ccip.dev methodology
What is counted
-
Message: one mainnet CCIP message from
api.ccip.chain.link/v2(environment=mainnet). Testnets are excluded. -
Day: the UTC date of the message’s send time.
-
Value transferred: Σ over a message’s tokens of
amount / 10^decimals × USD price.- Live data uses the latest DefiLlama price when the message is first valued. A live message that is still unpriced is valued again when its day is finalized (00:07 UTC, and again at 06:00 UTC, the next day).
- History uses DefiLlama’s daily price for the send day.
- Data-only messages add $0.
-
Price lookup: live data and history use the same order.
- DefiLlama’s price for the token’s own chain and address.
- The same token on another CCIP chain, from the same CCIP token group. Siblings on Ethereum, Base, Arbitrum, Optimism, Polygon, BSC, Avalanche and Solana are preferred, and the token’s own decimals are used.
- DefiLlama’s
coingecko:<id>price. The CoinGecko id comes from CoinGecko’s free coin list (platform plus contract address). CoinGecko is used only to map ids; no CoinGecko price is stored. - Otherwise the token is unpriced.
-
Price guards:
- A single token transfer valued above $10 billion is treated as a bad price and counted as unpriced.
- History only: a missing daily price is filled from the nearest day within ±2 days. The fill happens only when the price series has points on both sides of the day; when two days are equally near, the earlier one is used.
- History only: a daily price more than 20× or less than 1/20 of the median of the token’s other prices within ±7 days (or of its nearest 3 prices) is dropped as a glitch. The final history build dropped 107 such prices. For example, DefiLlama lists elizaOS at $125,176 on its launch day, 2025-11-07. Without the guard that day totals $18.8 trillion; with it, $74.7 million.
-
Unpriced: a token none of the steps can price adds $0. The message is still counted, and it is reported in
unpriced_messages. Since 2023-07-06, 68,893 of 1,565,729 messages (4.4%) are unpriced. -
Fees: the message’s fixed fee, in USD. Fees come from each message’s detail. Live messages have them from 2026-10-05, valued at the fee token’s latest price; earlier days get them from the fee backfill, which values each fee at its send day’s price. The site shows fees from the first day that has them. A live fee that couldn’t be priced when its detail was fetched is priced again, at the latest price, when its day is finalized.
- Some fee tokens have no market price of their own. When such a token is a chain’s wrapped gas token (WBTC on
Bitlayer, WTAO on Bittensor EVM) or the same token as a priced one (LINK on BNB Chain and Polygon), its fee is
valued at the coin it wraps or matches, through DefiLlama’s
coingecko:<id>price, at the token’s own decimals. Each such token’s decimals were read on chain. - Canton’s fees are valued as CC (Canton Coin), its only CCIP fee token with a known identity, at CC’s 10 decimals.
- For a coin DefiLlama has no history for, the fee backfill uses CoinGecko’s free public daily history (at most 365 days back; credit: CoinGecko). Today that applies to Mova’s MOVA, whose price data ends 2026-09-07 when it was delisted; later Mova fees stay unpriced. The live Worker uses DefiLlama only.
- A fee of 0 and a fee paid in one of four test “LINK” tokens (total supply 100, not LINK; two on Ethereum and two on Morph) count as $0.
- A fee paid before its token first traded is valued at the token’s first market price: Pharos, Monad, 0G and Plasma. Sonic uses FTM’s price on that day instead, since FTM converted 1:1 into S. Fees after a token’s first price are never filled this way.
- BTC-backed gas tokens (WBTCN on Corn, PBTC on Botanix) are valued as BTC.
- Fee tokens of chains that have left the CCIP docs (such as Mind and Everclear) were identified from the docs’ git history, and their decimals read on chain where the chain still answers.
- A few fees still have no free price: Mova fees after 2026-09-07 and about 8 fees whose token can’t be identified (one on Tron, Base zunETH and one rarely used Canton token). They are left out of the fee totals.
- Some fee tokens have no market price of their own. When such a token is a chain’s wrapped gas token (WBTC on
Bitlayer, WTAO on Bittensor EVM) or the same token as a priced one (LINK on BNB Chain and Polygon), its fee is
valued at the coin it wraps or matches, through DefiLlama’s
-
Fees paid in LINK: the part of a day’s fees whose fee token is LINK. That is LINK in CCIP’s LINK token group (LINK on the chains where it moves through CCIP), plus LINK on the 14 chains where LINK is a fee token but is not in that group: OP, BNB Chain, Gnosis, Polygon, Arbitrum, Avalanche, Cronos zkEVM, Bittensor EVM, HyperEVM, Kaia, Hedera, Etherlink, Nexon Henesys and Memento. Available for every day that has fees.
