feat(git): stub git clone benchmark

.changeset/benchsdk-client-run-key.md

@@ -0,0 +1,5 @@
+---
+"@benchsdk/client": minor
+---
+
+`createRun` accepts an optional `runKey`: callers passing the same key (per org + benchmark) get-or-create one shared run instead of each opening its own. `BenchmarkRun.runKey` reports the key a run was created with.

.changeset/benchsdk-runner-shapes-run-key.md

@@ -0,0 +1,11 @@
+---
+"@benchsdk/runner": minor
+---
+
+Verbs-only CLI: `bench run <file.bench.ts>` is the one mutating command. The benchmark is declared in the file and materialized (upserted) as a side effect of running it, and a run is opened as a side effect too — there are no imperative `bench create benchmark` / `bench create run` commands.
+
+`bench run` gains `--shape <name>`: a bench file can declare named `shapes`, each swapping in its own platform identity (`slug`/`name`, optional `kind`) and a stable knob (`staggerDelayMs`) while reusing the same task and participants. This collapses the per-shape slug/name/knob triple that was duplicated across package scripts and CI.
+
+`bench run` gains `--run-key <key>`: sibling processes passing the same key (per org + benchmark) get-or-create one shared run instead of each opening its own, so provider jobs running in parallel land in a single, directly-comparable run. Each process registers only the participants it runs. The key binding is permanent, so callers that need a fresh run (e.g. a CI re-run) vary the key (e.g. include `GITHUB_RUN_ATTEMPT`).
+
+`--slug` remains a working alias for `--benchmark`.

.changeset/participant-sized-runs.md

@@ -0,0 +1,5 @@
+---
+"@benchsdk/client": minor
+---
+
+`createRun` no longer requires `totalTasks`: omit it to open a participant-sized run, whose total is the sum of what its participants declare when they register. `BenchmarkRun.participantSized` reports which kind a run is.

