WebVM S1L 단일 커널 벤치 증거 추가

Author: eddmpython

docs/operations/benchmarking.md

@@ -51,7 +51,7 @@ S1은 현재 공개 속도 간판이다. S1L은 외부 후보가 S1의 병렬 wo
 | S0 python ready latency | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0-pyproc-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0-webvm-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0-jupyterlite-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0-marimo-wasm-2026-07-15.json) | 페이지 또는 런타임 시작부터 첫 Python 명령 성공까지. 조건은 각 artifact note를 따른다 |
 | S0C python cold ready latency | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0c-pyproc-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0c-webvm-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0c-jupyterlite-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s0c-marimo-wasm-2026-07-15.json) | cold profile/cache-clear 조건. warm S0와 한 표에 섞지 않음 |
 | S1 numpy sharded matmul | Speed Lab 반복 봉투 | N/A 가능 | N/A 가능 | N/A 가능 | 병렬 worker 모델이 다르면 single-lane으로 재정의하지 않음 |
-| S1L single-kernel numpy latency | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-pyproc-2026-07-15.json) | 측정 필요 | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-jupyterlite-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-marimo-wasm-2026-07-15.json) | S1 대체가 아니라 별도 single-lane 보조 축 |
+| S1L single-kernel numpy latency | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-pyproc-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-webvm-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-jupyterlite-2026-07-15.json) | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s1l-marimo-wasm-2026-07-15.json) | S1 대체가 아니라 별도 single-lane 보조 축 |
 | S2 process map | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s2-pyproc-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s2-webvm-na-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s2-jupyterlite-na-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s2-marimo-wasm-na-2026-07-15.json) | `PyProc.map` process pool 계약. 같은 API 후보가 없으면 N/A |
 | S3 browser server | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s3-pyproc-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s3-webvm-na-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s3-jupyterlite-na-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s3-marimo-wasm-na-2026-07-15.json) | pyproc `VirtualOrigin`/ASGI Service Worker URL contract. 같은 계약 후보가 없으면 N/A |
 | S4 machine resume | [측정됨](../../mainPlan/browser-os-north-star/benchmarks/s4-pyproc-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s4-webvm-na-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s4-jupyterlite-na-2026-07-15.json) | [N/A](../../mainPlan/browser-os-north-star/benchmarks/s4-marimo-wasm-na-2026-07-15.json) | signed `.pymachine`, trusted open, `resume.py` resource reopen. 같은 계약 후보가 없으면 N/A |

docs/operations/testing.md

@@ -139,7 +139,8 @@ S1L single-kernel NumPy latency는 S1을 대체하지 않는 별도 보조 축
 
 ```bash
 npm run bench:artifact -- --scenario S1L --candidate pyproc --command "<command>" --sample 10067,0 --sample 9633,0 --sample 10073,0 --out .tmp/pyproc-s1l.json
-npm run bench:compare -- .tmp/pyproc-s1l.json --out .tmp/s1l-compare.md
+npm run bench:artifact -- --scenario S1L --candidate webvm --command "WebVM terminal S1L" --sample 12825,0 --sample 11406,0 --sample 11349,0 --out .tmp/webvm-s1l.json
+npm run bench:compare -- .tmp/pyproc-s1l.json .tmp/webvm-s1l.json --out .tmp/s1l-compare.md
 ```
 
 체크리스트:

