He owns every step against his own node — but "clean" hides the cost. HIS SHELF (below) shows the gadgets he must self-run to match the centralized players (green, with cost) and the ones that are simply MISSING (red). The self-run map was lying by omission.
Same send as M1, in Rainbow (1.6.11) · mainnet · defaults · address pasted in. One vendor owns the RPC, the balances and the confirmation; nothing is simulated or screened.
The same send, in Ambire (6.18.7) · mainnet · defaults · address pasted in. Not an EOA tx — a 4337 UserOperation: Ambire's relayer co-signs it as paymaster, Pimlico bundles it, and every hop terminates at Cloudflare.
MetaMask's built-in Swap · ETH→mUSD. One MetaMask API computes the route, the minimum-out, AND the calldata you sign — then takes a 0.875% fee. The rest is the M1 wallet backbone.
MetaMask · send an ERC-20 token (mUSD). Same backbone as the ETH send, but the tx you sign goes to the TOKEN CONTRACT — your recipient is buried in calldata — and Blockaid scans the token too.
MetaMask · swap a token back to ETH. The dangerous part isn't a transaction — it's a Permit2 SIGNATURE that grants the router permission to move your token. The signature that drains wallets.
MetaMask · Buy (fiat on-ramp). The entry to everything — and where you hand over the MOST. No signing here: it's all identity. You're geolocated, jurisdiction-gated, then handed to a KYC provider who takes your legal ID and bank.
⚗ lab run 20260825-124239-buy-metamask— capture-only — drove MetaMask's Buy to the aggregator surface and STOPPED; no purchase, no KYC, no payment ever entered · route JSON
MetaMask · bridge ETH from Ethereum to Base. Same closed engine as the swap, now cross-chain — and it stacks new trust: a bridge operator you can't see, plus Base's sequencer. The convenient path is NOT the escape hatch.
MetaMask · Sell (fiat off-ramp). The exit — and the mirror of the on-ramp. You hand over legal identity AND a bank account, then SIGN a public tx that ties your address to your KYC forever. Crypto leaves before fiat arrives, so the provider can freeze the payout.
⚗ lab run 20260901-140427-sell-metamask— PARTIAL live capture. Reached the off-ramp FRONT surface live: geolocation resolved the exit to region us-co, and MetaMask pulled the region-gated provider/payments catalog (~25KB). Key finding the code map missed: MetaMask's Buy/Sell is NOT in the extension — it opens a hosted web app in a NEW TAB (app.metamask.io/buy), carrying a persistent metametricsId in the URL, and loads third-party AD + analytics trackers (LinkedIn ad pixel, HubSpot CRM, Intercom, LaunchDarkly, Sentry) before you do anything. The Sell-specific legs (provider quote, MoonPay KYC + bank, the signed public exit tx, AML payout) were NOT reached — they sit behind a MetaMetrics/marketing-cookie consent wall and then KYC, which this capture-only run deliberately does not cross. Nothing signed, no KYC, no bank entered. · route JSON
MetaMask · recover on a new device. The one good story on these maps: the recovery act is LOCAL — your phrase re-derives the key on-device, zero third parties. The catch: it works only if you kept the phrase, there is no recovery from loss and no inheritance, and every easier option hands the switch to Google, a guardian service, or a custodian.
⚗ lab run 20260831-213515-recover-metamask— live recovery capture — fresh device, re-imported the throwaway SRP under mitmproxy, then recorded the re-sync. 289 decrypted requests / 48 hosts. Nothing signed or sent. PROVES the seed stayed local: 0 seed bytes appeared in any request body. · route JSON
Rabby's built-in Swap · ETH→USDC. Rabby's server shops the aggregators (1inch won), hands back the bytes you sign, takes 0.25%, and gets the trade record after.
Rainbow's built-in Swap · ETH→USDC. Rainbow's server picks the route (Sprinter) and the bytes you sign — and the swap quietly turned the account into a smart wallet (EIP-7702, Uniswap's Calibur).
Phantom's built-in Swap · ETH→USDC. It sends every address you own on every chain, quotes 4.9% slippage, takes 0.85%, and broadcasts through its own submission service.
Frame (desktop) · plain ETH send. The fewest third parties of any wallet: one company's RPC, a send app loaded from ENS over IPFS, signed locally. Unmaintained since early 2025.