.github/workflows/git-benchmarks.yml

@@ -0,0 +1,126 @@
+name: Git Benchmark
+
+on:
+  push:
+    branches: [master]
+    paths:
+      - 'benchmarks/git/**'
+      - 'benchmarks/src/util/**'
+      - 'benchmarks/src/merge-results.ts'
+      - 'package.json'
+      - '.github/workflows/git-benchmarks.yml'
+  schedule:
+    - cron: '0 0 * * 1' # Weekly on Monday at midnight UTC
+  workflow_dispatch:
+    inputs:
+      iterations:
+        description: 'Iterations per provider'
+        required: false
+        default: '10'
+      provider:
+        description: 'Provider to run (leave empty for all)'
+        required: false
+        default: ''
+      dry_run:
+        description: 'Run without committing results'
+        required: false
+        default: false
+        type: boolean
+
+concurrency:
+  group: git-benchmarks
+  cancel-in-progress: true
+
+jobs:
+  bench:
+    permissions:
+      contents: read
+    name: Bench ${{ matrix.provider }}
+    runs-on: namespace-profile-default;permissions.additional_grant=vault/object:*:list;permissions.additional_grant=vault/object:*:describe
+    timeout-minutes: 60
+    strategy:
+      fail-fast: false
+      matrix:
+        provider: ${{ (github.event.inputs.provider != '' && fromJson(format('["{0}"]', github.event.inputs.provider))) || fromJson('["github","gitlab","bitbucket","tensorlake"]') }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: pnpm/action-setup@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 24
+          cache: 'pnpm'
+      - name: Install dependencies
+        run: pnpm install --frozen-lockfile
+      - name: Clear stale results from checkout
+        run: rm -rf results/git/
+      - name: Run git benchmark
+        run: |
+          . benchmarks/scripts/load-vault-secrets.sh '^(GITHUB_GIT_REPO_URL|GITHUB_TOKEN|GITLAB_GIT_REPO_URL|GITLAB_TOKEN|BITBUCKET_GIT_REPO_URL|BITBUCKET_TOKEN|TENSORLAKE_GIT_REPO_URL|TENSORLAKE_API_KEY|BENCHMARKS_PLATFORM_API_KEY)$'
+
+          ITERATIONS="${{ (github.event_name == 'schedule' && '50') || github.event.inputs.iterations || '10' }}"
+
+          # Every provider job passes the same --run-key, so all matrix jobs
+          # get-or-create one shared platform run and each registers only its
+          # own participant. The attempt is in the key so workflow re-runs open
+          # fresh runs instead of rejoining the completed ones.
+          RUN_KEY="${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+
+          npx tsx packages/benchsdk-runner/dist/bin.js run benchmarks/git/git.bench.ts \
+            --provider ${{ matrix.provider }} \
+            --iterations $ITERATIONS \
+            --run-key "$RUN_KEY"
+      - name: Upload results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: git-results-${{ matrix.provider }}
+          path: results/git/
+          if-no-files-found: ignore
+          retention-days: 7
+
+  collect:
+    name: Collect Git Results
+    runs-on: namespace-profile-default;permissions.additional_grant=vault/object:*:list;permissions.additional_grant=vault/object:*:describe
+    needs: [bench]
+    if: always()
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - uses: pnpm/action-setup@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 24
+          cache: 'pnpm'
+      - name: Install dependencies
+        run: pnpm install --frozen-lockfile
+      - name: Download all artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: artifacts/
+          pattern: git-results-*
+      - name: Merge results
+        run: npx tsx benchmarks/src/merge-results.ts --input artifacts --mode git
+      - name: Commit and push results
+        if: github.event_name != 'push' && github.event.inputs.dry_run != 'true'
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "github-actions[bot]@users.noreply.github.com"
+          git add results/git/
+          git diff --cached --quiet && echo "No changes to commit" && exit 0
+          git commit -m "chore: update git benchmark results [skip ci]"
+
+          branch="${GITHUB_REF#refs/heads/}"
+          for attempt in 1 2 3 4 5; do
+            git fetch origin "${branch}"
+            git rebase --autostash "origin/${branch}" || { git rebase --abort; exit 1; }
+            if git push origin "HEAD:${branch}"; then
+              echo "Pushed on attempt ${attempt}"
+              exit 0
+            fi
+            echo "Push rejected (attempt ${attempt}); will rebase and retry"
+          done
+          echo "Failed to push after multiple attempts" >&2
+          exit 1

.github/workflows/sandbox-dax-benchmarks.yml

@@ -8,6 +8,8 @@ on:
       - 'benchmarks/sandbox/providers.ts'
       - 'benchmarks/src/merge-results.ts'
       - 'benchmarks/scripts/dax-benchmark.sh'
+      - 'benchmarks/sandbox/generate-dax-svg.ts'
+      - 'dax.svg'
       - 'package.json'
       - '.github/workflows/sandbox-dax-benchmarks.yml'
   schedule:
@@ -149,6 +151,13 @@ jobs:
           pattern: results-*
       - name: Merge results
         run: npx tsx benchmarks/src/merge-results.ts --input artifacts
+      - name: Generate DAX SVG
+        run: |
+          if [ -f results/sandbox-dax/latest.json ]; then
+            pnpm run generate-dax-svg
+          else
+            echo "No DAX results found; skipping chart"
+          fi
       - name: Ingest results to platform
         if: github.event_name != 'push' && github.event.inputs.dry_run != 'true'
         continue-on-error: true
@@ -248,7 +257,7 @@ jobs:
         run: |
           git config user.name "github-actions[bot]"
           git config user.email "github-actions[bot]@users.noreply.github.com"
-          git add results/sandbox-dax/
+          git add results/sandbox-dax/ dax.svg
           git diff --cached --quiet && echo "No changes to commit" && exit 0
           git commit -m "chore: update dax benchmark results [skip ci]"
 

