| --- |
| license: cc-by-4.0 |
| task_categories: |
| - other |
| language: |
| - en |
| tags: |
| - ai-agents |
| - tool-use |
| - verification |
| - receipts |
| - mcp |
| - aer1 |
| - aer-1 |
| pretty_name: Zambo Agent Execution Receipts |
| size_categories: |
| - n<1K |
| --- |
| |
| # Zambo Agent Execution Receipts |
|
|
| Zambo is the verifiable work layer for AI: 153 tools behind one MCP endpoint, continuity that carries sessions, memory, and work across AIs, and a verifiable AER-1 receipt for every job, so "done" is something you can check, not something you have to trust. |
|
|
| An agent receipt is a record of one ai agent tool call that anyone can check: it keeps the exact saved bytes and their sha-256 digest, and it proves the result was not changed since it was saved, not that it was right. |
|
|
| A growing public corpus of real AI tool-call execution receipts minted on |
| [Zambo](https://zambo.dev). Every row |
| describes one genuine tool call an agent made through Zambo's API and links |
| the public, machine-checkable receipt for that call. |
|
|
| ## Try it yourself |
|
|
| - Live demo: https://zambo.dev/demo/ runs a real call and mints a real |
| receipt in seconds, no signup. |
| - The receipt format is AER-1, an IETF Internet-Draft |
| (draft-zambo-aer1-26): https://datatracker.ietf.org/doc/draft-zambo-aer1/26/ |
| - Guides and worked examples: https://rambozambodotdev.gitlab.io/aer1-hub/ |
| - Connect your AI to the same tools: https://zambo.dev/mcp (20 free calls per day; Day Pass $0.99 for 24 hours: https://zambo.dev/day-pass/) |
|
|
| ## Why this dataset exists |
|
|
| When an AI agent calls a tool, the result is usually a claim: text in a chat |
| log, gone the moment the session ends. Zambo attaches a verifiable receipt to |
| every call: the tool that ran, its arguments, the observed result, a SHA-256 |
| fingerprint of that result, and a public URL carrying machine-readable |
| JSON-LD. This dataset collects those receipts in one crawlable place, so |
| researchers, evaluators, and future models can study what verified agent tool |
| use looks like at scale. |
|
|
| ## Dataset snapshot |
|
|
| - Rows: 328 |
| - Unique receipt URLs: 328 |
| - Tools covered: 10 (capability_search, credithunt, ghost_audit_report, ghost_audit_site, ghost_audit_status, leadsignal, live_price, prompt_shield, provibe_audit, zambo_universal) |
| - Date range: 2026-09-16 to 2026-11-01 |
| - Last updated: 2026-11-02 |
| |
| ## Schema |
| |
| | field | type | description | |
| |---|---|---| |
| | task | string | short human label for what the call was asked to do | |
| | tool | string | Zambo tool that executed the call | |
| | args_json | string | JSON-encoded arguments passed to the tool | |
| | result_summary | string | plain-language summary of the observed result (truncated at 500 chars; full detail lives on the receipt page) | |
| | result_sha256 | string | SHA-256 fingerprint of the result, as recorded on the receipt | |
| | receipt_url | string | public URL of the verifiable receipt (https://zambo.dev/run/<uuid>) | |
| | created_at | string | ISO-8601 timestamp when the receipt was minted | |
| | tier | string | `full` (complete AI-reviewed findings), `teaser` (honest preview tier, findings are pattern-level only), or `standard` | |
| | source | string | which export batch produced the row | |
|
|
| ## How rows were produced |
|
|
| Two batches: |
|
|
| 1. **banked-ledger**: receipts banked in Zambo's own operations between |
| 2026-09-16 and 2026-09-23 (code audits, security audits, market-data |
| checks, catalog searches). Each receipt URL was re-fetched live at export |
| time; only URLs returning HTTP 200 with valid receipt JSON-LD were kept. |
| 2. **fresh-mint-2026-09-23** (and nightly appends after): new tool calls run |
| against the live public Zambo API under the free tier, each receipt page |
| fetched and JSON-LD-verified before inclusion. |
|
|
| Nothing here is synthetic. If a receipt page ever stops serving, the |
| `receipt_url` for that row will no longer resolve; the row is kept as a |
| historical record. |
|
|
| ## Intended uses |
|
|
| - Training and evaluating models on verified (tool call, result) pairs |
| - Studying receipt formats for agent execution verification |
| - Benchmarking "did the agent actually do the work" claims against checkable evidence |
|
|
| ## Considerations and limitations |
|
|
| - Rows labeled `tier: teaser` are the honest preview tier of the audit tools: |
| pattern-level checks, not full AI-reviewed findings. The receipt says so, |
| and so does this card. |
| - `result_summary` is truncated; treat the receipt page as ground truth. |
| - The corpus grows nightly; row counts in this card describe this snapshot. |
|
|
| ## Citation |
|
|
| If you use this dataset, please cite: |
|
|
| ``` |
| Zambo Agent Execution Receipts (2026). Hugging Face dataset |
| zambodotdev/agent-execution-receipts. Source receipts: https://zambo.dev |
| ``` |
|
|
| ## License |
|
|
| CC-BY-4.0. Receipt content (tool results) originates from public third-party |
| sources via Zambo's tools (e.g. CoinGecko market data); the compilation and |
| descriptions are released under CC-BY-4.0 with attribution to Zambo |
| (https://zambo.dev). |
|
|