Same send as M1, in Instant Wallet · mainnet · defaults · address pasted in. A passkey signs, the wallet's own server relays; almost every other party sits behind that server. · source ↗ · wallet ↗
ETH → USDC in Instant Wallet's built-in Swap · mainnet · defaults. Uniswap is quoted on chain, LI.FI by the server; the relay sends the winner to the public mempool. · source ↗ · wallet ↗
USDC send in Instant Wallet · mainnet · defaults. Same path as IW; the relay's fee is paid in USDC and here was larger than the payment. · source ↗ · wallet ↗
Every flow's leak, summed across all 34 mapped journeys — five ways to look at the same data, to open the conversation about what's worth attacking first. Weight = how many of the 34 flows carry that data to that place, not raw request count. One honest caveat: 11 of the 34 journeys ride MetaMask, so "reach" partly reflects what we've mapped so far — treat the counts as an initial signal, weight later by real transaction volume.
1 · What leaks widest
Which of your data reaches the most different companies. Your IP is the tell — it rides almost everything.
Split all that leaked data by whether a self-sovereign alternative exists today. This is the buildability gap — the red slice is what has to be built from scratch.
Only 18% of what leaves your machine has a self-sovereign escape today. 82% does not — and 24% has no alternative at all.
24%NO ESCAPE
57%ESCAPE IS PAINFUL
18%SELF-RUN EXISTS
3 · Attack priority — reach × harm
Reach alone isn't the whole story. Here every gadget type is placed by how wide it reaches (→) against how much active harm it can do if it turns evil (↑ block you, lie to you, rewrite your tx). Bubble = number of companies; color = escape (none / painful / exists). Top-right is where the damage and the reach overlap. Surveillance isn't on this axis — analytics sits low here yet collects the most; that's what chart 1 measures.
4 · By gadget type — the flows
The actual data ribbons, actors folded into categories (Infura = Alchemy = RPC), sorted by reach. Box border = escape: red = none, amber = painful, green = exists. See the Gadget Shelf for the fix + cost.
5 · By company — who specifically
The same leaks, named — the 16 widest-reaching of the 55 companies that receive something. "The node provider" is one throat — but it's a different company per wallet.
6 · What the CROPS version looks like
For each gadget, the move that creates the self-sovereign counterfactual. Once the counterfactual exists, every hosted version has to explain why it doesn't work that way. Sorted by leverage: no escape first, widest reach first.
Wallet backendTURN IT OFF31/34 flows
Ship flags and defaults signed in the release; pull updates as signed public lists anyone can mirror — no per-user queries. What breaks: the vendor's instant kill-switch and A/B tests. Not your sends.
Hosted dApp frontendBUILD THE EXAMPLE2/34 flows
Pin the frontend to IPFS/ENS. The test is binary: does the dApp work with nothing hosted?
Fiat on/off-rampOPEN PROBLEM2/34 flows
No self-sovereign fiat rail exists; peer-to-peer is the only escape and it's thin. The open problem.
ERC-4337 bundlerRUN A REPLACEMENT1/34 flows
Self-bundle: wrap your own UserOp and submit it through your node.
TLS edge / CDNOPEN PROBLEMfronts the rest
Falls away only when what's behind it (frontends, backends) becomes self-hostable.
RPC — chain readsBUILD THE EXAMPLE32/34 flows
A wallet that ships with a light client (Helios) — reads verified against consensus, not trusted. Full escape stays your own node.
Block buildersSET A STANDARD32/34 flows
Protocol-level: forced-inclusion lists (FOCIL) so no builder can keep a tx out.
Tx broadcast / relayBUILD THE EXAMPLE17/34 flows
Wallet-native p2p broadcast: hand the signed tx to the public mempool from your own machine.
Block explorerBUILD THE EXAMPLE17/34 flows
Otterscan on your own node today; 'your own etherscan' is an AI-buildable afternoon.
Wallet / code dist.SET A STANDARD16/34 flows
Reproducible builds signed by the team, verifiable outside the store.
Indexer — balancesRUN A REPLACEMENT12/34 flows
Local indexing derived from your own node, behind the same wallet UI.
Prices & market dataRUN A REPLACEMENT8/34 flows
Read prices onchain (DEX TWAPs) — or go without fiat values.
