DEEPER RESEARCH / 13 NOVEMBER 2026
A more separate set of keys.
A wallet waiting for its first move.
The clearest new clue is a change in operating access. Old credentials remain in the wider network of linked vaults, while TIBBIR’s treasury and fee wallet now use a more separate set. One new credential has access to the fee wallet alone.
This builds on RocketDoc’s original trace. His five-chain mapping and the 0.69 ETH payment are the starting point. The checks below examine what that change could mean, rather than presenting his findings as ours.
1. The old keys were removed here, but remain elsewhere.
At our checked Base block, the removed treasury key 0 was still installed in six other known linked vaults. Removed key 2 remained in three vaults in its separate group. Key 0 also remained in three checked Robinhood Chain vaults.
In plain English: the old credentials were not retired across the whole network. Their access changed specifically around these TIBBIR accounts. This fits a targeted handover or a separation of workspaces.
There is a credible alternative to “new keys mean new people.” Splits’ own privacy guidance recommends different signing credentials for different workspaces. The same people could create new credentials to separate their work. That is a documented possibility, not evidence that this is what they did.
Go deeper · the exact permissions
| Credential | Treasury | Fee wallet |
| Retained key A2 | One of three signers | Not a direct signer |
| New keys A4 and A5 | Two of three signers; together meet the threshold | Either can sign alone |
| New key A6 | Not a direct signer | Can sign alone |
A4 is treasury slot 3, A5 is slot 4, and A6 is the fee wallet’s third new credential. These labels identify public keys, not people. The fee wallet remains owned by the treasury, whose separate owner address remains unchanged. That owner path is additional authority; “two of three” does not describe every possible route to control.
Treasury: 2 signatures required / 3 installed keys
A2 (retained) + A4 (new) + A5 (new)
Fee wallet: 1 signature required / 3 installed keys
A4 (new) + A5 (new) + A6 (new)
owner → treasury
// A6 has a separate spending role in these accounts.
// A key is not a person's name or an employment record.
A separate operating role could handle payments without signing wider treasury actions. No per-payment allowance is established by this threshold. Exact signer reads and permission map ↗
2. We reconstructed the five-chain approvals.
The changes were not merely five similar transactions. Their public signatures belong to shared approval bundles: one for the treasury rotation and one for the fee-wallet setup. We reconstructed the proofs and accounted for all five operations in each bundle.
This strengthens the coordinated-change reading. It also keeps the interpretation grounded: five chains here are one synchronised administrative action, not five independent product launches.
Go deeper · signatures, hidden operations and automation
Both full Merkle trees resolve to the five observed operations, with no unresolved branches in those trees. The treasury’s separate light-operation tree also resolves to those same five chains. This does not exclude other approvals made separately.
We independently verified four P-256 assertions from the Robinhood transactions, including the new A4 signature that sent 0.69 ETH. The signed origin is https://app.splits.org; the relying-party hash matches splits.org. The assertions report user verification and backed-up credentials. They do not identify the human, device vendor or custody arrangement, and are not independent device attestations. A zero signature counter does not prove first use.
The checked Robinhood event history contains no module-enable or upgrade events for either account. The new signer set is not, on its own, evidence of an AI agent taking control. Splits’ documented agent interface also separates proposing a transaction from having authority to sign it.
Ethereum rotation ↗ · Base rotation ↗ · All ten operations and reconstructed proofs ↗ · WebAuthn specification ↗
3. We tested one undeployed-vault explanation much harder.
Could the fresh 0.69 ETH address be a Splits vault that has not been deployed yet? We tested a specific, plausible setup: owned by the treasury, initially signed by A4, with one signature required.
We checked every salt from 0 to 4,294,967,295. None produced the recipient’s address. This rules out that precise setup across the full 32-bit salt range, far beyond the earlier small selection of known salts.
It does not settle the wallet’s identity. A different original signer, owner, threshold, factory or larger salt could still produce an undeployed vault. An ordinary unused wallet remains possible too.
