bm25_search_go

GitHub
mediumruntimeGo
results

Measured by wall-clock runtime in seconds — lower is better.

2.1s0.03s
sota
Claude-Opus-4.6
reward 0.830

Usage

Run the reference answer to verify your environment is set up correctly

$ harbor run -p tasks/bm25_search_go

Test model

$ harbor run -p tasks/bm25_search_go \
  -a claude-code -m claude-opus-4-6

Description

Optimize a Go search engine that computes exact BM25 top-10 results for each query over a deterministic synthetic corpus. Both corpus generation and query execution are included in the measured time. Goroutines are allowed but only standard library packages may be used.

Files

path
permission
/app/engine.go✎ Edit
/app/tokenize.go✎ Edit
/app/score.go✎ Edit
/app/rank.go✎ Edit
/app/batch.go✎ Edit
/app/corpus.go✎ Edit
/app/types.go✎ Edit
/app/main.goRead-only

Rules

  • 01Standard library only. No external modules.
  • 02Single-process only. Goroutines are allowed.
  • 03Wrong checksum or hit count = score 0.

Tags

gosearchbm25inverted-indexranking

Model Results

Click a row to view its trajectory in Live Lab

model
reward
score
Claude-Opus-4.6
0.830
Kimi-K2.6
0.640
DeepSeek-V4-Pro
0.600
Qwen-3.6-Plus
0.560
Gemini-3.1-Pro
0.540
MiMo-V2.5-Pro
0.510
GLM-5
0.510
Grok-4-20
0.510
GPT-5.4
0.500
MiniMax-M2.7
0.500
Hunyuan-3-Preview
0.320
<- all tasks