01 · ZEUSLOBSTER
Hermes Base Board
Base, observed. Not assumed.
Cost-aware research for humans and AI agents. A working public, read-only proof of concept.
Erik Staebell · Founder | September 2026
02 · THE PROBLEM
Activity is not evidence of opportunity.
A visible spread is only the beginning.
Fees, liquidity, gas, latency and attribution can change what a signal means. Raw dashboards leave users to assemble those checks themselves.
Who we are learning from: Base researchers, builders and agent developers.
03 · THE PRODUCT
One evidence foundation.
Observe. Investigate. Explain.
Hermes Base Board combines historical research with continuously monitored stablecoin, liquidity, payment and contract activity. Research Pro adds a CLDv4-gated research workspace and Stable Trading Lab.
Read-only tools · Paper quotes · No trade execution or custody
04 · WORKING PROOF
Public and inspectable.
From a concept to a working research product.
Public Board and Research Pro pages are shipped. A representative August 24–31 study collected 502,261 transaction details across 2,240 sampled blocks and 56 windows.
Historical research counts—not customers, product transactions or a complete chain census.
05 · HOW IT WORKS
The spread is not the trade.
A disciplined evidence pipeline.
Observe the signal → check freshness and attribution → account for fees, impact and modeled gas → publish the result and limitations.
Negative findings are useful. Paper evidence does not guarantee executable profit.
06 · AGENT FORECAST ARENA
Put forecasts up against outcomes.
Our supporting agent-evaluation demonstration.
Recorded events compare Granite Sentinel, Qwen Coder Oracle, Gemma Calibrator and Qwen Scout using objective questions and reproducible Brier scoring. Lower loss is better.
Local-model benchmarks. External practice is unranked; open competition and run reporting remain roadmap work.
07 · EARLY VALIDATION
Engineering proof. Early feedback.
Adoption is the next question to answer.
The public product is available to explore. An external builder shared a custom Arena client for benchmark retrieval and local scoring. That feedback exposed a need for opt-in run acknowledgements.
We do not present that experiment as verified rankings, retention or established customer traction.
08 · BUSINESS MODEL
Find the workflow users return to.
Validate before scaling.
Potential paid offers include research reports, alerts, historical analysis and agent/API access. Research Pro currently checks 1+ CLDv4 ownership; recurring subscription revenue is not established.
CLDv4 access and the associated CLD/$CLAW ecosystem are disclosed separately from investment and product revenue claims.
09 · NEXT EIGHT WEEKS
Turn proof into repeat use.
A focused validation plan.
Weeks 1–2: design-partner interviews and workflow selection. Weeks 3–4: history, freshness and exports. Weeks 5–6: consenting user pilots and repeat-use measurement. Weeks 7–8: willingness-to-pay research and a paid-pilot target.
Proposed milestones—not completed work. Independent review and production readiness remain separate work.
10 · FOUNDER & ASK
Built firsthand.
Erik Staebell, supported by AI-assisted development.
Community moderation, AMAs and project operations led to hands-on building. Zeus is an AI development assistant, not a second human founder. We seek product guidance, design partners and funding to improve reliability, evidence delivery and distribution.
Base Batches: investment consideration. Creator Grant: public educational content drawn from the same product. No endorsement or selection implied.
EXPLORE THE PROOF
See it for yourself.
Guided product demonstration ↗
Explore recorded Arena benchmarks ↗
zeuslobster.com · zeus.lobster@gmail.com