2 October 2026 · Agentic game · isometric 3D
Digger 3D · Sushi 1.1.0
Completed a deterministic Digger game with grid simulation and isometric Three.js presentation.
What was checked
- TypeScript check and Vite production build passed.
- 28 of 28 self-tests passed.
- All five levels passed connectivity, emerald-access and bag-support validation.
- Playwright title/game capture completed; no console errors were reported.
Limits of this evidence
TTFT and peak process/system RAM were not recorded. Prefill includes cache reuse and is not a cold-prefill measurement. Functional checks are transcribed from the run report and were not rerun during archival.
Configuration
The model, serving engine and agent used for this execution.
| Field | Value | Detail |
|---|---|---|
| Model · primary | Qwen 3.8 Flash Next · primary | EXL3 2.625 bpw |
| Runtime | Sushi 1.1.0 · MLX 0.32.3 | inference engine and exact build |
| Harness | Pi coding agent | the agent or client that drove the run |
| Hardware | Apple M5 Pro · 64 GB unified memory | hardware recorded for this execution |
| Workload | Digger 3D · revision 1 | Long-horizon coding-agent challenge; not directly comparable with a fixed task suite. |
| Repetitions | 1 | executions recorded for this workload; one means an indicative result, not an estimate |
Inference measurements
Engine-side rates for this run. Each figure states its own statistic; figures from different runs are not interchangeable.
| Metric | Value | Definition |
|---|---|---|
| Cache hit ratio | 93.79 % | aggregate · n=1 |
| Decode · per-request median | 39.45 tok/s | median · n=128 |
| Decode · output-weighted | 35.99 tok/s | total output tokens / summed per-request decode time · n=128 |
| Prefill · median | 456.15 tok/s | median; includes cache reuse · n=128 |
| Prefill · mean | 418.91 tok/s | mean; includes cache reuse · n=128 |
| MTP acceptance | 73.48 % | accepted draft tokens / drafted tokens · n=122523 · 90,036 accepted out of 122,523 drafted tokens. |
Agent session counters
Agent-wide usage totals. These include reasoning, tool calls and waiting, and are not model decode speed.
| Metric | Value | Definition |
|---|---|---|
| Session elapsed | 117.8 min | first instruction to final response · n=1 |
| Session start to end | 118.0 min | elapsed · n=1 |
| Matched model requests | 128 requests | matched · n=1 |
| Tool calls | 171 calls | total · n=1 |
| Context compactions | 2 compactions | total · n=1 |
Context and token counters
Token totals as reported by the harness or engine. Cached and cumulative figures are not additive with fresh input.
| Metric | Value | Definition |
|---|---|---|
| Maximum prompt context | 113,288 tokens | maximum · n=1 |
| Output tokens | 194,071 tokens | total · n=1 |
| Fresh input | 502,196 tokens | total · n=1 |
| Cached input | 7,582,258 tokens | total · n=1 |
Recorded check counters
Counts from the checks recorded with this run.
| Metric | Value | Definition |
|---|---|---|
| Self-tests passed | 28 tests | total · n=1 |
| Levels validated | 5 levels | total · n=1 |
Request records & provenance
128 matched requests in run order, paired monotonically by exact prompt/output token counts. Includes numeric token counts, throughput and MTP counters only; raw prompts/responses, event timestamps and server-log row IDs are omitted.
Run ID: 2026-10-02-qwen38-flash-next-sushi-digger-3d
One execution, with checks transcribed from the retained record. A completed session is not a quality score. Measurement method · All runs