-
Fee groups: each fee counts in one of four groups, by its fee token.
- LINK: the LINK tokens above.
- Gas tokens: a chain’s gas token or its wrapped form, such as WETH, WBNB, WPOL, WAVAX, wSOL, WHYPE, APT, CC and GRAM.
- Stablecoins: GHO, pathUSD and other USD stablecoins, plus gas tokens that are stablecoins: USDT0 and gUSDT on Stable, xDAI on Gnosis, GHO on Lens and USDC on Arc.
- Other: every other fee token, such as the test tokens above and the few tokens that can’t be identified. Other is a day’s fees minus the three groups above.
A token’s group comes from the CCIP docs: each chain’s listed fee tokens, grouped by symbol (a fixed list of stablecoin symbols; every other symbol is a gas token). Fee tokens the API reports that the docs list under another address, or not at all, were checked on chain and added by hand. A token in the wrong group would move USD between groups, but never changes total fees.
-
LINK paid: each LINK fee’s amount,
amount / 10^decimals, counted whether or not the fee has a USD price. The decimals come from the CCIP token registry, are 18 for Ethereum LINK, and were read on chain for the chains above, except Memento’s, which comes from the CCIP docs. -
All-time fees (home page): every day in history plus today so far, from the first day with fees. Until fee data reaches back to 2023-07-06, the line reads “Fees since
” instead of “All-time fees”. -
Run-rate: the fees of the last 30 complete days × 365 / 30. It is shown once all 30 of those days have fees.
-
Fee rankings (Top pages, “Fees”): lanes, senders and source chains ranked by the fees their messages paid in the window, largest first, with value moved breaking ties. One with no fee data in the window is left out. Tokens have no fee ranking: a fee is paid per message and is not split across a message’s tokens. On a fee ranking, Share is of the fees of the top 100.
-
Typical fee (Flow and Top lanes): the median of a route’s priced fees over the last 30 complete UTC days, with the range from the 10th to the 90th percentile. Percentiles use the nearest rank: of a route’s n fees in ascending order, the p-th percentile is the fee at rank ⌈p × n / 100⌉. A route needs at least 5 priced fees in the window. When at least 20 of them were paid in LINK, the LINK fees and the other fees (gas tokens, plus the few paid in stablecoins or other tokens) each get their own median. The figures come from
cost.json, rebuilt at each finalize run (00:07 and 06:00 UTC). -
Take rate (day pages and History): a day’s fees as basis points of the value its messages moved,
fee_usd / usd_value × 10,000; one basis point is 0.01%. A day without fee data, without fees or without value moved has none. It is not a profit margin: it compares what senders paid in fees with the value their messages carried. -
Fee per message (day pages): a day’s fees divided by its messages, data-only messages included. Messages whose fee has no price still count in the divisor, so fee per message reads low on routes like Ronin.
-
Delivery time: receipt minus send, for
SUCCESSmessages. The median is reported per day.