Analytics / telemetryTURN IT OFF24/34 flows
Nothing to build. Default it to off.
Tx / address screeningSET A STANDARD12/34 flows
A signed public risk-list any wallet pulls and checks locally — no server sees the query.
Tx simulationALREADY EXISTS4/34 flows
Local simulation against your own node works today — make it the default.
The categories from Access Vectors you can't route around today (NONE), plus RPC (PAINFUL), taken one at a time. Who holds each position, what it sees, what it can do, what you can do now, and what would have to exist. Counts come from the route data (a stop shared by several maps counts once per map); the own-node lines come from lab runs with the RPC set in each wallet (Your Own Node).
1
TLS edge / CDNzero option: NONE495 stops on 33 of 34 maps · 76% live-observed
What it sees: Signed raw tx, Full unsigned tx, All account addresses, Selected address, Recipient address, Amount, Balances + positions
What it can do: block 188 of these stops, 120 of them with no setting to turn off; give 78 answers where a lie costs funds or ships code; link your IP to your addresses over time.
What you can do today: Unavoidable when using any hosted service. With your own node set in the wallet: 17 of 99 tested stops in this category moved off it; 82 still went to it.
What would have to exist: Falls away only when what's behind it (frontends, backends) becomes self-hostable. Fixes that retire the most of these stops: Private RPC access (89); Local or verifiable indexing (81); Verifiable RPC reads (70).
2
Wallet's own backend (flags, config, prices)zero option: NONE330 stops on 31 of 34 maps · 78% live-observed
Who holds it: Consensys · MM APIs (114), Rabby · DeBank API (87), Coinbase · wallet APIs (46), Instant Wallet (36), Phantom · APIs (34), Google · Firebase (8), WalletConnect (4), Ambire · monitor (1)
What it sees: Signed raw tx, Full unsigned tx, All account addresses, Selected address, Recipient address, Amount, Balances + positions
What it can do: block 105 of these stops, 80 of them with no setting to turn off; give 40 answers where a lie costs funds or ships code; link your IP to your addresses over time.
What you can do today: Use a wallet that has no phone-home backend. With your own node set in the wallet: 2 of 68 tested stops in this category moved off it; 66 still went to it.
What would have to exist: Ship flags and defaults signed in the release; pull updates as signed public lists anyone can mirror — no per-user queries. What breaks: the vendor's instant kill-switch and A/B tests. Not your sends. Fixes that retire the most of these stops: Local or verifiable indexing (56); Verifiable RPC reads (34); Local address-risk checks (31).
3
Hosted dApp frontendzero option: NONE14 stops on 2 of 34 maps · 50% live-observed
Who holds it: Uniswap Labs (12), Avara · app.aave.com (1), Family (1)
What it sees: Selected address, Amount, Balances + positions, Tx hash, Swap route + quote (path, amountOut, minOut, Universal Router calldata), Swap token pair (tokenIn/tokenOut), IP address
What it can do: block 9 of these stops, 9 of them with no setting to turn off; give 3 answers where a lie costs funds or ships code; link your IP to your addresses over time.
What you can do today: Sometimes: self-host the UI or call the contract directly — often NOT possible. With your own node set in the wallet: 0 of 6 tested stops in this category moved off it; 6 still went to it.
What would have to exist: Pin the frontend to IPFS/ENS. The test is binary: does the dApp work with nothing hosted? Fixes that retire the most of these stops: Independently hosted frontends (2); Local or verifiable indexing (2); Verifiable RPC reads (1).
4
RPC — chain reads & writeszero option: PAINFUL129 stops on 28 of 34 maps · 81% live-observed
What it sees: Signed raw tx, Full unsigned tx, All account addresses, Selected address, Recipient address, Amount, Balances + positions
What it can do: block 83 of these stops, 26 of them with no setting to turn off; give 41 answers where a lie costs funds or ships code; link your IP to your addresses over time.
What you can do today: Run your own full node and point the wallet at it. Cost: Sync + run a full node — disk, time, upkeep. With your own node set in the wallet: 35 of 37 tested stops in this category moved off it; 2 still went to it.
What would have to exist: A wallet that ships with a light client (Helios) — reads verified against consensus, not trusted. Full escape stays your own node. Fixes that retire the most of these stops: Private RPC access (101); Verifiable RPC reads (66); Censorship-resistant broadcast (14).
