Notice: Editorial analysis that does not constitute financial advice or a recommendation regarding any asset. All figures in the series in this article come from a single extraction frozen on Oct-27-2026: TVL series by chain for Arc and Starknet, breakdown by protocol and by token for the 126 tenants of both chains, volume of Arc decentralized markets, and prices at 00:00 UTC, all from DefiLlama, which is an aggregator and is marked as such. The migration mechanics come from the official Extended notice dated Oct-8. The decompositions between price and flow, crossover percentages, and ratios are our own calculations based on those figures and are labeled in the text. Press figures are cited with their source and date. CleanSky does not receive commissions or referral payments from any mentioned protocol.
What crossed from Starknet to Arc when Extended Perps, a perpetual futures market, moved its settlement on Oct-22-2026 at 13:00 UTC was 110,125,286.90 USDC, 28.26589 cirBTC, and 680.65106 WETH. The day before, on the other side, there were 109,327,204.84 USDC, 872,629.36 USDT, 28.35921 WBTC, and 680.60498 ETH. Stated this way, token by token, the transfer crossed 99.9324% of the dollars, 99.6709% of the bitcoin, and 100.0068% of the ether. In aggregator dollars, the same operation represents $114,240,586 leaving Starknet and 113,986,347 entering Arc—a gap of $254,239, or 0.2225%—a chain that loses 73.3106% of its TVL (the value deposited in it) and another that gains $104,542,202, which is 9.44 million less than what it received. The three accounts disagree for the same reason: chain TVL mixes units with price, and 72.33% of Arc's TVL on Oct-22 was tokenized bitcoin on the day bitcoin fell by 3.28%.
What units crossed from Starknet to Arc in the Extended migration?
Extended Perps settled on Starknet until Oct-22 and has since settled on Arc, the Circle chain. DefiLlama, the aggregator from which all series in this article originate, publishes two parallel series for each protocol and each chain: the dollar value and the inventory in units of each token. The latter allows for verifying if a move was complete, because one bitcoin remains one bitcoin even if its price changes. The comparable points are 00:00 UTC on Oct-22 in Starknet and 00:00 UTC on Oct-23 in Arc, and the crossover percentages are our own calculation: arrival units divided by departure units, with USDT grouped with USDC.
| Magnitude | Starknet, Oct-22-2026 00:00 UTC | Arc, Oct-23-2026 00:00 UTC | Variation | Crossover |
|---|---|---|---|---|
| Dollars (USDC + USDT → USDC) | 110,199,834.20 | 110,125,286.90 | −74,547.30 | 99.9324% |
| Bitcoin (WBTC → cirBTC) | 28.35921 | 28.26589 | −0.09332 | 99.6709% |
| Ether (ETH → WETH) | 680.60498 | 680.65106 | +0.04608 | 100.0068% |
| Dollar Value (DefiLlama) | 114,240,586 (out) | 113,986,347 (in) | −254,239 | 99.7775% |
The last row includes both operands because the series do not drop to exactly zero: the protocol goes from $114,240,713 to $127 in Starknet and from $63 to $113,986,410 in Arc. Ether crosses above 100% by 0.04608 ETH, about $110: noise between two captures.
Why don't Extended's USDT and WBTC appear on Arc?
Because Arc does not support them. The official notice on Oct-8 established the mechanics, according to the version we read on Oct-25 (the page was rewritten in the past tense on the 26th and no longer includes the times): USDT and WBTC deposits were closed that same day, the deadline to convert remaining balances expired at 13:00 UTC on Oct-21, the conversion was 1-to-1 with no fee, USDC traveled via CCTP—Circle's protocol that burns on the source chain and mints on the destination chain—WBTC was converted to cirBTC, and ETH to WETH. Extended added a 0.50% premium on the converted amount, paid within the following eight hours.
That premium is the first candidate to explain the $254,239 gap, and it does not explain it. At 00:00 UTC on Oct-22, 872,629.36 USDT and 28.35921 WBTC remained to be converted; 0.50% of both amounts to 4,363.15 USDC and 0.14180 cirBTC, approximately $15,871 by our calculation at the implicit price of the Arc snapshot: 6% of the gap, and in the opposite direction, as the premium is a credit.
The gap is distributed differently, and measured token by token, it is almost entirely split in two. In units, 74,547.30 dollars and 0.09332 bitcoin are missing, minus the $110 of surplus ether: $81,974 of inventory. The other $172,328 is price, because the two snapshots in the series are one day apart, and between them, the implicit price of bitcoin fell from $83,992.52 to $81,158.97 and ether from $2,474.93 to $2,382.33. The residue is $64: the gap is not lost money, it is $82,000 that did not cross plus one day's price movement. To put the $254,239 into perspective: in the six previous days, the protocol's daily variation in Starknet was always downward, between $134,204 and $581,639.