Senders and projects
- Sender identity: a message is credited, in this order, to the project whose verified label in
labels/projects/names its source chain and sender address (“sent by”); to the project whose label names the same EVM address on another EVM chain (“inferred”, see below); to the project that claims its first token, when its sender is the wallet that signed it (“via token”, see below); and otherwise to the sender address itself, across every chain it sends from. On 2026-10-13, xStocks’s xBridge (one address on five EVM chains) and its Solana sender add up to 319 of the day’s 1,569 messages (20.3%). - Verified labels: a label is verified when the project’s own pages (its site, docs, GitHub, or app configs and JS bundles) publish the address, or by on-chain control: the address’s CCIP sends arrive through a Chainlink KeystoneForwarder report whose header names a CRE workflow, and the public CRE Workflow Registry on Ethereum (
0x4Ac54353FA4Fa961AfcC5ec4B118596d3305E7e5) records that workflow’s owner as an address the project’s own pages publish. Such a label’s source comment cites both the project page that names the owner and the workflow (name, workflowId, owner, registry). Anything else, such as an explorer name, a workflow name or a shared deployer, stays a lead withverified = false. On 2026-11-06, Re’s four NAV broadcasters were verified this way: Re’s docs name its Safe, and the Safe owns the workflowcre-nav-ccip-prod. - Label spread: an EVM address with no label on its chain takes the label the same address has on another EVM chain, marked “inferred”. A label never spreads across chain families, never from or to Solana or another non-EVM chain, never to an address two projects label, and never for a project whose label file sets
spread = false. The weekly label candidates pull request lists every address a label spread to, so each is confirmed or blocked. - Via token: when a wallet calls the CCIP router itself (
sender = origin), the sender is only an address, but the project behind the message is the token’s issuer. Such a message is credited to the verified project that claims its first token (message_tokens.idx = 0) with a[[tokens]]entry: either the exact token on its chain, or any token in the same CCIP registry group. A token or group two projects claim is credited to neither. It counts only when no label names the sender on any chain, so a labeled wallet keeps its label and a message an app contract sent is never credited this way. The project’s totals are its sent-by messages plus its via-token ones, and the address loses what moved; an address left with nothing drops out of the rankings. A claim stands on the project’s own pages, which must show the token is on CCIP and give its address (the project’s own site, including its app configs and JS bundles, counts). General-purpose assets that no single app’s users bridge (USDC, USDT, LINK, cbBTC, WETH and the like) are not claimed, so their direct sends stay with the address. Every day since 2023-07-06 is covered: the list API gives every message an origin. - Rankings: Top › Senders, the daily and weekly top sender and today’s top senders rank identities, not chain-and-address pairs. The daily and weekly top sender is the identity with the most messages; Top › Senders ranks by value or by fees, as its other tabs do. A top sender’s share is of all the period’s messages. When half or more of the top sender’s messages came via token, the daily and weekly copy says “via its token”.
- Unique senders still counts chain-and-address pairs: the same address on two chains counts twice, and an identity that sends from several addresses counts once per pair.
- Data messages: messages that carry no tokens. Every day in history counts them; the daily card and History show their share.
- Label coverage: the share of the last 7 days’ messages, value and fees credited to a project (sent by, inferred or via token), computed at each finalize and shown on the status page with its via-token part: “Labeled: 74% of messages (29% sent by a labeled project, 45% via its token), 61% of value (last 7 days)”.
- Project pages (
/projects/) list every verified project: its totals over 7 days, 30 days and all time, its last 30 days, its chains and top lanes, its addresses (inferred ones marked), the tokens it claims, how its 7- and 30-day messages split between its own sends and via its token (its lanes and data split count its own sends only) and its label’s source. Its totals and chart include via-token messages, and its chains, including those it was credited on via its token, read “Seen on”. The split reads “N sent by its contracts · M via its token, by wallets bridging it”. - Public files: from 2026-10-14 the files carry sender identities in additive fields, so
schema_versionstays 1 and older readers keep working.top/sender.jsonaddsidentitiesandidentities_by_fees; itswindowsandby_feeskeep the older<selector>:<address>keys.day-leaders.jsonandtoday.jsonname sendersproject:<slug>oraddress:<address>, withchains,addresses(the number of addresses),inferred,x,kindandshare_pct.history.jsonandtoday.jsonadddata_messages;status.jsonaddslabel_coverage(messages_pct,usd_pct,fees_pct,window);moments.jsonaddsarrivals.sendersand each week’sproject.projects.jsonis new.- Every sender identity also carries
via_token_messages,via_token_usd,via_token_fee_usdandvia_token(true when every message came via token);label_coverageaddsvia_token_messages_pct,via_token_usd_pctandvia_token_fees_pct; aprojects.jsonwindow addsvia_token_messagesandvia_token_usd, and a project lists itstokens. A via-token-only identity publishesxas null, so the posts never tag it. On the site, a sender credited this way is marked “via token” when all its messages came via its token and “incl. N via token” when some did, with the title “Credited via its token: wallets bridging the project’s token through CCIP themselves”.
CRE-sent messages
The Chainlink Runtime Environment (CRE) runs workflows on Chainlink nodes. A workflow reaches a chain by having a node submit a signed report to that chain’s KeystoneForwarder; the forwarder calls the workflow’s contract, which can send a CCIP message. ccip.dev counts a CCIP message as sent by a CRE workflow when the signer of its source transaction (its origin) is a known CRE node key on the source chain.
