The 5 steps, each with its receipt
Expand any step to see its verifiable receipt: which tool ran, the hash of its inputs, the hash of its outputs, the committed fingerprint, and when it executed. Press Verify this receipt and your browser recomputes the SHA-256 fingerprint from the receipt bytes and compares it to what was committed.
1 measure the error spike
Why this step: Errors are the symptom. Before blaming any deploy, the agent measures exactly when /checkout started failing and how far it deviated from baseline.
Show the verifiable receipt
- Receipt ID
- bc570d4d-b0a8-4582-bb67-68e922421ea9
- Tool
- code_exec
- Inputs hash
- sha256:4fc9f0547f5a633530a60d0b3e165eb3b2e63c58cf3b45bdb1069fc3dfb11aa2
- Outputs hash
- sha256:fdcc84f2f2ac27d221875f8f322a97f3f991f8eac4209dc9aa22a73878aac3fd
- Committed fingerprint
- sha256:49c37f11d502dc0c4828af2fdaf95dab0353a56dad2527075dd0d07fe1631de3
- Executed
- 2026-10-04T18:43:50.787Z to 2026-10-04T18:43:50.866Z
2 list recent deploys
Why this step: A sudden error spike usually has a trigger. Deploys are the most common trigger, so the agent lists every checkout-api deploy near the incident window.
Show the verifiable receipt
- Receipt ID
- 45d1b382-3dbd-4e58-aa4a-64ef19267d0f
- Tool
- code_exec
- Inputs hash
- sha256:b43d008b2c5e1d2525566e4b0a9fdc2736fbed8a8e0c87a6863a0059bffd17a0
- Outputs hash
- sha256:911db0ce212fd5cd019ae30d8ebfb0bc40406c4175d31ea639ce040917b1cbb9
- Committed fingerprint
- sha256:b4fff84ec5d095bbb2c870f5e0772fbd5876b00d9ac7ec3a3fa12ba4f0380763
- Executed
- 2026-10-04T18:43:50.866Z to 2026-10-04T18:43:50.909Z
3 read support tickets
Why this step: Logs show what the servers saw; tickets show what users felt. The agent checks whether real customers reported the same failure in the same window.
Show the verifiable receipt
- Receipt ID
- f472b49e-aa79-45bd-a46f-6738cf8f21c5
- Tool
- code_exec
- Inputs hash
- sha256:e589193c752003d502f2e0988f2984ad1ff76244468230e9329cae592bd95756
- Outputs hash
- sha256:f34e1c93ba331e89a26ea88102e0562d1f5e528122cd6bbe3d3c233fa72d9bda
- Committed fingerprint
- sha256:2aa99e9408a1430b1477858b049abd30f6524d9ff78659c0f9cec04ac194cae9
- Executed
- 2026-10-04T18:43:50.909Z to 2026-10-04T18:43:50.965Z
4 correlate the findings
Why this step: Three sources, three timelines. The agent lines them up on one clock: first error minute, latest deploy before it, first user report. Causality is argued from timing, not guessed.
Evidence chain: this step builds on step 1 step 2 step 3
Show the verifiable receipt
- Receipt ID
- fd9190a3-1d55-49f5-b656-b3b0d07228e8
- Tool
- code_exec
- Inputs hash
- sha256:7886c28623a4b955329c5a41784774cb731bff88f6040fbe3c18304de8c551ee
- Outputs hash
- sha256:f996c2408c5eccf8647ba169d54b555820d0648199f5b8f8e4f44bad1ac165a4
- Committed fingerprint
- sha256:3d10661c199c84381dfb601ef57642f902374fb4e96468bef56aadadec400533
- Executed
- 2026-10-04T18:43:50.965Z to 2026-10-04T18:43:50.998Z
5 recommend a fix
Why this step: A recommendation without evidence is an opinion. This step turns the correlated findings into one actionable answer, each reason pointing back to a receipt that proves the underlying measurement.
Evidence chain: this step builds on step 1 step 2 step 3 step 4
Show the verifiable receipt
- Receipt ID
- f5db7af4-c04b-4ac0-837a-3b6b8d1dafe0
- Tool
- code_exec
- Inputs hash
- sha256:5f9da5f9a058586b33609cf818578e65d8652d4a7d7c92523f0c643b29a5e8c5
- Outputs hash
- sha256:9c2c2c50040e96822da483bb9398986aef8b2d522f8009465e2f1cfcc693b977
- Committed fingerprint
- sha256:1b3f42222ce606883e040e27053642390352a7188c6073c79e08458578ff8221
- Executed
- 2026-10-04T18:43:50.999Z to 2026-10-04T18:43:51.035Z
The tamper test
Hashes are either doing real work or they are decoration. This button takes a copy of step 1's receipt, flips one byte in the copy, and asks the verifier to check it. The original receipt is never touched.
The recommendation and its evidence chain
Roll back deploy d-4417 immediately
This is not an opinion. Each reason below points to a receipt above that proves the underlying measurement:
- Checkout errors began at 2026-10-02T14:15:00Z, one minute after d-4417 shipped at 14:14:00Z. Proved by step 1 and step 2.
- d-4417 refactored the payment retry path, the exact code path now throwing 500s. Proved by step 2.
- Six customers reported checkout failures starting 14:16:00Z, matching the log timeline. Proved by step 3.
- No other checkout-api deploy exists in the window. Proved by step 2.
The step 4 receipt commits the correlation of all three timelines, and the step 5 receipt commits this recommendation with its evidence chain: 1, 2, 3, 4. Confidence: high. Next step: re-run the error-spike measurement after the rollback to confirm errors stop.
Give your agent receipts
WhyTrail's receipts are AER-1 verifiable receipts, the open format. The fastest way to add them to your own agent:
pip install aer1kit
Three lines and every action your agent takes mints a receipt you can verify like the ones above. Then try the live demo: run a real call, get a real receipt, forge it, watch the fingerprint catch you. Need unlimited runs? $0.99 for 24 hours. Or go straight to zambo.dev.