What did Starknet lose when Extended left, and what stayed?
Starknet's chain TVL went from $157,848,175 to $42,128,739 between Oct-22 and Oct-23: $115,719,436 less, or 73.3106%. Of that drop, $114,240,586 is the protocol that moved, 98.722% of the total. The tenant that left was worth 72.3738% of the chain the day before it departed.
A residue of $1,478,850 remains in Starknet between Oct-22 and Oct-23-2026. Breaking it down requires a token-by-token analysis: the price effect is the initial units multiplied by the variation in implicit price, and the flow effect is the variation in units multiplied by the final price. Summing the tenants that the aggregator breaks down, price explains $1,740,307 of the drop and flow explains $393,321—more than the chain lost, because some tenants rose that day and the sum offsets them.
| Starknet, Oct-22 → Oct-23-2026 (DefiLlama) | Oct-22 | Oct-23 | Variation | Price | Flow |
|---|---|---|---|---|---|
| Chain TVL | 157,848,175 | 42,128,739 | −115,719,436 | — | — |
| Extended Perps | 114,240,713 | 127 | −114,240,586 | — | transfer |
| Ekubo | 26,286,703 | 25,736,363 | −550,340 | −531,160 | −19,181 |
| Endur | 12,419,490 | 11,702,549 | −716,941 | −770,667 | +53,726 |
| Vesu | 11,327,146 | 10,895,056 | −432,090 | −398,985 | −33,105 |
| Extended Spot | 2,745,564 | 2,312,221 | −433,343 | −38,582 | −394,761 |
| Sum of breakdown items (includes Haiko, −913 price) | — | — | — | −1,740,307 | −393,321 |
That table answers whether the move dragged anyone else down. In Ekubo (the chain's decentralized market), Vesu (lending), and Endur (STRK liquid staking), almost everything is price: Starknet's native token fell 7.17% that day ($0.057344 → $0.053234) and ether fell 3.49%. Of the $393,321 in negative flow, 394,761 is Extended Spot, the same operator's spot market, with withdrawals during the migration; among all other tenants, $52,286 remains.
Why did Arc's TVL rise by 104.54 million and not Extended's 113.99?
Arc's chain TVL went from $530,164,328 to $634,706,530 at the same two points: $104,542,202 more. Since the arriving protocol contributed $113,986,347, the rest of the chain decreased by $9,444,145 that day. The question of whether those 9.44 million left Arc has a measurable answer: they did not leave, they are price.
| Arc, Oct-22 → Oct-23-2026 (DefiLlama) | Oct-22 | Oct-23 | Variation | Price | Flow |
|---|---|---|---|---|---|
| Chain TVL | 530,164,328 | 634,706,530 | +104,542,202 | — | — |
| Extended Perps | 63 | 113,986,410 | +113,986,347 | — | transfer |
| Morpho Blue | 304,467,013 | 301,997,147 | −2,469,866 | −7,953,023 | +5,483,157 |
| Aave V4 | 184,289,910 | 177,865,504 | −6,424,406 | −3,556,472 | −2,867,934 |
| Circle Gateway | 6,209,637 | 9,722,042 | +3,512,405 | −1,779 | +3,514,185 |
| Uniswap V3 | 25,824,885 | 22,248,079 | −3,576,806 | no breakdown | no breakdown |
| Uniswap V4 | 4,544,786 | 4,347,903 | −196,883 | no breakdown | no breakdown |
| Sum of breakdown items | — | — | — | −11,511,275 | +6,129,408 |
The price effect of Arc's breakdown tenants between Oct-22 and Oct-23-2026, an $11,511,275 drop, alone exceeds the $9,444,145 total, and the flow is positive: $6,129,408, with Morpho Blue adding $5,483,157 in USDC and Circle Gateway, Circle's USDC bridge, adding $3,514,185. While the chain's value was decreasing, dollars were flowing in.
One gap remains open in that table. The aggregator does not publish a token breakdown for the two Uniswap instances on Arc, so the $3,576,806 lost by Uniswap V3 remains unallocated between price and flow: 38% of the drop. That same day, Uniswap V3 did $47,278,491 in volume on Arc, 8.4 times the $5,620,767 from the previous day, with the entire chain multiplying its volume by 5.7. A protocol that loses 13.9% of its TVL on the day of its peak activity allows for several interpretations, none of which are measured here.
