Acc. XXV · Solitaria fugax — single-agent evaluation build, cultivated 2026
Sasuke vs Itachi — single-agent browser fight prototype
Systems R&D engineer · Agent evaluation · 2026 · Independent studio
A browser one-on-one fight prototype built end to end by a single agent in a single session with no harness, no gates, and no orchestration — the control case against the orchestrated, gated builds elsewhere in this catalogue.
Solitaria fugax — single-agent evaluation build, cultivated 2026
Field notes
The context
Every other entry in this catalogue runs through structured agent workflows — harnesses, verification gates, orchestration. The studio needed a control case: what does a capable but ungoverned single agent produce on the same class of problem, given one session and no scaffolding around it?
Cultivation
What I built
- Commissioned and ran the evaluation: a browser one-on-one fight prototype built entirely by Sakana AI Fugu in a single agent session.
- Deliberately withheld harness, verification gates, and orchestration scaffolding to isolate the agent's unassisted output.
- Reviewed the resulting build against the orchestrated, gated projects elsewhere in the lab to characterise the difference.
Root system
How it works
A single agent, single session, browser-based build with no external harness or gate structure — the deliberate absence of architecture is the point of the evaluation.
Harvest
Outcomes
- Produced a working comparison point: what a different agent product does with an interactive-systems brief absent any orchestration.
- This is a throwaway evaluation prototype — no live deployment, no ongoing development.
- The working repository is not currently available for inspection.
Herbarium
The work itself




An orchestrated, gated workflow's value is easiest to see in contrast — this single-session, ungoverned build is what the harness and verification gates elsewhere in the studio's projects are measured against.