- The keys are the signers of KeystoneForwarder
report()transactions, read from each chain’sReportProcessedlogs (config/cre-keys.json, refreshed daily). Only production forwarders count, never the simulation forwarders. - Checked: on 2026-11-05, every one of 20 sampled tagged messages had its source transaction sent to the chain’s forwarder, and none of 10 untagged messages from the same senders did.
- Matching on the CCIP sender does not work: the contract that receives a CRE report usually hands off to another contract that sends, so the sender is not a forwarder receiver.
- A lower bound. It misses a workflow whose contract queues a request and sends CCIP in a later transaction, and a key rotation until the next key refresh is merged; the last 30 days are re-tagged after each key change.
- Sampled keys. Keys come from KeystoneForwarder reports sampled in one window a week (5,000 blocks, up to 5 reports) from 2025-10-17 on, and from the last day’s reports daily since 2026-11-05. A key that signed only outside the sampled windows is missed, so counts stay a lower bound. Where no free RPC serves old logs, a chain’s history has no weekly samples: on 2026-11-05 that was Jovay and Monad, and BNB Chain had 1 of 55 weeks.
- History is tagged from 2025-10-17, the CRE Workflow Registry’s first record.
- In the replay, each lane-day that carried CRE-sent messages gets its own small sample of teal comets on top of the usual ones. CRE is under 1% of messages, so a share to scale would almost never show; the teal comets mark where and when CRE workflows sent CCIP, not how much.
Broadcasts
- What is posted: a daily report after each day is finalized, a weekly report on Mondays (Monday–Sunday, UTC), a post for each record or milestone a newly finalized day sets, and one for each Reserve deposit. They go to the Telegram channel @ccipdev and to X.
- Same numbers as the site: posts and their cards use the site’s data and number formats. A daily change is against the mean of the 7 days before; a weekly change is against the week before.
- Records follow the Records page’s rules: a tie keeps the earlier day, and the LINK share counts days with 100 or more messages. Milestones are the all-time message, value and fee thresholds the Records page lists.
- Tags: an X post tags only the player it is about, and only with an X handle its own website or docs link to; everyone else, and any player without a verified handle, is named. The daily and weekly top sender is never tagged. Reserve deposits tag @chainlink.
Moments
Besides the daily and weekly reports, @ccipdev posts moments: short posts about one chain, token or week. Each is computed from the same data as the site, after each day is finalized. They go to the Telegram channel and to X; the poll goes to Telegram only.
- A new chain is welcomed the day after its first CCIP message, on either side of a lane. Its number is how many chains had sent or received a message by that day; chains that arrive on the same day are numbered in the order of their CCIP chain selectors, compared as text, so no two share a number. Its first partners are the chains on its lanes that day, by value.
- A countdown names the next all-time milestone (the messages, value and fee thresholds of the Records page) when it is 2 to 7 days away at the pace of the last 7 days. Under 2 days the milestone post itself is about to come.
- Movers compare last week’s top 10 (Monday to Sunday) with the week before’s, on three boards: chains by fees paid, tokens by value and lanes by value. A new #1 leads (fees first, then tokens, then lanes); otherwise the biggest climb does. Climbers gained 3 places or more, or entered the top 10.
- The spotlight is the chain whose messages (sent and received) grew the most against the week before, among chains with at least 100 messages last week and 50 the week before, and whose first message came no later than the first day of the week before, so a partial first week cannot inflate its rise. A chain is not spotlighted again within 4 weeks.
- New on CCIP lists the lanes and tokens whose first message came Monday to Friday. A token is counted by its registry group, so a token’s new chain is not a new token. A lane of a newly welcomed chain is left out.
- A new sender is posted the day after it enters the day’s top 10 senders by messages, with 20 or more that day, when its first message on any chain came within the 30 days ending that day. Each sender is posted once. Its count is of its messages since its first day.
- The project spotlight is the labeled project that gained the most messages on the week before, with at least 50 that week; it is posted on the Wednesday after the week. The spotlight card shows a project with no messages the week before as “none the week before”, with no claim that it is its first week; the post leaves out the growth figure.
- The poll asks, in turn by week, which chain will pay the most fees, whether fees will beat last week’s total (by more than 10%, within 10%, or less), or which lane will move the most value. The answer is taken from our data once the week is complete.
