Claims you can reproduce
gograph’s checked-in benchmark uses a controlled Go fixture with manually reviewable ground truth. It verifies implicit interface implementations, interface-dispatched callers, and a composed change-context workflow.
The current v1.5.5 report was generated with a complete precise graph. Every declared evidence item was found by the corresponding gograph workflow:
| Scenario | gograph evidence | Text-search evidence | Process calls |
|---|---|---|---|
| Implicit interface implementations | 2/2 | 0/2 | 1 vs 1 |
| Interface-dispatched caller | 2/2 | 1/2 | 1 vs 1 |
| Composed change context | 4/4 | 4/4 | 1 vs 3 |
What the benchmark does not claim
This is a deterministic structural-evidence benchmark. It does not claim a fixed token reduction, universal speedup, hallucination rate, or complete end-to-end coding-agent success rate. The fixture is small, elapsed times depend on the host, and text search returns different evidence.
Reproduce it
git clone https://github.com/ozgurcd/gograph
cd gograph
make build
make benchmark
The harness records the binary version, source revision, runner, suite, and fixture SHA-256 digests, graph precision and build health, complete raw outputs, evidence coverage, output size, process-call count, and repeated timing.
Review the suite and fixture, the complete v1.5.5 result, and the methodology.