mainPlan/browser-os-north-star/03-progress-ledger.md

@@ -2195,3 +2195,44 @@ NEXT:
 1. WebVM의 S1L 또는 Python shell 단일 계산 latency를 분리할 가치가 있는지 판정한다.
 2. README benchmark 섹션에 S0-S4를 모두 올릴 경우, 숫자보다 계약 차이를 먼저 설명한다.
 3. 다음 구조 강화는 benchmark artifact schema v2 또는 product-facing capability matrix 중 하나로 잡는다.
+
+## 2026-07-15 - WebVM S1L single-kernel benchmark 합류
+
+완료:
+
+- WebVM terminal에서 Python 3.7.3과 NumPy 사용 가능 여부를 확인했다.
+- WebVM을 S1L single-kernel NumPy latency 후보로 측정했다.
+- [s1l-webvm-2026-07-15.json](benchmarks/s1l-webvm-2026-07-15.json)을 생성했다.
+- [s1l-compare-2026-07-15.md](benchmarks/s1l-compare-2026-07-15.md)를 pyproc, WebVM, JupyterLite, marimo WASM 4자 표로 갱신했다.
+- [06-speed-comparison.md](06-speed-comparison.md), [benchmarking.md](../../docs/operations/benchmarking.md), [testing.md](../../docs/operations/testing.md), [externalS1](../../tests/attempts/externalS1/README.md)의 S1L 상태를 갱신했다.
+
+실측:
+
+- Browser: Edge `150.0.4078.65`.
+- WebVM URL: `https://webvm.io/`.
+- WebVM runtime: Python `3.7.3`, NumPy import 가능.
+- 작업: WebVM terminal에서 1024x1024 f64 NumPy matmul을 `time.perf_counter()`로 측정. 128x128 warmup 뒤 3 samples. NumPy import와 matrix construction은 sample timing에서 제외했다.
+- 유효 sample: 12825ms, 11406ms, 11349ms.
+- checksum은 세 sample 모두 `3.2149321267`로 동일했다.
+- WebVM S1L median: 11406ms, p95 12825ms, min 11349ms, max 12825ms, maxErr 0.
+- artifact 생성 commit: `e8ecb78025d7e8e034ef5bf075f7d41647f2812d`, `worktreeDirty: false`.
+
+검증:
+
+- WebVM terminal probe: `PY_READY 3.7.3`, `NUMPY_SPEC True`.
+- WebVM S1L run: `PYPROC_S1L_SAMPLE` 3개와 `PYPROC_S1L_DONE` 확인.
+- `npm run bench:artifact -- --scenario S1L --candidate webvm --browser-version 150.0.4078.65 --engine "WebVM/CheerpX Python 3.7.3 NumPy" --source "https://webvm.io/ via playwright-cli" --command "WebVM terminal, Python perf_counter S1L 1024 f64 numpy matmul, 128 warmup, 3 samples" --note "warmed WebVM session; numpy import and matrix construction excluded from sample timings; checksum stayed 3.2149321267 across all samples" --sample 12825,0 --sample 11406,0 --sample 11349,0 --out mainPlan/browser-os-north-star/benchmarks/s1l-webvm-2026-07-15.json` PASS.
+- `npm run bench:compare -- mainPlan/browser-os-north-star/benchmarks/s1l-pyproc-2026-07-15.json mainPlan/browser-os-north-star/benchmarks/s1l-webvm-2026-07-15.json mainPlan/browser-os-north-star/benchmarks/s1l-jupyterlite-2026-07-15.json mainPlan/browser-os-north-star/benchmarks/s1l-marimo-wasm-2026-07-15.json --out mainPlan/browser-os-north-star/benchmarks/s1l-compare-2026-07-15.md` PASS.
+
+판정:
+
+- WebVM S1L은 측정 가치가 있었다. 단, 이것은 WebVM이 S1을 충족한다는 뜻이 아니다.
+- S1은 여전히 pyproc의 4-worker sharded NumPy matmul 계약이고, WebVM/JupyterLite/marimo WASM은 S1에서는 N/A다.
+- S1L은 single-kernel 보조 축이다. 4자 S1L 결과는 marimo WASM 9355ms, pyproc 10067ms, JupyterLite 10149ms, WebVM 11406ms다.
+- 이로써 S0, S0C, S1L은 4자 실측 표가 닫혔고, S1/S2/S3/S4는 pyproc OS 계약 + 외부 N/A 봉인 구조가 닫혔다.
+
+NEXT:
+
+1. pyproc의 공개 속도 간판은 S1 병렬 worker pool로 유지한다.
+2. README benchmark 섹션에 S0-S4와 S1L을 올릴 경우 숫자보다 계약 차이를 먼저 설명한다.
+3. 다음 구조 강화는 benchmark artifact schema v2 또는 product-facing capability matrix 중 하나로 잡는다.