Why does the TVL of Arc, a stablecoin chain, move with bitcoin?
Arc is the Circle chain and its gas is paid in USDC, but its TVL is mostly tokenized bitcoin: cirBTC, Circle's wrapped bitcoin, which has been collateral for borrowing against Circle Mint assets since Sep-21-2026, with Morpho as the first approved protocol. On Oct-22, there was $383,481,820 of cirBTC among the tenants broken down by token.
That figure allows for two denominators, and it is important not to confuse them. Relative to the chain TVL of that day, $530,164,328, cirBTC is 72.33%; relative to the subtotal of tenants that breakdown tokens, 499.75 million, it is 76.73%. The first is the one comparable to TVL headlines.
| Date (00:00 UTC) | Breakdown cirBTC | Arc Chain TVL (DefiLlama) | Share (lower bound) |
|---|---|---|---|
| Oct-22-2026, before transfer | 383,481,820 | 530,164,328 | 72.33% |
| Oct-23-2026, after | 376,467,301 | 634,706,530 | 59.31% |
| Oct-25-2026 | 373,921,659 | 628,869,327 | 59.46% |
| Oct-27-2026 | 370,915,377 | 618,460,158 | 59.97% |
The share is a lower bound for the same reason: any cirBTC that might be in the two Uniswap instances is not counted. With this composition, the day's bitcoin drop subtracted $11,511,275 in price effect from the breakdown tenants without anyone withdrawing anything: this is our own calculation using the implicit prices of each snapshot, not 3.28% applied to cirBTC. The arrival of Extended diluted the share by thirteen percentage points at once, from 72.33% to 59.31% of the chain TVL; since then, it has been rising again because the denominator is decreasing.
Did Aave V4 lose deposits on Arc on the day of the move?
Aave V4 appears with $2,867,934 in negative flow, and its USDC inventory drops from 36,994,814.46 to 32,490,913.87 units: 4,503,900.59 fewer USDC between Oct-22 and Oct-23-2026. Read as deposits, this looks like a withdrawal coinciding with the move, but the reason it is not lies in what DefiLlama measures for a lending protocol.
The TVL it publishes for Aave V4 is available liquidity: deposits minus outstanding credit. Thus, if a borrower takes a million, liquidity drops by a million without the deposit moving. The on-chain reading of the hub shows the Main Spoke deposit at 143,178,544 USDC on Oct-25, 2,061 USDC more than on the 17th: flat. The 4.5 million that disappear from liquidity between the 22nd and 23rd are credit that someone drew down. The same inventory shows the other side: deposited cirBTC rose by 22.06 units (1,121.32271 → 1,143.38708). This is the reading error we already documented when measuring the USDC concentration of that hub.
Why does DefiLlama say Extended fell 98% in Starknet and kept 2.3 million?
Because in DefiLlama there are two entries with the same name: a parent protocol, extended, which sums perpetuals and the spot market, and the child, extended-perps, which is the one that moved. The parent in Starknet went from $116,986,277 to $2,312,348 between Oct-22 and Oct-23, a 98.02% decrease; the child went from $114,240,713 to $127. The 2.3 million that seem to have stayed are Extended Spot, a different protocol that continues to settle on Starknet: $2,312,221 on Oct-23, which with the $127 residue from the perpetual gives exactly the parent's $2,312,348.
This bifurcation explains two figures that circulated. It is the parent, and not the market that moved, that describes the "98% down to 2.3 million dollars" in the coverage from Oct-23 and Oct-24. And no point in the DefiLlama series between Oct-19 and Oct-27-2026 shows 120 million dollars for Extended: the perpetual was worth $114,240,713 in Starknet and $114,395,780 summing its three chains on Oct-22; the parent, with spot, was $117,141,344. The weekly peak was on Oct-19, $115,428,481 for the perpetual and $118,170,988 for the parent. The 120 million figure traveled in three pieces—Crypto Briefing on Oct-22, the TokenPost headline on Oct-23, and AMBCrypto on Oct-24—and none of the sources cited in this article attribute them in the text to a specific source. The "approximately $117.7 million" from the previous Crypto Briefing piece on Oct-21 is in the range of the parent.