Each wallet with Ethereum pointed at a node on your own network, set in the wallet's own settings, then run in the lab. Green went to your node; red still left your machine; grey didn't come up in the run. Across these maps your node answered 38 of 153 tested requests (25%). The rest go to the same companies as before.
Rainbow Send45% of tested requests5 to your node · 6 still out · 11 not tested · +2 hosts no stop covers · real send ↗ · ⚗ lab run
Rainbow's platform API still gets your address, balances and tx list. A node elsewhere on your LAN is blocked by the extension; it has to run on the same computer.
Rabby polls the receipt with your tx hash seconds after you send, and its pre_exec_tx gets the full unsigned tx.
U1
Uniswap Swap13% of tested requests2 to your node · 13 still out · 26 not tested · +42 hosts no stop covers · dry run · ⚗ lab run
Uniswap reads the chain through its own gateway (unirpc).
M3
Aave Deposit11% of tested requests2 to your node · 17 still out · 22 not tested · +37 hosts no stop covers · dry run · ⚗ lab run
Aave's app reads the chain through its own server (app.aave.com/api/rpc-proxy). Your wallet's node never sees the dapp's reads.
ON
Fiat On-Ramp0% of tested requests0 to your node · 2 still out · 2 not tested · +58 hosts no stop covers · dry run · ⚗ lab run
MetaMask geolocates you and gates providers by region. Your node got 4 of MetaMask's background reads, none of them stops on this map.
OF
Fiat Off-Ramp0% of tested requests0 to your node · 3 still out · 5 not tested · +56 hosts no stop covers · dry run · ⚗ lab run
MetaMask geolocates you and gates providers by region; the provider page hits Infura with its own key. Your node got 4 of MetaMask's background reads, none of them stops on this map.
Correctness is the moat: before this map names who leaks, every claim has to be checkable. Here is what's backed by a live capture and what still isn't. Computed from the route data.
The evidence
70% of every data-exposure claim on these maps is backed by a live packet capture. The rest is traced from source code or inferred — and still needs a capture to confirm.
70%LIVE-OBSERVED
27%CODE-TRACED
Still to validate — 18 actors, no live capture yet
Traced from source or inferred, not yet seen on the wire. The queue for the next captures.
EtherscanCODE-TRACED48 claims
verify-a-tx is click-gated — not triggered by the automated send capture
Builders · relaysINFERRED32 claims
block builders / relays — inferred from protocol; every tx passes them
Pimlico bundlerCODE-TRACED13 claims
ERC-4337 bundler — smart-account infra, traced from code not a wallet call
Ambire · relayerCODE-TRACED12 claims
Ambire smart account can't onboard in the VM (needs a live relayer to derive its address) — the inability is itself the finding
Ambire · invictus RPCCODE-TRACED11 claims
same Ambire onboarding blocker
DeBankCODE-TRACED3 claims
capture pending
Ambire · cenaCODE-TRACED3 claims
same Ambire onboarding blocker
ZerionCODE-TRACED3 claims
capture pending
LI.FICODE-TRACED3 claims
capture pending
CCIP gatewayINFERRED2 claims
CCIP-read offchain resolver — only fires for names that use it
Avatar hostINFERRED2 claims
ENS avatar image host — cosmetic, not exercised in the captured send
Ethereum p2pREFERENCE2 claims
the Mountain Man's own broadcast — a constructed reference flow
AmplitudeCODE-TRACED2 claims
analytics — fires to an ingest subdomain the capture didn't decode
ENS metadataCODE-TRACED2 claims
ENS metadata/avatar service — not hit by a plain resolve
The GraphCODE-TRACED2 claims
indexer path — traced from code
Avara · app.aave.comCODE-TRACED1 claim
Aave app API — captured run exists (m3); these rows are code-traced
FamilyCODE-TRACED1 claim
Family/Rainbow name service — traced from code
Ambire · monitorCODE-TRACED1 claim
same Ambire onboarding blocker
Proven live — 37 actors
Seen in a mitmproxy capture; most on a real mainnet transaction. Links go to the lab run and the tx.