- Tags: an X post tags only what it is about (the welcomed chain, the spotlighted chain, the headline chain of the movers when the headline is a chain, the new sender or the spotlighted project), and only with a verified handle. A project credited only through label spread, and any address, is never tagged. Everyone else is named in plain text, and the daily and weekly reports tag no one.
Monthly video
On the 1st of each month @ccipdev posts a 30-second video of the month before: the replay, cut to that month. It shows only the month’s days and the chains and lanes that moved in it. Its value counter, its fees counter and its leaderboard count the month alone, from zero. A chain whose first CCIP message came that month gets its “joins” beat, and an all-time milestone or record crossed that month gets its beat. The end card gives the month’s messages, value moved, fees paid and the share of fees paid in LINK, the all-time records set that month (by the Records page’s rules) and the chains that joined (three of each, then “+N more”), and the month’s busiest lane by value. The value moved is rounded down there and in the post, as the counter shows it. The post’s caption is written from the same month summary the end card shows, recorded with the video.
LinkMarine bot
The LinkMarine Discord app’s intel cards read the same public files as this site, so their numbers match its pages.
- SITREP (
/cciptoday): today’s totals fromtoday.json, as of itsupdated_at. “On pace for” divides today’s messages by the share of the UTC day gone, rounded to three figures; it compares with yesterday’s total fromhistory.json, and is left out before 01:00 UTC. The top lanes are today’s five biggest by value. - Chain dossier (
/ccipchain): all-time messages and value count both directions (messages from the chain and to it), fromreplay.json; the rank is by that value among every chain CCIP has seen. Fees are those paid on the chain as the source (top/src_chain.json), with its rank by the last 7 days’ fees when in the top 3. The week and the partners are the last 7 finalized days. - The vault (
/linkreserve):reserve.jsonas the Reserve page reads it. The weekly bars leave out the current week until it has a deposit. - Hall of records (
/cciprecords): the five all-time records by the Records page’s rules; the newest is the one set on the latest day. - Callsigns: three different tags from
docs/callsigns/tags.toml, each drawn with a chance proportional to the square root of its number of SFW memes in the chainlinkme.me archive. The rarity score is the sum of −ln(chance) over the three tags; the tier cut-offs and the percentile (“rarer than N in 100 Marines”) come from 200,000 simulated draws, so that about 60% of Marines are Common, 25% Uncommon, 12% Rare and 3% Legendary.
Coverage
History starts on 2023-07-06. This is the first CCIP mainnet message (2023-07-06 22:34:59 UTC). All-time figures are labeled “since 2023-07-06”.
How the history was collected:
- The API’s unfiltered message list cannot be crawled back to the start. The January 2026 depth wall was the Sui
poison message (2026-01-15 01:28:06 UTC). The unfiltered list’s 30-second timeouts also start somewhere between
September and December 2025. History is therefore crawled one source chain at a time, using the API’s
sourceChainSelectorfilter, which avoids both. - All 95 source chains were crawled to their first message, and none stopped early. The source list is the CCIP chain registry plus every chain seen in a message, so chains that only receive were checked as well.
- Three list pages fail with HTTP 500 whenever they contain one particular message. The crawler skipped past each such message and recorded the window after it. For each window, every lane was then queried to the window’s end: no message other than the skipped one is missing. The skipped messages are not counted.
| Source chain | Send time (UTC) | Message id | Window checked to |
|---|---|---|---|
| sui-mainnet | 2026-01-15 01:28:06 | 0xc8cc7f57a25e6e96e6be7bcd606a1a8d59d85d757d08e7e045e87a8cfe0bcbc7 |
2026-01-15 07:08:08 |
| aptos-mainnet | 2025-09-01 15:51:44 | 0xb3a18e5747c5083dc41b3e751482a2d0b30f62379f5d863df937b70edb175212 |
2025-09-02 02:30:29 |
| aptos-mainnet | 2025-09-01 14:43:49 | 0xb3a18e5747c5083dc41b3e751482a2d0b30f62379f5d863df937b70edb175212 |
2025-09-01 15:19:01 |
The two Aptos records share a message id but have different send times, so both are listed. The detail endpoint returns 404 for the Sui message.
Days with no messages (2023-08-03 to 2023-08-06 and 2023-08-31) have no row in history.json.
