Mulch Self Improver — Benchmark

Side-by-side vs Self Improving Agent — Rank #2 on ClawHub (legacy .learnings flow)

27.5%

Fewer chars (total)

3792

Mulch (chars)

5233

Baseline (chars)

~352

Tokens saved/session

1. Token efficiency (session + retrieval)

Metric	Baseline (legacy)	Mulch	Winner
Reminder (chars)	632	452	Mulch (shorter)
Session context (chars)	1318	1694	Baseline
Retrieval (2 queries, chars)	932	330	Mulch 65% less
Total (rem + ctx + ret)	2882	2476	Mulch 14% less

2. Troubleshooting (3 error lookups)

Metric	Baseline (legacy)	Mulch	Winner
Chars to get all 3 resolutions	1215	559	Mulch 54% less
Resolutions found (of 3)	3/3	2/3 or 3/3	Same or better

3. Style & memory (6 scenarios)

Metric	Baseline (legacy)	Mulch	Winner
Chars to get all 6 answers	One full file	Sum of 6 targeted searches	Mulch (targeted)
Scenarios found (of 6)	6/6	4/6–6/6	Same or better

4. Projected savings (chars ≈ token proxy)

Scenario	Baseline (chars)	Mulch (chars)	Saving	Per 100 sessions (tokens ≈ chars/4)
Session (rem + ctx + retrieval)	2882	2476	406 chars (~100 tokens)	~10k tokens
Troubleshooting (3 errors)	1215	559	656 chars (~164 tokens)	~16k tokens per 100 troubleshoot rounds

Run: docker run --rm mulch-self-improver-test benchmark. Credits: Mulch Self Improver by Austin Dixson; Mulch by Jaymin West.