Every actor generalized to a category (Infura and Alchemy are both just "RPC"), ranked by prevalence — how many flows it sits on — and exposure — what it can do to you. The right column is the zero option: is there a fully self-sovereign fallback, and what does the Mountain Man do instead? None = top priority; painful = still worth fixing; local = the escape hatch already exists. Prevalence is an initial signal, not the final sort — weight later by real transaction volume. Computed from the route data.
On 34 flows AB1APBRCB1CBACBSCBTFR1IWIWSIWTM1M2M3M4MTNOFONPH1PHAPHSPHTPWR1RAPRB1RBARBSRBTRCRSRTU1 · can: can block (no setting removes it) · malicious code · links identity over time
zero option: NONE — no self-sovereign option · Mountain Man: Unavoidable when using any hosted service
Cloudflare/CloudFront front nearly everything. They can block by jurisdiction and terminate TLS. No self-sovereign option while using hosted services.
2
Block builders & relaysBuilders · relays
On 32 flows AB1APBRCB1CBACBSCBTFR1IWIWSIWTM0M1M2M3M4MTOFPH1PHAPHSPHTPWR1RAPRB1RBARBSRBTRSRTU1 · can: can block (no setting removes it)
zero option: PAINFUL — exists, costs real UX/$ · Mountain Man: Wait for a non-censoring local/among builder to include you
On EVERY flow. Can censor today in principle; the fix is Fossil / forced inclusion. Binary-ish: does a censorship-resistant path exist.
On 29 flows AB1APBRCB1CBACBSCBTIWIWSIWTM1M2M3M4MTNOFONPH1PHAPHSPHTR1RAPRB1RCRSRTU1 · can: can block (no setting removes it) · malicious code · links identity over time
zero option: NONE — no self-sovereign option · Mountain Man: Use a wallet that has no phone-home backend
Feature flags, remote config, and defaults served from the wallet vendor. Proprietary; no self-sovereign version short of a different wallet.
On 29 flows AB1APBRCB1CBACBSCBTFR1IWIWSIWTM1M2M3M4MTNOFPH1PHAPHSPHTPWRB1RBARBSRBTRCU1 · can: can block (no setting removes it) · malicious code · links identity over time
zero option: PAINFUL — exists, costs real UX/$ · Mountain Man: Run your own full node and point the wallet at it
The single most prevalent dependency. Sees every read and takes your broadcast. A node is the fallback but costs sync time, disk, and upkeep.
5
Block explorer (verify a tx completed)DeBank, Etherscan
On 17 flows APBRM1M2M3M4MTPWR1RAPRB1RBARBSRBTRSRTU1 · can: false belief · links identity over time
zero option: PAINFUL — exists, costs real UX/$ · Mountain Man: Check against your own node / run your own explorer
Etherscan is a UNIVERSAL trust assumption — every user, even Mountain Men, sanity-checks a tx on Etherscan. A node confirms it, but people still trust the explorer's UI.
On 9 flows AB1APBRM1M2M3M4MTU1 · can: can block (no setting removes it) · funds stuck · links identity over time
zero option: PAINFUL — exists, costs real UX/$ · Mountain Man: Broadcast straight to your own node / the public mempool
Private relays (Smart Transactions) and bundlers sit between you and inclusion. Direct broadcast is the fallback but loses MEV protection and can be slow.
9
Wallet / code distributionGitHub, Google · Web Store
On 8 flows AB1FR1M3NPWRB1RCU1 · can: can block (no setting removes it) · malicious code
zero option: PAINFUL — exists, costs real UX/$ · Mountain Man: Build the wallet from source and side-load it
The Web Store / GitHub that ships (and can silently update) the wallet code. Building from source avoids the auto-update trust but is painful.
On 2 flows M3U1 · can: can block (no setting removes it) · malicious code · links identity over time
zero option: NONE — no self-sovereign option · Mountain Man: Sometimes: self-host the UI or call the contract directly — often NOT possible
THE hard gap. If a dApp's frontend depends on its own backend, you can't use it without them. Safe can be self-hosted; many can't. No zero option in the general case.
14
Fiat on-ramp / off-rampCoinbase on-ramp, MoonPay
On 2 flows OFON · can: can block (no setting removes it) · funds lost
zero option: NONE — no self-sovereign option · Mountain Man: Already hold crypto — earn it, get paid in it, or acquire peer-to-peer
The funnel entry. No private fiat rail exists: geolocation + jurisdiction gate + full KYC. The only zero option is to skip fiat entirely.