mainPlan/browser-os-north-star/06-speed-comparison.md

@@ -1,6 +1,6 @@
 # 06. 속도 정면 비교 계약
 
-상태: pyproc S0/S1/S2/S3/S4 기준 artifact 기록. S0와 S0C는 pyproc, WebVM, JupyterLite, marimo WASM을 같은 표에 합쳤다. WebVM, JupyterLite, marimo WASM은 같은 S1, S2, S3, S4 계약을 N/A artifact로 봉인. S1L은 pyproc, JupyterLite, marimo WASM을 같은 브라우저에서 측정.
+상태: pyproc S0/S1/S2/S3/S4 기준 artifact 기록. S0, S0C, S1L은 pyproc, WebVM, JupyterLite, marimo WASM을 같은 표에 합쳤다. WebVM, JupyterLite, marimo WASM은 같은 S1, S2, S3, S4 계약을 N/A artifact로 봉인.
 
 ## 목표
 
@@ -42,6 +42,7 @@
 | 2026-07-15 | S1 pyproc | [s1-pyproc-2026-07-15.json](benchmarks/s1-pyproc-2026-07-15.json) | [s1-compare-2026-07-15.md](benchmarks/s1-compare-2026-07-15.md) | GREEN, `size=1024`, 3 samples, median 3.95x, shard p95 2606ms, maxErr 0 |
 | 2026-07-15 | S1 external candidates | [webvm N/A](benchmarks/s1-webvm-na-2026-07-15.json), [jupyterlite N/A](benchmarks/s1-jupyterlite-na-2026-07-15.json), [marimo-wasm N/A](benchmarks/s1-marimo-wasm-na-2026-07-15.json) | [s1-compare-2026-07-15.md](benchmarks/s1-compare-2026-07-15.md) | 같은 4-worker sharded NumPy matmul 계약 없음 |
 | 2026-07-15 | S1L pyproc | [s1l-pyproc-2026-07-15.json](benchmarks/s1l-pyproc-2026-07-15.json) | [s1l-compare-2026-07-15.md](benchmarks/s1l-compare-2026-07-15.md) | GREEN, 3 samples, median 10067ms, p95 10073ms, maxErr 0 |
+| 2026-07-15 | S1L WebVM | [s1l-webvm-2026-07-15.json](benchmarks/s1l-webvm-2026-07-15.json) | [s1l-compare-2026-07-15.md](benchmarks/s1l-compare-2026-07-15.md) | GREEN, 3 samples, median 11406ms, p95 12825ms, maxErr 0 |
 | 2026-07-15 | S1L JupyterLite | [s1l-jupyterlite-2026-07-15.json](benchmarks/s1l-jupyterlite-2026-07-15.json) | [s1l-compare-2026-07-15.md](benchmarks/s1l-compare-2026-07-15.md) | GREEN, 3 samples, median 10149ms, p95 10153ms, maxErr 0 |
 | 2026-07-15 | S1L marimo WASM | [s1l-marimo-wasm-2026-07-15.json](benchmarks/s1l-marimo-wasm-2026-07-15.json) | [s1l-compare-2026-07-15.md](benchmarks/s1l-compare-2026-07-15.md) | GREEN, 3 samples, median 9355ms, p95 11424ms, maxErr 0 |
 | 2026-07-15 | S2 pyproc | [s2-pyproc-2026-07-15.json](benchmarks/s2-pyproc-2026-07-15.json) | [s2-compare-2026-07-15.md](benchmarks/s2-compare-2026-07-15.md) | GREEN, 3 samples, serial median 73ms, process pool median 43ms, speedup 1.61x, maxErr 0 |
@@ -60,7 +61,7 @@
 | S0 python ready latency | [artifact](benchmarks/s0-pyproc-2026-07-15.json) | [artifact](benchmarks/s0-webvm-2026-07-15.json) | [artifact](benchmarks/s0-jupyterlite-2026-07-15.json) | [artifact](benchmarks/s0-marimo-wasm-2026-07-15.json) | pyproc 3471ms, WebVM 3472ms, marimo WASM 8385ms, JupyterLite 12352ms. 조건은 각 artifact note를 따른다 |
 | S0C python cold ready latency | [artifact](benchmarks/s0c-pyproc-2026-07-15.json) | [artifact](benchmarks/s0c-webvm-2026-07-15.json) | [artifact](benchmarks/s0c-jupyterlite-2026-07-15.json) | [artifact](benchmarks/s0c-marimo-wasm-2026-07-15.json) | pyproc 3660ms, marimo WASM 10136ms, JupyterLite 11796ms, WebVM 46534ms. cold profile/cache-clear 전용. S0 warm/observed 표와 섞지 않음 |