Go deeper · how the negative result was checked
The calculation reproduced seven known project vaults. Additional controls passed at salts 0, 1, 2,147,483,648 and 4,294,967,295. All 256 scan chunks completed without a match. The contract’s salt is a uint256, so 32 bits is a tested range, not a contract limit.
// Public, read-only address prediction; no deployment.
factory.getAddress(
treasury,
[newSpendingKeyA4],
1,
salt // all 2^32 values tested for this configuration
)
// Target: 0x284ddfe7f4e412287a00ac1402fc2f0bab0b16e5
// Result: no match in this explicitly bounded test.
A separate bounded test of ordinary CREATE addresses from ten known accounts, at nonces 0–512, also returned no match. Neither test identifies a private key or a human owner. Test scope, controls and results ↗
Latest recipient check: 21:35:12 UTC on 13 November 2026, Robinhood block 113,670,916: still 0.69 ETH, zero outgoing nonce, no deployed code and no TIBBIR. Open the dated reads ↗
What this could mean for TIBBIR
The best-supported reading is that the project is organising who can operate its treasury and spend its fee reserves. A separate spending credential and a small native-ETH transfer fit preparation for operating work—deploying a contract, paying an expense or funding another account.
The next action from the recipient would be much more informative than the amount alone. A contract deployment would point toward infrastructure; a payment to an identified service would point toward operating expenditure; a factory transaction could identify a new vault.
The remaining links are grouped here: who holds the new credentials; whether the same people changed their keys; what the 0.69 ETH is intended for; and whether any of this supports a specific Ribbita product or a new economic role for TIBBIR. This update improves the operational picture without inventing those answers.
Research coverage and source credits
The Base scan covered every block from 53,658,966 to 57,061,216 for the exact Splits signer-addition and vault-creation events, across all emitting addresses. Its only matches for A4, A5 and A6 were the five expected additions in the treasury and fee wallet. The single rate-limited range was successfully rechecked.
We also paginated all 740 factory logs returned by Ethereum Blockscout. None disclosed the recipient’s creation or the new keys in initial creation parameters. A separate broader Ethereum signer-event scan encountered provider limits and remains partial; its covered ranges and errors are retained, not reported as a universal “no match.”
Publication follow-up: a complete Ethereum scan of blocks 26,402,000–26,408,685 and Base scan of blocks 56,672,200–57,072,570 checked four exact event formats used by Splits, Coinbase Smart Wallet and Safe P-256 signers. Each returned only the five expected A4/A5/A6 additions in these two project accounts. This is bounded event coverage, not a claim that the credentials have never existed elsewhere.
We also searched 38 public Ribbita app files observed in the latest saved capture, plus 11 older retained files, for these addresses and multiple public-key encodings. The known TIBBIR token address was present as a control; the treasury, fee wallet, owner, recipient and replacement credentials were not matched. Server code, other encodings and unvisited routes remain outside this search. Saved files, timestamps and exact search values ↗
The follow-up also checked the following recent intervals. In each completed scan, all matches were the expected additions in the treasury and fee wallet; none supplied another account or human attribution.
| Chain | Blocks checked | Coverage | Expected key matches |
|---|
| Arbitrum | 523,377,152–523,977,152 | complete | 5 |
| Optimism | 159,453,373–159,503,373 | complete | 5 |
| Robinhood | 112,670,071–113,670,071 | complete | 5 |
Credits: RocketDoc (@rocketdoc_eth) for the passkey and five-chain trace; PostAurum for distinguishing paired fee reserves from the fresh-wallet payment; TibbiЯacer for the fee-funded liquidity hypothesis. The latter two specific replies were supplied without exact post permalinks; their notes remain in the saved community record. TIBBIR Universe performed the additional tests described here.
Embedded RocketDoc post and archived source ↓ · Community source notes ↗ · New research evidence ↗