EVIDENCE
Evidence from real repositories.
Transcripts, commands, artifacts, and before/after results β captured from Loctree running on actual codebases.
Loctree turns agent "done" into structural evidence
Loctree on Loctree itself: a before/after structural snapshot replaces an agent's claim of "done" with evidence reviewers can verify.
> Prompt: Verify agent's "done" β capture before/after structural diff
$ loct health
Dead exports: 8
Cycles: 0
...
> Action: Removed verified dead surfaces Β· captured after-state
$ loct health
Dead exports: 2
Known test-only remains βPROOF LEDGER
More exhibits.
A proof log of real sessions. Real diffs, reproducible commands, agent-verified outcomes.
BENCHMARKS Β· SCALE TARGETS
Loctree on production-grade trees.
Preview values are marked. Real numbers replace them as runs land.
SCAN TARGETSGo stdlibRust compilerCPythonNode.jsTauriTypeScript compiler
Go stdlib
- FILES
- 17,182
- SCAN
- preview
- SURFACES
- β
Pure Go, deeply nested packages. Scan target.
Rust compiler
- FILES
- 35,387
- SCAN
- preview
- SURFACES
- β
rustc tree β large workspace stress test.
CPython
- FILES
- preview
- SCAN
- preview
- SURFACES
- β
Python interpreter sources + stdlib.
Node.js
- FILES
- preview
- SCAN
- preview
- SURFACES
- β
Mixed JS + C++ β multi-language sensor exercise.
Tauri
- FILES
- preview
- SCAN
- preview
- SURFACES
- β
Rust β TS bridge. Command graph validation.
TypeScript compiler
- FILES
- preview
- SCAN
- preview
- SURFACES
- β
Self-hosting compiler tree, deep TS-only graph.
CONTRIBUTE
Open a PR. Submit a case. Add a benchmark.
Evidence works because it's reproducible. Bring yours.