 | S1 numpy sharded matmul | `npm run bench:speed -- --out <path>` | [N/A](benchmarks/s1-webvm-na-2026-07-15.json) | [N/A](benchmarks/s1-jupyterlite-na-2026-07-15.json) | [N/A](benchmarks/s1-marimo-wasm-na-2026-07-15.json) | pyproc만 같은 S1 계약 충족 |
-| S1L single-kernel numpy latency | [artifact](benchmarks/s1l-pyproc-2026-07-15.json) | 미측정 | [artifact](benchmarks/s1l-jupyterlite-2026-07-15.json) | [artifact](benchmarks/s1l-marimo-wasm-2026-07-15.json) | WebVM 제외 3자 측정 완료 |
+| S1L single-kernel numpy latency | [artifact](benchmarks/s1l-pyproc-2026-07-15.json) | [artifact](benchmarks/s1l-webvm-2026-07-15.json) | [artifact](benchmarks/s1l-jupyterlite-2026-07-15.json) | [artifact](benchmarks/s1l-marimo-wasm-2026-07-15.json) | 4자 측정 완료. S1 대체가 아니라 single-lane 보조 축 |
 | S2 process map | [artifact](benchmarks/s2-pyproc-2026-07-15.json) | [N/A](benchmarks/s2-webvm-na-2026-07-15.json) | [N/A](benchmarks/s2-jupyterlite-na-2026-07-15.json) | [N/A](benchmarks/s2-marimo-wasm-na-2026-07-15.json) | pyproc serial 73ms, process pool 43ms, median speedup 1.61x. 외부 후보는 같은 `PyProc.map` process pool API 계약 없음 |
 | S3 browser server | [artifact](benchmarks/s3-pyproc-2026-07-15.json) | [N/A](benchmarks/s3-webvm-na-2026-07-15.json) | [N/A](benchmarks/s3-jupyterlite-na-2026-07-15.json) | [N/A](benchmarks/s3-marimo-wasm-na-2026-07-15.json) | pyproc `VirtualOrigin` POST roundtrip median 18ms, p95 18ms. 외부 후보는 같은 pyproc `VirtualOrigin`/ASGI Service Worker URL contract 없음 |
 | S4 machine resume | [artifact](benchmarks/s4-pyproc-2026-07-15.json) | [N/A](benchmarks/s4-webvm-na-2026-07-15.json) | [N/A](benchmarks/s4-jupyterlite-na-2026-07-15.json) | [N/A](benchmarks/s4-marimo-wasm-na-2026-07-15.json) | pyproc export 76ms, open 2264ms, image 10.8MB, resume rows 2-2. 외부 후보는 같은 pyproc signed `.pymachine`/trusted open/`resume.py` resource reopen contract 없음 |
@@ -92,7 +93,8 @@ S1L artifact는 S1을 single-lane으로 바꿔치기하지 않기 위한 보조
 ```bash
 npm run bench:artifact -- --scenario S1L --candidate jupyterlite --command "manual single-kernel S1L run" --sample 9844,0 --sample 10149,0 --sample 10153,0 --out .tmp/jupyterlite-s1l.json
 npm run bench:artifact -- --scenario S1L --candidate marimo-wasm --command "manual single-kernel S1L run" --sample 11424,0 --sample 9355,0 --sample 9239,0 --out .tmp/marimo-wasm-s1l.json
-npm run bench:compare -- .tmp/pyproc-s1l.json .tmp/jupyterlite-s1l.json .tmp/marimo-wasm-s1l.json --out .tmp/s1l-compare.md
+npm run bench:artifact -- --scenario S1L --candidate webvm --command "WebVM terminal S1L run" --sample 12825,0 --sample 11406,0 --sample 11349,0 --out .tmp/webvm-s1l.json
+npm run bench:compare -- .tmp/pyproc-s1l.json .tmp/webvm-s1l.json .tmp/jupyterlite-s1l.json .tmp/marimo-wasm-s1l.json --out .tmp/s1l-compare.md
 ```
 
