29 September 2026 · Agentic game · browser FPS
Whiteout Protocol · TensorFold 0.3.6.3
Completed a separate Whiteout Protocol implementation and build with TensorFold.
What was checked
- Production build passed; the JavaScript bundle was 623624 bytes and gzip output was 165179 bytes.
- A 75-second smoke test reported zero console errors.
- Aim-down-sights self-tests passed at FOV 75, 65, and 55 with zero-pixel offset.
- Gameplay checks covered deployment, combat, seven kills, wave clear, score tally, wave 2, and restart.
- Two post-feedback browser runs completed the menu-to-game transition without console errors.
Configuration
The model, serving engine and agent used for this execution.
| Field | Value | Detail |
|---|---|---|
| Model · verifier | Qwen 3.8 27B · verifier | 4-bit · 4 bpw |
| Checkpoint · verifier | Qwen3.8-27B MLX-Serve 4-bit | verifier model artifact |
| Model · drafter | Qwen 3.8 27B · drafter | Q4 · 4 bpw |
| Runtime | TensorFold 0.3.6.3 · source commit 1911880 | 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 | Whiteout Protocol · revision 1 | Same canonical challenge as the Sushi baseline. This run spanned 2026-09-29 to 2026-09-30 and included a long backend outage. |
| Context window | 131,072 tokens | maximum context configured for the serving runtime |
| 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 |
|---|---|---|
| Decode · median | 22.9 tok/s | median · n=273 |
| Decode · 10th percentile | 17 tok/s | 10th percentile · n=273 |
| Decode · 90th percentile | 35.8 tok/s | 90th percentile · n=273 |
| Decode · peak | 59.3 tok/s | peak · n=273 |
| TTFT · cached median | 7.8 s | median · n=273 |
| TTFT · cached 90th percentile | 20.9 s | 90th percentile · n=273 |
| Cold prefill · worst observed | 218.9 s | worst observed · n=1 |
| Peak KV checkpoint | 6.23 GiB | peak · n=1 |
| Resident model weights | 20 GiB | approximate · n=1 |
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 | 849.6 min | first user message to final report; includes outage · n=1 |
| Backend outage | 428.6 min | total · n=1 |
| Elapsed excluding outage | 421.0 min | effective elapsed · n=1 |
| Requests retained after relaunch | 273 requests | post-relaunch retained · n=1 |
| Backend errors before relaunch | 4 errors | total · n=1 |
| Backend errors after relaunch | 0 errors | total · n=1 |
| Tool calls | 368 calls | total · n=1 |
| Context compactions | 5 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 | 114,700 tokens | approximate maximum as reported · n=1 |
Recorded check counters
Counts from the checks recorded with this run.
| Metric | Value | Definition |
|---|---|---|
| Smoke test duration | 75 s | duration · n=1 |
| Console errors | 0 errors | smoke test · n=1 |
Request records & provenance
The source report retains 273 successful post-relaunch request measurements in aggregate, but the public archive does not contain per-request rows. Original prompt, prompt fingerprint, session transcript, server logs, and local paths are excluded.
Run ID: 2026-09-29-qwen38-27b-tensorfold-whiteout-protocol
One execution, with checks transcribed from the retained record. A completed session is not a quality score. Measurement method · All runs