps, but none of those. (i'd assumed the opposite on the gate at first; that's fixed now.)

& the oracle's code isn't as unreadable as i first said. sui hands back each package version's disassembly (sui_getObject w content), so v3 & v4 can be diffed. same interface, 9 function bodies changed. the ones that matter:

• update_price_from_pyth: last_updated becomes chain time instead of pyth's publish time & an update older than the stored one is skipped.
• update_pyth_identifier_for_coin: on a re-point, clears the coin's old feed ids from the reverse map (v3 left them behind).
• update_max_age: a 120 s floor. add_coin_to_oracle: validates the threshold & the coin kind.
• smaller: a div becomes a mul in parse_price_to_number's confidence branch (unused by USD feeds) & a supra key type goes from u8 to u64.

P4's $14.16 booking on dec 3 ran under v3. on dec 9 v4 went live at 10:11, raj's sdk bump followed at 10:20 & the APT/AVAX flip at 12:46 to 12:52 is consistent w using v4's new re-point cleanup. the move registry record that claims to be this oracle's source (@altest.sui/allor, 0xb152f5a1…) is a separate v1 copy published 2025-09-07 by another wallet. the real one, @alphalend/oracle, has no public source. the only public alphalend move repo is interfaces whose function bodies are mostly abort 0 stubs.

the other accounts

commits since 2025-04-01 by account, from the github api (2026-10-04). an account's name here is a fact about the repo, not an accusation. most of these people did normal sdk work & i've said so where that's all there is.

two of those deserve a line each. jangid: his public github profile listed his company as "AlphaFi" & his bio as "Building #AlphaFi #Crypto #DeFi" (archived jul 12 2026↗) until the company field was cleared on oct 13 2026, two and a half weeks after the shutdown. on aug 11 2026 raj posted "we've been running an AI auditor on our own Move code for over a year... open sourcing it today" ↗ w a link to AlphaFiTech/sui-ai-commons↗ (born aug 9 in the org as AlphaFiTech/open-agents). that address now redirects to jangid/sui-ai-commons↗. the repo's own notes say it was transferred out of alphafi's org into his personal account on sep 24 2026, the day before the shutdown announcement (his commit recording the move, ba39b145, is stamped 18:01 utc, 29 hours before the tweet) & that alphafi's org rulesets didn't come w it. it's the only repo i can find that left the org around the wind-down. it's his repo now, so i archived it on oct 6 in case any of that changes: the notes↗, the commit↗ & the repo↗. sid & harshaalphafi: four commits on sep 22 & 23 2026 ("find the right position cap id for a user w multiple positions", "paginate receipts to find all user receipts", "add slush position code in util") teach the sdk to split a withdrawal across several position caps in one wallet. PR #103↗ (archived) merged them on sep 27, 34.5 hours after the tweet. never published to npm & none of the 7,050 slush withdrawals from sep 23 to 30 used more than one cap. so: wind-down code, written two days before anyone was told, shipped to nobody.

also in the org: alphafi-betterstack, a telegram /alert bot that opens a better stack incident & phones whoever's on call (created 2026-05-13). the cron runner & a "metarepo" are private.

what raj said in public

every post here was re-read by its post id through the fxtwitter api on 2026-10-05 & links to the original. quotes are trimmed, never reworded.

on oct 6 2026, the day after this page was built, raj made his x account private. the links below now show nothing unless you follow him. archive.org has copies of all but one: the IKA post, for one, was archived the second it went up↗.

• 2024-08-24: "How we think at @AlphaFiSUI"↗: "Simple > Complex, Users > Company, Product > Marketing...". noting it for later.
• 2025-05-30: on another protocol pinning a stablecoin at $1 (reply↗): they "should have had checks in the code to avoid oracle malfunction... That's how we do at AlphaLend". 22 days earlier alphalend had given ALPHA its first price by mapping it onto SUI's feed.
• 2025-11-13 & 12-05: "AlphaLend will be open sourced soon."↗, then "Open sourcing AlphaLend loading soon..."↗. ten months on, no alphalend source is public.
• 2025-12-06: on EVM chains (post↗): "someone can upgrade the logic, spoof you, and do whatever they want", followed by "Thank you @SuiNetwork". three days later, on sui, the lending admin upgraded the oracle's logic (v4, 10:11 utc) & re-pointed ALPHA's price entry that same afternoon.
• 2025-12-12: the IKA liquidation threshold, above. that afternoon (17:38 utc), replying about audits (reply↗): "all in all, audits are not high ROI." the only audits in alphafi's github are two "preliminary" pdfs from may 2025.
• 2025-12-14: at 23:01 utc (3am in dubai), when pyth-sponsored prices on sui stalled (post↗), he "just finished writing a cron job that updates Pyth prices for Sui assets". so he writes the price-feed plumbing himself.
• 2025-12-15: "$DEEP and $WAL markets on AlphaLend"↗, pitched w "Strict borrow limits so that you can withdraw anytime". twelve days earlier P4 had borrowed 47.5M DEEP & 6.2M WAL inside two hours against ALPHA booked at $14.16.
• 2026-06-27: "AlphaLend is built for the bad days"↗ (screenshot), listing "no socialized debt", "oracle sanity checks" & "conservative risk caps". at the time alphalend's oracle was pricing ALPHA off AVAX's feed, as it had since nov 25 2025 (bar a two-day gap in december).
• 2026-07-14: "AlphaLend continues to lead where it matters"↗ (screenshot), incl. "Flow limits cap the blast radius at $250K per market". true for USDC that day, then raised 20x & switched off around the fee wallet's sep 6 borrows (above). the bad debt the foundation covered ten weeks later was $17.1M.
• 2026-07-22 & 08-11: "$USDSUI just started paying us!"↗ & hashi BTC on alphafi "from day one of mainnet"↗. sui's own hashi launch post↗ quotes him by name as a partner.
• 2026-09-03: "Looking forward to Basecamp"↗. alphafi was announced as a diamond sponsor of sui basecamp 2026 around then, w raj on the main stage. sui.io/basecamp no longer lists alphafi.
• 2026-09-26 11:59, 13 hours after the shutdown tweet, a reply under alphafi's thread: "wind-down update"↗: "$50M+ withdrawn so far", "Only $5.37M in loans left to repay", "For help repaying or withdrawing, reach out on Discord." it doesn't mention ALPHA, the oracle, the bad debt, or whose money paid it. that's the last post i can find from him.