 S2 artifact는 `test:browser`의 `PyProc.map` timing을 paired sample로 승격한다. 같은 process pool에서 같은 함수와 입력을 serial과 parallel로 돌린 wall time만 받는다.
@@ -129,6 +131,6 @@ npm run bench:compare -- .tmp/pyproc-s4.json .tmp/webvm-s4-na.json --out .tmp/s4
 
 ## 다음 작업
 
-1. WebVM의 S1L 또는 Python shell 단일 계산 latency를 분리할 가치가 있는지 판정한다.
-2. pyproc의 속도 간판은 S1 병렬 worker pool로 유지하고, README 속도 문구는 이 비교 계약을 통과한 숫자만 갱신한다.
-3. S0-S4 비교 축을 제품 README에 올릴 경우 숫자보다 계약 차이를 먼저 설명한다.
+1. pyproc의 속도 간판은 S1 병렬 worker pool로 유지하고, README 속도 문구는 이 비교 계약을 통과한 숫자만 갱신한다.
+2. S0-S4와 S1L 비교 축을 제품 README에 올릴 경우 숫자보다 계약 차이를 먼저 설명한다.
+3. 다음 구조 강화는 benchmark artifact schema v2 또는 product-facing capability matrix 중 하나로 잡는다.

mainPlan/browser-os-north-star/benchmarks/s1l-compare-2026-07-15.md

@@ -3,3 +3,4 @@
 | marimo-wasm | GREEN | 3 | 9355 | 11424 | 9239 | 11424 | 0 | 150.0.4078.65 | 4e30f3a8 | mainPlan/browser-os-north-star/benchmarks/s1l-marimo-wasm-2026-07-15.json |
 | pyproc | GREEN | 3 | 10067 | 10073 | 9633 | 10073 | 0 | 150.0.4078.65 | 5c867ee8 | mainPlan/browser-os-north-star/benchmarks/s1l-pyproc-2026-07-15.json |
 | jupyterlite | GREEN | 3 | 10149 | 10153 | 9844 | 10153 | 0 | 150.0.4078.65 | 98934a90 | mainPlan/browser-os-north-star/benchmarks/s1l-jupyterlite-2026-07-15.json |
+| webvm | GREEN | 3 | 11406 | 12825 | 11349 | 12825 | 0 | 150.0.4078.65 | e8ecb780 | mainPlan/browser-os-north-star/benchmarks/s1l-webvm-2026-07-15.json |

mainPlan/browser-os-north-star/benchmarks/s1l-webvm-2026-07-15.json

@@ -0,0 +1,51 @@
+{
+  "schemaVersion": 1,
+  "scenario": "S1L",
+  "candidate": "webvm",
+  "name": "single-kernel numpy matmul latency",
+  "command": "WebVM terminal, Python perf_counter S1L 1024 f64 numpy matmul, 128 warmup, 3 samples",
+  "commit": "e8ecb78025d7e8e034ef5bf075f7d41647f2812d",
+  "worktreeDirty": false,
+  "startedAt": "2026-07-15T03:07:45.194Z",
+  "finishedAt": "2026-07-15T03:07:45.195Z",
+  "browser": {
+    "path": null,
+    "version": "150.0.4078.65"
+  },
+  "host": {
+    "platform": "win32",
+    "release": "10.0.26200",
+    "cpuModel": "AMD Ryzen 7 8845HS w/ Radeon 780M Graphics     ",
+    "cpuCount": 16,
+    "totalMemBytes": 33557934080,
+    "freeMemBytes": 17717940224
+  },
+  "engine": {
+    "name": "WebVM/CheerpX Python 3.7.3 NumPy"
+  },
+  "source": "https://webvm.io/ via playwright-cli",
+  "note": "warmed WebVM session; numpy import and matrix construction excluded from sample timings; checksum stayed 3.2149321267 across all samples",
+  "ok": true,
+  "metrics": {
+    "samples": [
+      {
+        "latencyMs": 12825,
+        "maxErr": 0
+      },
+      {
+        "latencyMs": 11406,
+        "maxErr": 0
+      },
+      {
+        "latencyMs": 11349,
+        "maxErr": 0
+      }
+    ],
+    "sampleCount": 3,
+    "medianMs": 11406,
+    "p95Ms": 12825,
+    "minMs": 11349,
+    "maxMs": 12825,
+    "maxErr": 0
+  }
+}

