AskGrokWallet · a real monthly record
What the gate actually did for 30 days
Pulled live on 2026-09-24 from the public demo pool. It is not a client's record — a client's version carries their own policy constraints and a name. Everything below is the real thing, unedited, including the parts that look small.
Decisions
| Outcome | Count | Meaning |
|---|---|---|
| Auto-allowed | 15 | inside the policy, no human needed |
| Approved by a human | 10 | escalated, then a person said yes |
| Refused | 7 | stopped (detail below) |
| Executed | 3 | settled, with a receipt |
| Still pending | 7 | waiting on a human when the window closed — nothing executed |
| Total | 42 |
The seven refusals
| Why it stopped | Count |
|---|---|
| A human declined it in the approval inbox | 3 |
| Daily budget already spent | 2 |
| Hit a "never" rule | 1 |
| Action type forbidden by policy | 1 |
The distinction matters: three of these were stopped by a person, four by the rules. Reporting them as one number would overstate automation.
The ledger behind the numbers
entries 68
anchors 19
integrity intact: true
latest anchored entry 67
chain Ethereum Sepolia (chainId 11155111)
anchor transaction 0xe5c9d3c5bfdc29427b6f9f3001c3b94dcacb246a869c4b1a7f870ba3217d393c
confirmed in block 11759509, status success
An anchor is the log's head hash written into a blockchain transaction every hour. Its point: the state of the log at that moment is fixed in public, and afterwards not even we can change it.
Check it yourself
# the whole log, readable by anyone curl -s https://askgrokwallet.io/api/receipts/chain | head -c 400 # the anchor above, on a block explorer: search 0xe5c9d3c5… on Sepolia # one record, end to end curl -sLO https://github.com/richard7463/askgrokwallet/releases/download/verify-receipt-v1.0.1/SHA256SUMS curl -sLO .../verify-receipt-v1.0.1/verify-receipt.mjs shasum -a 256 -c SHA256SUMS node verify-receipt.mjs receipt.json
What this proves — and what it does not
Proves. Every number above comes from signed records; each has a fixed position in the chain; the head is inside a successful on-chain transaction. The counts distinguish policy refusals from human refusals.
Does not prove. That nothing was omitted — that needs an enforcement point that refuses to execute an unrecorded action, which is not built yet. The classification of a refusal is derived from the signed reason text by a fixed rule, not independently verified; a reader who wants to check the classes must hold the receipts. And a quiet month would be reported as a quiet month: zero decisions is information, not a gap in the data.
If it had been a client's month
| This sample | A client's version |
|---|---|
| Public demo pool, all activity | That operator's strategy account only |
| Refusals grouped by reason class | Each refusal mapped to the clause it would have broken |
| Amounts hidden | Amounts disclosable to the funder, still hidden from the public |
None of that is "more proof". It is the same record, said in the language the reader needs.
For the person funding the bot → · Verification procedure for auditors →