diff --git a/.changeset/benchsdk-client-run-key.md b/.changeset/benchsdk-client-run-key.md
new file mode 100644
index 00000000..90795488
--- /dev/null
+++ b/.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.
diff --git a/.changeset/benchsdk-runner-shapes-run-key.md b/.changeset/benchsdk-runner-shapes-run-key.md
new file mode 100644
index 00000000..bf198f3c
--- /dev/null
+++ b/.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`.
diff --git a/.changeset/benchsdk-runner-slug-flag.md b/.changeset/benchsdk-runner-slug-flag.md
deleted file mode 100644
index f100a1c7..00000000
--- a/.changeset/benchsdk-runner-slug-flag.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"@benchsdk/runner": minor
----
-
-Add `--slug` and `--name` CLI overrides for `benchmarkSlug`/`benchmarkName`, so one `*.bench.ts` can report under several platform benchmarks (e.g. the sandbox TTI entrypoint reporting sequential/staggered/burst runs of the same workload).
diff --git a/.changeset/participant-sized-runs.md b/.changeset/participant-sized-runs.md
new file mode 100644
index 00000000..6d8b8121
--- /dev/null
+++ b/.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.
diff --git a/.github/workflows/sandbox-tti-benchmarks.yml b/.github/workflows/sandbox-tti-benchmarks.yml
index 19a5bb09..0db92f3d 100644
--- a/.github/workflows/sandbox-tti-benchmarks.yml
+++ b/.github/workflows/sandbox-tti-benchmarks.yml
@@ -116,19 +116,26 @@ jobs:
         run: |
           . benchmarks/scripts/load-vault-secrets.sh '^(ARCHIL_API_KEY|ARCHIL_REGION|ARCHIL_DISK_ID|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|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[@]}" ;;
diff --git a/benchmarks/sandbox/tti.bench.ts b/benchmarks/sandbox/tti.bench.ts
index bbb7a80c..a1b9aa7f 100644
--- a/benchmarks/sandbox/tti.bench.ts
+++ b/benchmarks/sandbox/tti.bench.ts
@@ -3,17 +3,23 @@
  * first command (`node -v`) succeeding, excluding destroy. Declarative —
  * exports `config` + `task`; `bench run` owns the entrypoint.
  *
- * One workload, three launch shapes — the only thing that differs is the
- * framework's own knobs, so they're CLI flags rather than separate files:
- *   sequential  one at a time (the config default: concurrency 1)
- *   burst       --concurrency N: all slots open at once, so this launches that
- *               many real sandboxes simultaneously — raise N deliberately
- *   staggered   --concurrency N --stagger-delay-ms D: task i starts at i * D
- * Each shape reports under its own platform benchmark via `--slug`/`--name`.
+ * One workload, three launch shapes. Each shape is its own platform benchmark
+ * but the same task, so they're declared once in `shapes` and picked with
+ * `--shape`; the scale knobs (`--iterations`/`--concurrency`) stay on the CLI:
+ *   sequential  the base config (concurrency 1) — no `--shape`
+ *   burst       --concurrency N: all slots open at once, launching that many
+ *               real sandboxes simultaneously — raise N deliberately
+ *   staggered   its 200ms delay is baked into the shape; task i starts at i * D
  *
  *   bench run benchmarks/sandbox/tti.bench.ts --iterations 5 --provider e2b,modal
- *   bench run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 10 --concurrency 10
- *   bench run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 10 --concurrency 10 --stagger-delay-ms 200
+ *   bench run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 10 --concurrency 10
+ *   bench run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 10 --concurrency 10
+ *
+ * To rank providers against each other, run each provider with the same
+ * `--run-key`: they get-or-create one shared run and each claims its own worker.
+ *
+ *   bench run benchmarks/sandbox/tti.bench.ts --shape burst --provider e2b   --run-key "$GITHUB_RUN_ID"
+ *   bench run benchmarks/sandbox/tti.bench.ts --shape burst --provider modal --run-key "$GITHUB_RUN_ID"
  */
 import '../src/env.js';
 import path from 'node:path';
@@ -52,6 +58,13 @@ export const config = defineBenchmarkConfig({
   benchmarkKind: 'sandbox',
   iterations: 2,
   concurrency: 1,
+  // The launch shapes: same task, distinct platform identities. `--shape burst`
+  // just swaps the slug/name — the caller brings `--concurrency`. Staggered
+  // also carries its defining 200ms delay so no caller has to remember it.