tests/attempts/externalS1/README.md

@@ -26,9 +26,10 @@ WebVM, JupyterLite, marimo WASM은 모두 브라우저 안 실행 표면을 제
 | 날짜 | probe | 환경 | 핵심 수치 | 결론 | 다음 |
 |---|---|---|---|---|---|
 | 2026-07-15 | 공식 문서 판정 + `bench:artifact --na` | Edge/Windows, 공식 문서 기반 | pyproc S1 기준 artifact 3.95x, [WebVM N/A](../../../mainPlan/browser-os-north-star/benchmarks/s1-webvm-na-2026-07-15.json), [JupyterLite N/A](../../../mainPlan/browser-os-north-star/benchmarks/s1-jupyterlite-na-2026-07-15.json), [marimo WASM N/A](../../../mainPlan/browser-os-north-star/benchmarks/s1-marimo-wasm-na-2026-07-15.json) | 같은 S1 계약은 pyproc만 충족 | 외부 후보 비교는 S1L 또는 S0로 분리한다 |
-| 2026-07-15 | JupyterLite REPL S1L 실측 | Edge/Windows, JupyterLite demo REPL | [JupyterLite S1L](../../../mainPlan/browser-os-north-star/benchmarks/s1l-jupyterlite-2026-07-15.json) median 10149ms, p95 10153ms, maxErr 0 | JupyterLite는 S1은 N/A, S1L은 측정 가능 | marimo WASM은 S1L, WebVM은 S0로 분리한다 |
-| 2026-07-15 | marimo playground S1L 실측 | Edge/Windows, marimo.app playground | [marimo WASM S1L](../../../mainPlan/browser-os-north-star/benchmarks/s1l-marimo-wasm-2026-07-15.json) median 9355ms, p95 11424ms, maxErr 0 | marimo WASM은 S1은 N/A, S1L은 측정 가능 | WebVM은 S0로 분리한다 |
+| 2026-07-15 | JupyterLite REPL S1L 실측 | Edge/Windows, JupyterLite demo REPL | [JupyterLite S1L](../../../mainPlan/browser-os-north-star/benchmarks/s1l-jupyterlite-2026-07-15.json) median 10149ms, p95 10153ms, maxErr 0 | JupyterLite는 S1은 N/A, S1L은 측정 가능 | marimo WASM과 WebVM으로 S1L을 확장한다 |
+| 2026-07-15 | marimo playground S1L 실측 | Edge/Windows, marimo.app playground | [marimo WASM S1L](../../../mainPlan/browser-os-north-star/benchmarks/s1l-marimo-wasm-2026-07-15.json) median 9355ms, p95 11424ms, maxErr 0 | marimo WASM은 S1은 N/A, S1L은 측정 가능 | WebVM S1L 가능 여부를 확인한다 |
+| 2026-07-15 | WebVM terminal S1L 실측 | Edge/Windows, WebVM terminal | [WebVM S1L](../../../mainPlan/browser-os-north-star/benchmarks/s1l-webvm-2026-07-15.json) median 11406ms, p95 12825ms, maxErr 0 | WebVM은 S1은 N/A, S1L은 측정 가능 | S1L 4자 비교표를 닫는다 |
 
 ## 판정
 
-진행 중. S1은 pyproc의 병렬 worker pool 증거 축으로 유지한다. WebVM, JupyterLite, marimo WASM은 같은 S1을 single-lane이나 boot 비교로 재정의하지 않고 N/A artifact로 봉인했다. JupyterLite와 marimo WASM은 별도 S1L artifact로 측정했으므로, 다음 질문은 WebVM S0다.
+완료. S1은 pyproc의 병렬 worker pool 증거 축으로 유지한다. WebVM, JupyterLite, marimo WASM은 같은 S1을 single-lane이나 boot 비교로 재정의하지 않고 N/A artifact로 봉인했다. S1L은 네 후보 모두 별도 single-kernel NumPy latency artifact로 측정했고, S1의 대체 간판이 아니라 보조 축으로 남긴다.