Live ingest started on 2026-10-05. For that day, the API lists exactly 2,487 mainnet messages, and their ids match the stored ids one for one.
Known limitations
- Before 2026-10-05, a message carrying more than one token counts only its first token, because the list endpoint exposes one token per message. Since 2026-10-05, none of 2,360 token-carrying messages has carried more than one token, so the effect is expected to be small.
- Live data and history use different price methods (latest versus daily), so their values can differ slightly.
- History’s group fallback (step 2) uses today’s CCIP token registry. A token that has since left the registry has no sibling fallback in history.
- Our cumulative value is 7.2% below Chainlink’s official figure (see the Chainlink metrics table below). The gap was already there by 2026-05, and recent months match within 0.7%, so it comes from older history. Its causes are not yet measured. Candidates are multi-token messages before 2026-10-05, unpriced long-tail tokens, and Chainlink’s own outlier handling.
- History statuses are as of crawl time, so a message still in flight when crawled keeps that status. Live messages update until final.
- Some tokens are not in CCIP’s public token list. They are priced and named from DefiLlama where it knows them; otherwise they show only their address.
- The CCIP API publishes no rate limits or terms. We keep to 1 request per second per job and credit the source in every file.
Data sources
All data sources are free:
- CCIP API (
api.ccip.chain.link/v2): messages, chains and the token registry. - DefiLlama coins API: latest and daily token prices, with no API key.
- CoinGecko public API: coin-id mapping, and the fee backfill’s daily price history for a coin DefiLlama has none for (MOVA), with no API key.
- Public Ethereum RPC endpoints (keyless, e.g. rpc.mevblocker.io and 0xrpc.io): the Reserve’s LINK balance and transfer logs.
- Chainlist (
chainlist.org/rpcs.json, from DefiLlama/chainlist): the source of the keyless RPC and Blockscout endpoint lists inconfig/endpoints.json. Only the data is used. Every entry is checked live before it is written.
Using this data
The numbers on ccip.dev, and the JSON files behind them at https://data.ccip.dev/v1/, are free to use with credit: “Source: ccip.dev”.
- Freshness: the files update every few minutes, so fetch each at most once a minute.
- Limits: requests to
/v1/are rate-limited per IP.
Chainlink Reserve
- Balance: read every hour with
balanceOfon the LINK token for the Reserve,0x9A709B7B69EA42D5eeb1ceBC48674C69E1569eC6. - Transfers: every LINK
Transferinto or out of the Reserve since its first transfer (block 23,039,541, 2025-07-31). They come from Ethereum logs through free public RPC endpoints, 12 blocks behind the chain head. Each run checks that transfers in minus transfers out equal the balance at the last block scanned, and alerts if they differ. - Price at transfer: DefiLlama’s LINK price at the transfer’s block time, within 10 minutes. This is the market price when the LINK arrived, not the price Chainlink paid for it.
- Cost basis: every inbound transfer at its own time price, minus every outbound transfer at its own time price.
Value now is the net LINK at DefiLlama’s current price. A transfer DefiLlama has not priced yet counts as $0 in the USD totals (cost basis and weekly USD) until it is priced, which is retried every hour;
cost_basis.unpriced_transferssays how many are waiting. - Deposits: inbound transfers of at least 1,000 LINK. Smaller inbound transfers (a 1-LINK test and gifts of up to 7 LINK) count toward the balance and the cost basis, but not toward pace, weekly deposits or performance.
- Weekly: UTC weeks starting Monday. The weekly USD is the deposit-time value of the LINK deposited, not Chainlink’s revenue.
- LINK demand: LINK paid in fees is the LINK amount of each day’s LINK fees (see Fees). The weekly fee mix, and the weekly fees beside the Reserve’s deposits, use the same Monday UTC weeks as the weekly deposits, and leave out any week that fee data covers only in part. Fees and deposits are shown side by side only: the Reserve does not publish which revenue each deposit came from.
- Pace: the 4-week figures average the last four complete weeks. The milestone date assumes that pace continues.
- Cadence: the average gap between deposits; the next deposit is expected at the last deposit plus the median gap, and is flagged overdue 24 hours after that; the streak counts consecutive deposits no more than 8 days apart.