The rest of the coverage was accurate. The 73% drop in Starknet is exact when rounded (73.3106%), the 158 and 42 million are the correct extremes, the attributed cause—migration—is correct, and Crypto Briefing described the mechanics without errors: start time, unsupported tokens, deadline, and 0.50% premium. What was missing was verifying the amount and looking at the breakdown. Two discrepancies are noted. The ChainCatcher note picked up by KuCoin on Oct-9 set the conversion deadline at 12:00 UTC on Oct-21, compared to 13:00 in the official notice, which prevails. And the Starknet governance poll on investigating a move to its own L1—non-binding—opened on Oct-15 at 09:32 UTC and closed on the 22nd at the same time with 93.52% of the voting power: three and a half hours before the migration, which Extended had announced on Oct-8. These are dates, and no subtraction in this article depends on them.
Is the money Extended transferred staying in Arc?
The transfer was complete; retention remains an open question. Extended Perps in Arc went from $113,986,410 on Oct-23 to $110,644,944 on Oct-27 at 00:00 UTC: $3,341,466 less, a 2.93% drop in four days. Using the same method, only $95,250 of that decrease is price and $3,246,216 is flow. USDC falls from 110,125,286.90 to 106,760,723.80 units, while cirBTC rises from 28.26589 to 28.52380 and WETH from 680.65106 to 721.82269: dollars are leaving and collateral is entering.
The average is $811,554 in outflows per day, 3.2 times the gap left by the entire move. The live reading on Oct-27 distributes the protocol as $111,018,147 in Arc, 154,973 in Ethereum, and 127.74 in Starknet: the balance is dropping without appearing on another chain.
What remains unresolved from the Extended transfer to Arc and on what dates will it be checked?
Three gaps are declared. The token breakdown for the two Uniswap instances on Arc does not exist in the aggregator: $3,773,689 of the D-day variation remains unallocated between price and flow. The perimeter of categories the aggregator includes in Arc's chain TVL leaves a difference of $39,568 on Oct-22, 39,093 on Oct-23, and 38,083 on Oct-25—0.007% of 530 million—but on the 27th it jumps to $9,256,711 without knowing which tenant it comes from. Only the 59.97% share and the Oct-27 Extended reading depend on this point, and they pass their thresholds comfortably; no subtraction from the 22nd to the 23rd touches it. And the vault guarding Extended's collateral in Arc remains unlocated: the official explorer responds behind Cloudflare and the two public RPCs limit the log query window, so the D-day scan only identified a pass-through contract, with two mints of 9,997,000 USDC via CCTP in the 13:00 to 16:00 UTC window that the RPC allowed to be read.
The three conditions that would invalidate this reading were set on Oct-25-2026 and measured again on the 27th.
| Condition written on Oct-25-2026 | Condition that invalidates it | Measured on Oct-27 | Result |
|---|---|---|---|
| cirBTC share in Arc chain TVL | below 50% | 59.97% | maintained |
| Extended Perps remains on Arc | below 100 million dollars without equivalent multi-chain drop | $110,644,944 | maintained |
| Retroactive series review | Starknet drop below 70% or Arc entry below 110 million dollars | 73.3106% and $113,986,347 | maintained |
The third test exists due to our own precedent: when measuring Arc's 30th day, we published a TVL of $538,351,560 for the 16th on Oct-19, and the current series shows $535,571,580 for that date, $2,779,980 less (0.52%). An aggregator revises backward, and then subtractions are recalculated. The three control dates:
- Nov-8-2026 — re-read the daily series at 00:00 UTC with the three thresholds intact: cirBTC below 50% of Arc chain TVL, Extended Perps on Arc below 100 million dollars without a multi-chain drop, or a retroactive revision that leaves the Starknet drop below 70% or the Arc entry below 110 million.
- Nov-8-2026 — new threshold, for the four subsequent days: if the flow outflow of Extended Perps on Arc exceeds 10 million dollars accumulated since Oct-23 (currently 3.25 million), "moved" ceases to describe what happened and the phrase becomes "moved and is leaving."
- Dec-1-2026 — if Extended Spot stops appearing on Starknet, the aggregator's parent and child converge and the reading confusion explained in this article resolves itself. It is advisable to measure it before it happens.
The same operation yields four results depending on whether it is read in units, protocol value, or TVL of each chain, and all four are correct: reading a chain whose collateral is bitcoin in dollars is measuring the price of bitcoin.
Related articles: The Aave V4 hub on Arc and why its TVL is liquidity and not deposits. Arc on day 30: zero swaps in the cohort and credit against syrupUSDC. Arc vs. Tempo: the two stablecoin chains from Circle and Stripe. Appchain, Solana, or Ethereum L2: where perpetuals settle and why. What are decentralized perpetuals and how is their liquidity measured.