Skip to content

latest.md

Latest commit

 

History

History
54 lines (41 loc) · 2.59 KB

latest.md

File metadata and controls

54 lines (41 loc) · 2.59 KB

Benchmark Results

  • Generated at: 2026-03-28T17:20:15.198620+00:00
  • Corpus: 17 Markdown files, 247 indexed blocks
  • Full index: 5553.57 ms
  • Idle incremental: 844.25 ms

Benchmark summary

At a Glance

Metric Result Why it matters
Corpus 17 files · 247 blocks This is measured on the real repository corpus
Full index 5.55 s First-time indexing is short
Idle incremental 0.84 s Re-indexing after small changes is light
Avg semantic_search 68.13 ms Fast enough to use as the default retrieval path
Avg hydrate 41.63 ms Opening more context stays cheap
Search-only byte savings 63.96% Much less text is sent to the model
Search + hydrate byte savings 44.1% Even the guided path stays clearly smaller than opening full files

What These Results Mean

  • The compact retrieval path is much smaller than naive full-file opening.
  • Even after hydrating the top hit, the guided path still saves a lot of context.
  • More of the model's context window stays available for reasoning instead of repeated reading.

Aggregate Savings

Strategy Average byte savings Median byte savings Average word savings
semantic_search only 63.96% 65.1% 75.02%
semantic_search + hydrate(top1) 44.1% 45.67% 63.51%

Query Breakdown

Query Top hit Full files bytes Search bytes Search+hydrate bytes Search savings Guided savings
namespace model project global hybrid hybrid 9151 5486 8409 40.05% 8.11%
hydrate bounded neighborhood related mode Hydration Strategy 22069 5963 8845 72.98% 59.92%
current project resolution git remote overrides Project Identity 17837 6102 9114 65.79% 48.9%
storage stats freshness index status Acceptance Criteria 15996 5913 9206 63.03% 42.45%
Claude Code Codex Cursor OpenCode integrations OpenCode 18582 6613 10905 64.41% 41.31%
remember note decision lesson handoff pattern Memory note 25147 5659 9072 77.5% 63.92%

Method

Strategy Why it matters
Baseline without MCP Open the full source text of every unique Markdown file represented in the top-5 project search hits
Compact MCP path Use the semantic_search response only
Guided MCP path Use semantic_search and then hydrate only for the top Markdown hit

Savings are measured against the baseline using real UTF-8 byte counts and whitespace-delimited word counts taken from this repository corpus.