Chainlink staking
- Events: every
Staked,Unstaked,UnbondingPeriodStarted,UnbondingPeriodReset,OperatorAddedandOperatorRemovedlog of the two staking v0.2 pools on Ethereum since the first one (block 18,621,067, the pools’ deployment on 2023-11-21; the first stake was 2023-11-28): the community pool0xBc10f2E862ED4502144c7d632a3459F49DFCDB5eand the operator pool0xa1d76A7cA72128541E9FCAcafBdA3a92EF94fDc5. They come from free public RPC endpoints, 12 blocks behind the chain head, every hour.staking.jsonand the stakers files are published only once the scan has reached the head and every staker is classified. - Pool state: each pool’s cap, staking limits, unbonding and claim periods and principal are read from the pools
every hour, and the three reward streams from the reward vault
0x996913C8c08472f584Ab8834E925b06D0eb1D813. - Rewards: the yearly rate is a stream’s LINK a second times a 365-day year over the pool’s principal. Operators
earn their base stream and the delegated stream, so their rate counts both. This is what the vault pays out now,
not a promise: the streams end when their funding runs out (
ends_at) and change when Chainlink adds rewards. - Holders: an address with no code is a wallet; one whose code starts
0xef0100is an EIP-7702 wallet (“smart”); a contract whoseowner()is one of stake.link’s two vault controllers (0xac12290b097f6893322f5430627e472131fbc1b5for community vaults,0x4852e48215a4785ee99b640caced5378cc39d2a4for operator vaults) is a stake.link vault; any other contract (mostly multisigs) is a contract. Holders are checked again every 30 days. - Stakers and at max: a staker holds more than 0 LINK; at max means within 0.01 LINK of the pool’s maximum stake.
- Exit queue: a staker is unbonding from its
UnbondingPeriodStarteduntil its claim window opens 28 days later, then claimable for the 7 days of the window. An unstake ends it, and so doesUnbondingPeriodReset(staking more while unbonding resets the clock). The queue counts each staker’s principal now. - Conversion: of the unbonding periods whose claim window has closed, the share that ended in an unstake inside the window (give or take an hour). stake.link’s vaults unbond on a rolling schedule and almost never leave, so they are counted apart from everyone else. Likely to leave is the LINK unbonding now outside stake.link times the others’ conversion. Lapsed counts claim windows that closed without an unstake.
- Space forecast: the same estimate day by day: each coming UTC day’s LINK whose claim window opens, outside stake.link, times the others’ conversion, plus the windows open now. Conversion is a share of windows, not of LINK, and a leaver can unstake on any day of its window, so the days are an estimate of where space comes from, not when.
- Openings and refills: once the community pool first filled, an opening starts when an event leaves it below its cap (within 0.001 LINK) and ends when it is full again. The gap is the most space open at once; the refill time runs from the event that opened the space to the one that filled it. A grab is a stake made while the pool was open. Hunter of the week took space in the most openings of the last 7 days (of the week reported, in the Monday post), and needs at least two; an opening refilled within a block counts as a grab within a block for everyone who took part of it.
- Weeks: UTC weeks starting Monday.
- Operators: the roster is the operator pool’s allowlist; an operator is removed once
OperatorRemovednames it. Names come fromlabels/staking-operators.json, only where a public source ties the name to the staking address (Chainlink’s reference data directory, Dune’s spellbook or stake.link’s app config). Unnamed operators show their address.
stLINK peg
stLINK is stake.link’s rebasing liquid-staking token, redeemable 1:1 for LINK through stake.link in time. ccip.dev tracks
its price on Curve’s stLINK/LINK-ng pool (0x7e13…1761), the main market.
- History: hourly OHLC of stLINK in LINK from Curve’s price API (
prices.curve.finance), from 2025-04-19. Hours without swaps repeat the last price. - Live: every 5 minutes, one batch of
eth_calls: the pool’s spot (get_p) and EMA (price_oracle), swap quotes (get_dy) for 1k–250k in each direction, and stake.link’s PriorityPool, StakingPool, WithdrawalPool and FundFlowController state. Aget_dyrung that reverts reads as an unprofitable rung (0 out); any other failed call fails the read and the next RPC endpoint is tried. - Gap is spot − 1: negative is a discount, positive a premium.
- Curve depth is the largest quoted size whose swap beats 1:1 after the pool’s fees. Gas is not counted.
