Results / Qwen3.8 27B pi IQ2_XS (llama.cpp)
Qwen3.8 27B pi IQ2_XS (llama.cpp)
pillamacppomarchy (RTX 3060 12 GB, dual-GPU host)scene
The game opens in a new tab. It runs sandboxed, so a game that saves to browser storage may not keep its save.
Metrics
- Checklist score
- 100.0 / 100
- Design (rubric, 0–5)
- –
- Stop reason
- done
- Model time
- 29mins
- Tool time (wall minus model)
- 1min
- Wall time
- 30mins
- Tool calls
- –
- Tool errors
- –
- Errors by type
- –
- Loop-guard events
- –
- Tokens in / out
- 703,215.0 / 29,002.0
- Thinking tokens
- –
- Decode / prefill tok/s
- 17.0 / 347.0
- Prefix reuse
- 0.971
- Peak VRAM (GPU / engine)
- 3 / 11,156 MiB
- Peak engine RAM
- 12,862 MiB
- Engine
- llamacpp
- Model
- Qwen3.8 27B pi IQ2_XS (llama.cpp) (qwen38-27b-pi-iq2xs)
- Settings
- {"thinking": "medium", "reasoning_budget_tokens": 3072, "max_tokens": 7168, "sampling": "see research 2026-10-08-model-qwen38-27b-pi.md (server)", "note": "harness model id is a Strata alias name; llama-server serves its one model whatever the request names"}
- Node
- omarchy (RTX 3060 12 GB, dual-GPU host)
- Prompt
- scene-pagoda-v1
- Started
- 2026-10-30T06:26:28Z
Feature scores
- L
- 100.0
- R
- 100.0
- M
- 100.0
- C
- 100.0
- P
- 100.0
Screenshots
Landing page at desktop and phone widths, and the game at noon, at night, after breaking a block, and on a phone.
Checklist (6 pass, 0 partial, 0 fail)
| Feature | Test | Score | Evidence |
|---|---|---|---|
| L | page_loads | 1 | {"page": "index.html", "load_error": null, "errors": [], "failed_requests": []} |
| R | canvas_present | 1 | {"canvases": 1, "largest_frac_of_viewport": 1} |
| R | non_blank | 1 | {"luma_std": 32.8, "luma_mean": 205.7, "busy_cells_of_144": 64, "distinct_colours_4bit": 535} |
| M | animated_or_orbit | 1 | {"diff_time_2s_6s": 11.64, "diff_after_drag": 14.24, "diff_after_zoom": 17.54, "animated": true, "responds_to_drag": true, "responds_to_wheel": true} |
| C | colour_variety | 1 | {"hue_buckets_of_12": 7, "saturated_px": 3575, "hist": [39, 261, 113, 1220, 1598, 232, 17, 90, 0, 0, 0, 5]} |
| P | startup_cost | 1 | {"own_script_ms": 11, "top_functions": ["loop=5", "(anon)=4", "V=1"], "pass_below_ms": 1500, "fail_above_ms": 5000} |
Suite scene-1.1.0 · Chrome/155.0.8059.84 · WebGL: ANGLE (Intel, Vulkan 1.4.354 (Intel(R) Graphics (RKL GT1) (0x00004C8A)), Intel open-source Mesa driver)
Steers
None: an unsteered run gets the prompt and nothing else.