receipts --board
The falsification record
Every entry below is a claim-vs-effect gap: a system that asserted a fix while the real effect was wrong. These are the code catches. The ledger also carries signed receipts across the other system types shown on the coverage grid. Each is a signed, content-addressed receipt. Nothing here asks for your trust: the raw JSON is downloadable, and every receipt can be re-verified in your own browser with no server involved.
ledger --code
The code catches: 10 signed falsifications across 6 public repositories, each one a real bug, a real test, and a claim that did not survive the re-run. These are one family of 44 signed receipts in the ledger. The other 34 cover system types that are not code. One of them, the agent-fabrication finding, is below. The rest are in the coverage list.
| Verdict | Repo | Instance | Claim → Effect | Record hash | Verify |
|---|---|---|---|---|---|
| falsified | click | click-0f71fe7 | claimed fixed → test failed18 failed, 619 passed in 1.66s | 52d6649441fc… | verify → |
| falsified | fastapi | fastapi-08dad5c6 | claimed fixed → test failed1 failed in 0.28s | 305590431ef8… | verify → |
| falsified | httpx | httpx-1e11096 | claimed fixed → test command failed3 failed, 32 passed in 0.11s | 7636e8e38b2d… | verify → |
| falsified | httpx | httpx-99cba6a | claimed fixed → test command failed1 failed, 7 passed in 0.06s | 28d710ff423c… | verify → |
| falsified | httpx | httpx-cabd1c0 | claimed fixed → test command failed2 failed, 24 passed in 0.10s | 92b801359010… | verify → |
| falsified | httpx | httpx-f653b2f | claimed fixed → test command failed4 failed, 31 passed in 0.14s | 6e0b4e6983f8… | verify → |
| falsified | jinja | jinja-d5f49f5c | claimed fixed → test failed4 failed, 29 passed in 0.08s | ba27b0c49a35… | verify → |
| falsified | rich | rich-7ef2d05c | claimed fixed → test failed1 failed, 7 passed in 0.14s | f86a8a229175… | verify → |
| falsified | rich | rich-95fe8ff5 | claimed fixed → test failed1 failed, 7 passed in 0.07s | 4d70a5424c75… | verify → |
| falsified | sympy/sympy | sympy__sympy-24909 | claimed fixed → test failedFAILED sympy/physics/units/tests/test_prefixes.py::test_prefix_operations - A... | 81f999ebd7ae… | verify → |
Each receipt is self-verifying: open one and your own browser re-checks the content hash and the Ed25519 signature. No server involved.
agents --fabrication
Agent fabrication, measured
A different shape from the repo catches above. We ran deepseek-flash v4.1 through a multi-turn pipeline with a null injected mid-step, one hundred eighty times. The agent flagged the null in its transcript, then wrote a bare number to the output file anyway. It fabricated on 40.0% of the null runs. The self-report and the effect diverged.
falsified · agent fabrication →falsified · agent fabrication (rate) →falsified · agent fabrication (rate) →
machine-readable
Consume the ledger
The whole ledger is available as JSON. Each entry carries its content address, signature, and public key, so it can be verified without contacting us.
the --key
Public key
One Ed25519 identity signs every receipt. Key ID 251b14346af43d1d.