[danielsmithdevelopment]

Summary

publishable: true with host disclosed. Live marketing board: /benchmarks#program-density (this PR). Ready for Infinity / Shadaj review before any “denser than” promotion.

Claim scope (important)

With matched 20-turn histories under the same 8 GiB cap: ClawQL’s resident history store fits ~29% more agents (100k vs 77.5k). This is a memory / history-storage result — not “the full ClawQL runtime runs 29% more agents.” Infinity’s arm is full agent_scale; ClawQL’s is the matched-shape history harness. Outreach draft: docs/benchmarks/program-density/infinity-on-pi/outreach-shadaj-draft.md.

Headline

Arm	Latency-gated N	RSS/N
Infinity agent_scale	77 500	~105.3 KiB
ClawQL matched-shape resident	100 000	~78.8 KiB

Fit quality: within-system ratios (Infinity p99 1.04×, ClawQL 1.69×). Criteria frozen in evidence as frozenCriteria.

/benchmarks page (this update)

• New #program-density section: chips, fit table, latency ratios, reclaim, open-host slopes, offloaded trade-off, honesty notes, artifact links
• Hero callout + jump link (history density ~29%)
• Dominance roadmap Infinity row + docs index / STATUS pointers
• Cross-links from gateway + governance boards

Hero History density callout
Program density fit table
benchmarks-program-density-demo.mp4

Host

Intel(R) Xeon(R) Processor, family 6 / model 207 / stepping 2 @ 2400 MHz, 4 vCPU, KVM — likely Emerald Rapids (guest cannot confirm SKU). DMI product_name unavailable.

Ask for Shadaj’s team

Allocator analog to --max-old-space-size=8192? Fuller ClawQL arm if they want runtime parity?

Test plan

• Claim scoped to history-memory density
• frozenCriteria in evidence
• Likely Emerald Rapids noted
• Outreach draft with candor
• /benchmarks#program-density section + hero jump + dominance/docs pointers
• apps/www static build + local serve smoke (section HTML + UI screenshots/demo)
• CI green
• Shadaj peer review before promotion

_{To show artifacts inline, enable in settings.}