.github/workflows/sandbox-tti-benchmarks.yml

@@ -118,19 +118,26 @@ jobs:
         run: |
           . benchmarks/scripts/load-vault-secrets.sh '^(ARCHIL_API_KEY|ARCHIL_REGION|ARCHIL_DISK_ID|ARKER_API_KEY|BEAM_TOKEN|BEAM_WORKSPACE_ID|BL_API_KEY|BL_WORKSPACE|CLOUD_RUN_SANDBOX_URL|CLOUD_RUN_SANDBOX_SECRET|CLOUDFLARE_SANDBOX_URL|CLOUDFLARE_SANDBOX_SECRET|CSB_API_KEY|CREATEOS_SANDBOX_API_KEY|DAYTONA_API_KEY|DECLAW_API_KEY|E2B_API_KEY|HOPX_API_KEY|ISORUN_API_KEY|LIGHTNING_API_KEY|MODAL_TOKEN_ID|MODAL_TOKEN_SECRET|NSC_TOKEN|NORTHFLANK_TOKEN|NORTHFLANK_PROJECT_ID|OPENCOMPUTER_API_KEY|OPENCOMPUTER_API_URL|RUNLOOP_API_KEY|SAIL_API_KEY|SANDBOX0_TOKEN|SPRITES_TOKEN|SUPERSERVE_API_KEY|TENKI_API_KEY|TENSORLAKE_API_KEY|UPSTASH_BOX_API_KEY|VERCEL_TOKEN|VERCEL_TEAM_ID|VERCEL_PROJECT_ID|COMPUTESDK_ADMIN_API_KEY|BENCHMARKS_PLATFORM_API_KEY)'
 
-          # One entrypoint, three launch shapes: the knobs come from the CLI
-          # and --slug/--name pick which platform benchmark each reports to.
+          # One entrypoint, three launch shapes selected with --shape; the scale
+          # knobs come from the CLI. Every provider passes the same --run-key, so
+          # each shape's providers get-or-create one shared run and rank against
+          # each other (each still registers itself and claims its own worker).
+          # The shapes are distinct benchmark slugs, so one key value yields one
+          # run per shape; the run attempt is in the key so a workflow re-run
+          # opens fresh runs instead of rejoining the completed ones.
           BENCH=(npx tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts)
-          COMMON=(--provider ${{ matrix.provider }} \
-            --iterations ${{ github.event.inputs.iterations || (github.event_name == 'push' && '10') || '100' }})
+          RUN_KEY="${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          COMMON=(
+            --provider ${{ matrix.provider }}
+            --iterations ${{ github.event.inputs.iterations || (github.event_name == 'push' && '10') || '100' }}
+            --run-key "$RUN_KEY"
+          )
           CONCURRENCY=${{ github.event.inputs.concurrency || (github.event_name == 'push' && '10') || '100' }}
           # Sequential is one-at-a-time by definition, so it never gets
           # --concurrency (the config default pins it to 1).
           SEQUENTIAL=("${BENCH[@]}" "${COMMON[@]}")
-          STAGGERED=("${BENCH[@]}" "${COMMON[@]}" --concurrency "$CONCURRENCY" --stagger-delay-ms 200 \
-            --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)')
-          BURST=("${BENCH[@]}" "${COMMON[@]}" --concurrency "$CONCURRENCY" \
-            --slug sandbox-burst-local --name 'Sandbox burst TTI (local)')
+          STAGGERED=("${BENCH[@]}" "${COMMON[@]}" --shape staggered --concurrency "$CONCURRENCY")
+          BURST=("${BENCH[@]}" "${COMMON[@]}" --shape burst --concurrency "$CONCURRENCY")
           case "${{ github.event.inputs.mode }}" in
             sequential) "${SEQUENTIAL[@]}" ;;
             staggered)  "${STAGGERED[@]}" ;;

README.md

@@ -100,6 +100,10 @@ Our partners support our independent benchmarks.
 
 ![Object Storage Snapshot & Fork](./snapshot_fork_small.svg)
 
+### [DAX Sandbox Builds](#dax-sandbox-builds)
+
+![DAX Sandbox Builds](./dax.svg)
+
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](./LICENSE)
 
 **TTI (Time to Interactive)** = API call to first command execution. Lower is better.