- The 1:1 leg. A discount is closed by redeeming stLINK at 1:1. That is instant only against LINK queued for deposit (or from the pool, when stake.link allows instant withdrawals and nobody waits); the rest joins the withdrawal queue, paid out at most every 3 days inside a claim window. A premium is closed by minting at 1:1, instant only while no LINK is queued, into the withdrawal queue and then the pool’s room. Tradeable now is the smaller of the Curve depth and the instant capacity.
- Episodes. An episode opens when the gap stays at or past the private line (0.3%) for two 5-minute checks in a row, and closes when it stays under half that line (0.15%) or flips side for two. History before launch uses the same rule on hourly closes with one bar, so the episode list and an alert’s numbers can differ slightly.
- Episode peak. Public episode posts and cards show the episode’s own peak (“peak so far” while it is open), not the live gap at the moment they are drawn.
- Markers pin hours when 10 or more stake.link vaults began unbonding, or they staked 10,000 LINK or more. They show timing, not cause.
- Not financial advice.
Cross-check
@CCIPMetrics posts its “#CCIP today” totals just after midnight UTC. Each post is dated the day it is posted and covers the previous UTC day: on every day below, its transaction count equals our message count for the previous UTC day. Chainlink’s metrics site (metrics.chain.link) publishes only a monthly cumulative CCIP volume, with no daily figures, so it is compared by month in a separate table.
| Day (UTC) | ccip.dev messages | ccip.dev USD | CCIPMetrics messages | CCIPMetrics USD | Gap and reason |
|---|---|---|---|---|---|
| 2026-10-05 | 2,487 | 22,954,922 | 2,488 | 22,921,006.7 | Messages −1: the API lists exactly 2,487 mainnet messages for the day, matching our ids one for one. USD +0.15% after the 2026-10-07 finalize re-valued the 101 messages first valued before the pricing fallbacks were deployed (24 remain unpriced). Fees: ours $1,245.29, CCIPMetrics $1,252.4 (−0.57%). |
| 2026-10-04 | 1,121 | 17,405,830 | 1,121 | 17,335,113.8 | +0.41% |
| 2026-10-03 | 1,178 | 50,942,816 | 1,178 | 51,231,459.2 | −0.56% |
| 2026-10-02 | 1,977 | 74,446,837 | 1,977 | 74,349,254.0 | +0.13% |
| 2026-10-01 | 2,158 | 49,675,344 | 2,158 | 50,029,455.9 | −0.71% |
| 2026-09-27 | 1,031 | 14,521,483 | 1,031 | 13,512,588.1 | +7.47%. Without PRIME on Tempo ($906,683): +0.76% |
| 2026-09-26 | 1,083 | 12,213,602 | 1,083 | 10,004,805.9 | +22.08%. Without PRIME on Tempo ($2,113,332): +0.95% |
| 2026-09-25 | 1,889 | 40,894,816 | 1,889 | 39,809,129.0 | +2.73%. Without PRIME on Tempo ($904,538): +0.46% |
| 2026-09-24 | 2,309 | 41,650,681 | 2,309 | 38,846,952.5 | +7.22%. Without PRIME on Tempo ($2,709,166): +0.24% |
Notes on the table:
- 2026-10-01 to 10-05: the CCIPMetrics posts were read on x.com: 10-05, 10-04, 10-03, 10-02, 10-01.
- 2026-09-24 to 09-27: these figures come from search-engine copies of the posts, not from the posts themselves.
- PRIME on Tempo: token
0x20c00000000000000000000058d0b8b2cfdb358c. It is not in CCIP’s token list, and DefiLlama prices it at about $1.06. Until late September, CCIPMetrics appears to have valued it at $0. With it excluded, those days match within 1%. From 2026-10-01 our totals match CCIPMetrics with PRIME included.
| Month | ccip.dev month USD | Chainlink month USD | Gap |
|---|---|---|---|
| 2026-06 | 1,095,606,149 | 1,130,622,176 | −3.10% |
| 2026-07 | 1,319,783,375 | 1,327,119,940 | −0.55% |
| 2026-08 | 1,221,665,054 | 1,213,506,818 | +0.67% |
| Cumulative to 2026-08-31 | 22,381,511,846 | 24,117,366,970 | −7.20% (see Known limitations) |
Chainlink’s figures come from https://metrics.chain.link/api/metrics/latest (ccipVolume, generated 2026-09-25).
That page notes that “some outliers and anomalous data have been removed”.