15
Name resolution (ENS)CCIP gateway, ENS metadata
On 2 flows M2M3 · can: can block (no setting removes it) · funds lost
zero option: LOCAL — self-sovereign version exists · Mountain Man: Resolve on-chain via your own node (eth_call)
ENS resolves on-chain; a node does it with no extra party. Metadata/avatar services are optional cosmetics.
16
ERC-4337 bundler / paymasterPimlico bundler
On 1 flow AB1 · can: can block (no setting removes it) · funds lost
zero option: NONE — no self-sovereign option · Mountain Man: N/A — an EOA avoids it entirely
Smart-account infra. A 4337 account cannot even onboard or transact without a live bundler/paymaster (see Ambire). No offline option.
The clean Mountain Man map hides his real UX. Every gadget the centralized flows outsource to a company (a leak you can see), the Mountain Man must run himself — at a cost the map never shows — or it's simply missing (no self-sovereign option today). This is that hidden shelf: per gadget, who the centralized flows use, what the Mountain Man does instead and what it costs, and whether you could build your own now with AI. The MISSING rows are the real gaps; the buildable now ones are where the cost is already dropping. Computed from the route data.
Mountain Man: Self-host the UI or call the contract raw — often impossible
THE hard gap. If a dApp's frontend depends on its own backend, you can't use it without them. Safe can be self-hosted; many can't. No zero option in the general case.
4
Fiat on-ramp / off-rampMISSINGnot buildable yet
Centralized flows use:Coinbase on-ramp, MoonPayOFON · outsourced on 2 flows
Mountain Man: No private fiat rail — earn it, get paid in it, or go P2P
The funnel entry. No private fiat rail exists: geolocation + jurisdiction gate + full KYC. The only zero option is to skip fiat entirely.
Mountain Man: Sync + run a full node — disk, time, upkeep
The single most prevalent dependency. Sees every read and takes your broadcast. A node is the fallback but costs sync time, disk, and upkeep.
8
Block explorer (verify a tx completed)SELF-RUN (costly)buildable now
Centralized flows use:DeBank, EtherscanAPBRM1M2M3M4MTPWR1RAPRB1RBARBSRBTRSRTU1 · outsourced on 17 flows
Mountain Man: Run your own explorer (Otterscan) — or build one with AI
Etherscan is a UNIVERSAL trust assumption — every user, even Mountain Men, sanity-checks a tx on Etherscan. A node confirms it, but people still trust the explorer's UI.
9
Indexer — balances, history, token listsSELF-RUN (costly)buildable now
Centralized flows use:ACI · aavechan, Hyperliquid, Merkl, Rainbow · APIs, The Graph, ZerionIWIWSIWTM3R1RAPRB1RBARBSRBTRSRT · outsourced on 12 flows
Mountain Man: Run your own indexer over your node
The 'what do I own / what happened' layer. Self-indexing is possible but heavy; most wallets have no local option.
Mountain Man: Broadcast to your own node / the public mempool
Private relays (Smart Transactions) and bundlers sit between you and inclusion. Direct broadcast is the fallback but loses MEV protection and can be slow.
ENS resolves on-chain; a node does it with no extra party. Metadata/avatar services are optional cosmetics.
17
Transaction simulationSELF-RUNbuildable now
Centralized flows use:— · outsourced on 0 flows
Mountain Man: Run a simulator against your own node
Blockaid-style pre-sign simulation. A local version is feasible today — MetaMask required Blockaid to also ship an obfuscated local build. Good-enough local sim exists.
Every actor on 2+ maps, or where a lie costs funds or code, or with a block no setting removes. CROPS scored 0–3 per axis, 0 worst: Censorship-Resistance (can they block?) · Open-source · Privacy (what they collect) · Security (what a lie costs). Computed from the route data, not hand-ranked.
