Research · September 2026
State of AI Agent Verification 2026: What Verifiable Receipts and 34,700 Tool Calls Show
AI agent verification has a measurement problem: agents claim results and humans have no way to check them. Zambo's answer is the verifiable receipt, a public record of every tool call that pairs the result with a cryptographic hash and a timestamp, served at a permanent URL anyone can open. As of , the platform has processed 34,700 tool calls and published 40,260 public receipts. The supplied receipt sample records hash and public-page coverage, while the reliability harness recorded 7,149 test results across the full 138-tool surface. This page reports the numbers, the method, and what the data actually supports.
Checking live platform totals from /api/stats. The receipt composition and rankings remain the supplied sample.
For the multi-step version of an execution receipt, see what a workflow receipt verifies.
For the underlying record format, read what a verifiable receipt contains.
The supplied receipt corpus
The supplied receipt sample spans September 16 to 25. Treat it as a sample with receipt-level evidence, not a census or a basis for population-wide claims.
Each receipt carries fields for the task, tool, input arguments, result summary, a verifiable hash of the result (result_sha256), a public receipt URL (zambo.dev/run/<id>), a timestamp, a tier, and a source. There is no claim-vs-execution field in the schema and no chain-anchoring field in the schema, so this page reports neither: the data does not support those metrics, and they are not invented here.
Tier split
76 standard, 11 teaser, 6 full.
Daily receipt counts
| Date | Receipts |
|---|---|
| September 16 | 22 |
| September 17 | 19 |
| September 21 | 12 |
| September 23 | 30 |
| September 24 | 3 |
| September 25 | 7 |
Weekly: ISO week 38: 41, ISO week 39: 52.
- verifiable receipt
- A public record of one AI tool call that includes the input, the result summary, a cryptographic hash of the result, and a timestamp, served at a permanent URL. The hash lets anyone check that the published result matches what the tool returned.
- result hash
- The sha138 digest of the tool's result, recorded on the receipt as
result_sha256. The supplied corpus records whether receipts carry one.
The 138 tools that show up in the receipts
Ranked by receipt count in the supplied receipt corpus:
live_price: 20provibe_audit: 17ghost_audit_site: 16capability_search: 11zambo_universal: 9prompt_shield: 6leadsignal: 6credithunt: 6ghost_audit_status: 1ghost_audit_report: 1
Price lookups and site audits dominate the sample; capability_search shows up 11 times here. In the live platform totals, the all-time leader is capability_search with 4,270 of 34,700 calls (12.3% of all-time calls).
Across the whole platform, the shape is different from the sample. 19,403 calls (55.9% of all-time calls) are classified as work tool calls, meaning calls doing real work rather than discovery or testing.
- work tool call
- A tool call classified as doing real work for the user, as opposed to discovery, capability search, or test traffic.
Reliability harness results
The platform runs a reliability harness that records test results per tool. As of the last run on , it held 7,149 recorded results across the full 138-tool surface, with coverage marked "full". That means every tool in the catalog has recorded results, not just the popular ones.
- reliability coverage
- The share of the tool catalog with recorded harness results. The live coverage label and tool count appear in the reliability totals above.
Reading the numbers honestly
Three claims the data supports
- The supplied receipt sample records hash-backed, public verification evidence.
- Verification is continuous, not one-off. The supplied corpus spans multiple days, and the reliability harness last ran on 2026-10-06 03:06:04 UTC.
- Work traffic is half the platform. 19,403 of 34,700 all-time calls (55.9%) are work tool calls.
Three claims the data does not support
Chain-anchoring shares (the receipt schema has no anchoring fields), claim-vs-execution mismatch rates (no such field exists), and full-population statements are not supported by this corpus. This page does not make those claims.
Frequently asked questions
What is a verifiable receipt?
A verifiable receipt is a public record of one AI tool call. It pairs the input and result with a cryptographic hash and a timestamp, served at a permanent URL anyone can open and check. The supplied sample records whether receipts carry a valid result hash.
How many AI tool calls has Zambo processed?
The current call total, public receipt total, and work-call count appear in the live totals above, which refresh from Zambo's /api/stats endpoint every 60 seconds.
How fresh is the verification data?
The live totals refresh every 60 seconds from /api/stats. The receipt corpus and receipt rankings remain the supplied sample, while the reliability run time and result count are refreshed with the platform totals.
Do the receipts prove what the tool returned?
Each receipt records a sha256 hash of the result, so anyone can check whether the published result matches what the tool returned. The hash does not prove the tool itself was correct; it proves the record has not been altered after the fact.
Can I verify a receipt myself?
Yes. Every receipt in the sample has a public page at a zambo.dev/run/ URL with the result summary, hash, and timestamp. No account is needed.
What can the data not tell me?
The receipt schema has no chain-anchoring fields and no claim-vs-execution fields. The supplied receipt corpus is a sample, not a census, and its relationship to all-time calls changes as the live platform total grows.
Check your own result
Every verification record lives at zambo.dev, where you can run a free call and get your own verifiable receipt in seconds. Try the free tier (20 free calls per metered tool each day, no account) at https://zambo.dev and read the install guide at https://zambo.dev/install?src=research-2026.