Topic
Validating one implementation by executing an identical input on a reference implementation and comparing observable effects instead of recorded expectations.
Megapixel99's assay scan pairs duplicate functions whose outputs match on one fixed input ladder, a check it can apply to roughly a tenth of functions. A match means only that no probe split the pair, so the tool fails the run and hands it to a person.
Reality
A dev.to writeup attributes 612.9M tokens and $28.35 to four days of DeepSeek Harness work on a C# arbitrary-precision library, and the part worth reading is how the agent checked its own rewrite.
Reality
The project reports 2,931,581,192 matching checks over the full ChEMBL 37 structure table with zero blocking mismatch against a pinned RDKit 2026.03.1, and the pin is the line that sets your upgrade bill.
Reality
A dev.to walkthrough freezes 2050 fixture outputs behind a hash file and replays them byte for byte against the candidate binary. The gate itself is cheap and mostly right. The corpus is doing more of the work than the script.
Reality
The tfcompat harness runs the same HCL program through tofu and pulumi, then compares what the providers saw. The oracle is tofu, not a recorded expectation.
Reality