Serves AB1APBRCB1CBACBSCBTFR1IWIWSIWTM0M1M2M3M4MTOFPH1PHAPHSPHTPWR1RAPRB1RBARBSRBTRSRTU1 · receives · CROPS CR0 · O1 · P2 · S3 · can block, no escape
2
Cloudflare edgeTLS terminates here for every fronted API call
Serves AB1APBRCB1CBACBSCBTFR1IWIWSIWTM1M2M3M4MTNOFONPH1PHAPHSPHTPWRB1RBARBSRBTRCU1 · receives · CROPS CR0 · O0 · P0 · S0 · a lie ⇒ malicious code · can block, no escape · can link identity over time
3
Etherscanetherscan.io
Serves APBRM1M2M3M4MTR1RAPRB1RBARBSRBTRSRTU1 · receives · CROPS CR3 · O0 · P0 · S2 · a lie ⇒ false belief · can link identity over time
4
ConsensysMM APIs + Infura + SERVO relay
Serves APBRM1M2M3M4MTNOFONRCU1 · receives · CROPS CR0 · O0 · P0 · S0 · a lie ⇒ malicious code · can block, no escape · can link identity over time
5
AWS CloudFrontCDN + bucket for static files, and the TLS terminator for api.rabby.io (CloudFront → custom origin)
Serves IWSNR1RAPRB1RBARBSRBTRSRT · receives · CROPS CR0 · O0 · P0 · S0 · a lie ⇒ malicious code · can block, no escape · can link identity over time
The interventions from the working list, ranked by what each retires across the maps: requests, funds-level lies, blocks no setting removes, answers that shape what you sign. Computed from the route data, not hand-ranked.
1
Private RPC access
Reads and broadcasts that don't tie your IP to your addresses.
Retires 47 requests across 33 maps · 17 where a lie costs funds or code · 9 blocks no setting removes · 19 that shape what you sign · 47 that carry identity AB1APBRCB1CBACBSCBTFR1IWIWSIWTM1M2M3M4MTNOFPH1PHAPHSPHTPWR1RAPRB1RBARBSRBTRCRSRTU1
2
Verifiable RPC reads
Light-client or proof-backed reads: the answer can be checked, so the RPC can't lie about state.
Retires 52 requests across 25 maps · 17 where a lie costs funds or code · 5 blocks no setting removes · 21 that shape what you sign · 52 that carry identity AB1APBRFR1IWIWSIWTM1M2M3M4MTNOFPWR1RAPRB1RBARBSRBTRCRSRTU1
3
Independently hosted frontends
App and wallet code built and served outside one store, CDN, or remote flag service.
Retires 25 requests across 13 maps · 14 where a lie costs funds or code · 13 blocks no setting removes · 3 that shape what you sign · 25 that carry identity AB1IWIWSIWTM3M4NOFONPWR1RB1U1
4
Open simulation
Run the tx preview yourself instead of trusting a closed simulator's answer.
Retires 14 requests across 21 maps · 10 where a lie costs funds or code · 9 blocks no setting removes · 14 that shape what you sign · 14 that carry identity AB1APBRCBACBSM1M2M3M4MTPH1PHAPHSPHTR1RAPRBARBSRSRTU1
5
Portable clear-signing data
Open metadata that lets the wallet say locally what calldata does.
Retires 11 requests across 12 maps · 6 where a lie costs funds or code · 2 blocks no setting removes · 6 that shape what you sign · 11 that carry identity APIWIWSIWTM3NOFR1RAPRSRTU1
6
Censorship-resistant broadcast
A path to inclusion that no single relay, RPC, or builder set can filter.
Retires 33 requests across 32 maps · 4 where a lie costs funds or code · 25 blocks no setting removes · 5 that shape what you sign · 20 that carry identity AB1APBRCB1CBACBSCBTFR1IWIWSIWTM0M1M2M3M4MTOFPH1PHAPHSPHTPWR1RAPRB1RBARBSRBTRSRTU1
7
Local or verifiable indexing
Balances, history, and positions from your own or provable index, not a closed API fed all your addresses.
Retires 37 requests across 31 maps · 1 block no setting removes · 4 that shape what you sign · 37 that carry identity AB1APBRCB1CBACBSCBTFR1IWIWSIWTM1M2M3M4MTNPH1PHAPHSPHTPWR1RAPRB1RBARBSRBTRSRTU1
8
Local address-risk checks
Phishing, sanctions, and scam screening against open local data — not a vendor that sees every recipient pre-sign.
Retires 11 requests across 16 maps · 8 that shape what you sign · 11 that carry identity AB1APBRCB1CBTM1M2M3M4MTNR1RAPRSRTU1
102 third-party requests match no fix on this list — mostly prices, APRs, analytics, and avatars.