+  shapes: {
+    burst: { slug: 'sandbox-burst-local', name: 'Sandbox burst TTI (local)' },
+    staggered: { slug: 'sandbox-staggered-local', name: 'Sandbox staggered TTI (local)', staggerDelayMs: 200 },
+  },
   participants: providers,
   onComplete: (outcome) => {
     const { resultsDir, mode } = legacyShape(outcome.config);
diff --git a/package.json b/package.json
index f0a5d010..e3d0b3f3 100644
--- a/package.json
+++ b/package.json
@@ -5,29 +5,29 @@
   "type": "module",
   "scripts": {
     "typecheck": "tsc --noEmit",
-    "bench": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3",
+    "bench": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3",
     "bench:sequential": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts",
-    "bench:staggered": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200",
-    "bench:burst": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3",
+    "bench:staggered": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3",
+    "bench:burst": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3",
     "bench:sandbox:dax": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/dax.bench.ts",
-    "bench:blaxel": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider blaxel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider blaxel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider blaxel",
-    "bench:codesandbox": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider codesandbox && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider codesandbox && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider codesandbox",
-    "bench:daytona": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider daytona && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider daytona && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider daytona",
-    "bench:e2b": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider e2b && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider e2b && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider e2b",
-    "bench:hopx": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider hopx && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider hopx && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider hopx",
-    "bench:isorun": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider isorun && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider isorun && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider isorun",
-    "bench:modal": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider modal && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider modal && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider modal",
-    "bench:namespace": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider namespace && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider namespace && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider namespace",
-    "bench:northflank": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider northflank && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider northflank && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider northflank",
-    "bench:railway": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider railway && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider railway && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider railway",
-    "bench:render": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider render && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider render && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider render",
-    "bench:runloop": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider runloop && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider runloop && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider runloop",
-    "bench:vercel": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider vercel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider vercel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider vercel",
-    "bench:just-bash": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider just-bash && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider just-bash && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider just-bash",
-    "bench:sprites": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider sprites && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider sprites && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider sprites",
-    "bench:sandbox0": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider sandbox0 && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider sandbox0 && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider sandbox0",
-    "bench:opencomputer": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider opencomputer && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider opencomputer && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider opencomputer",
-    "bench:tensorlake": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider tensorlake && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-staggered-local --name 'Sandbox staggered TTI (local)' --iterations 3 --concurrency 3 --stagger-delay-ms 200 --provider tensorlake && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --slug sandbox-burst-local --name 'Sandbox burst TTI (local)' --iterations 3 --concurrency 3 --provider tensorlake",
+    "bench:blaxel": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider blaxel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider blaxel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider blaxel",
+    "bench:codesandbox": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider codesandbox && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider codesandbox && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider codesandbox",
+    "bench:daytona": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider daytona && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider daytona && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider daytona",
+    "bench:e2b": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider e2b && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider e2b && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider e2b",
+    "bench:hopx": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider hopx && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider hopx && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider hopx",
+    "bench:isorun": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider isorun && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider isorun && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider isorun",
+    "bench:modal": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider modal && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider modal && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider modal",
+    "bench:namespace": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider namespace && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider namespace && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider namespace",
+    "bench:northflank": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider northflank && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider northflank && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider northflank",
+    "bench:railway": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider railway && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider railway && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider railway",
+    "bench:render": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider render && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider render && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider render",
+    "bench:runloop": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider runloop && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider runloop && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider runloop",
+    "bench:vercel": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider vercel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider vercel && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider vercel",
+    "bench:just-bash": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider just-bash && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider just-bash && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider just-bash",
+    "bench:sprites": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider sprites && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider sprites && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider sprites",
+    "bench:sandbox0": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider sandbox0 && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider sandbox0 && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider sandbox0",
+    "bench:opencomputer": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider opencomputer && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider opencomputer && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider opencomputer",
+    "bench:tensorlake": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --provider tensorlake && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape staggered --iterations 3 --concurrency 3 --provider tensorlake && tsx packages/benchsdk-runner/dist/bin.js run benchmarks/sandbox/tti.bench.ts --shape burst --iterations 3 --concurrency 3 --provider tensorlake",
     "bench:browser": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/browser/browser.bench.ts",
     "bench:browser:browserbase": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/browser/browser.bench.ts --provider browserbase",
     "bench:browser:hyperbrowser": "tsx packages/benchsdk-runner/dist/bin.js run benchmarks/browser/browser.bench.ts --provider hyperbrowser",
diff --git a/packages/benchsdk-runner/src/__tests__/bench-config.test.ts b/packages/benchsdk-runner/src/__tests__/bench-config.test.ts
index d390074b..da7df4c4 100644
--- a/packages/benchsdk-runner/src/__tests__/bench-config.test.ts
+++ b/packages/benchsdk-runner/src/__tests__/bench-config.test.ts
@@ -85,6 +85,31 @@ describe('defineBenchmarkConfig', () => {
       defineBenchmarkConfig({ benchmarkSlug: 's', benchmarkName: 'n', participants, phases: [{ name: 'cold', iterations: 1 }, { name: 'cold', iterations: 1 }] }),
     ).toThrow('duplicate phase name');
   });
+
+  it('accepts a valid shapes map', () => {
+    const config = defineBenchmarkConfig({
+      benchmarkSlug: 's',
+      benchmarkName: 'n',
+      participants,
+      shapes: {
+        burst: { slug: 'sandbox-burst-local', name: 'Burst' },
+        staggered: { slug: 'sandbox-staggered-local', staggerDelayMs: 200 },
+      },
+    });
+    expect(config.shapes?.burst.slug).toBe('sandbox-burst-local');
+  });
+
+  it('rejects a shape without a lowercase slug', () => {
+    expect(() =>
+      defineBenchmarkConfig({ benchmarkSlug: 's', benchmarkName: 'n', participants, shapes: { burst: { slug: 'Burst' } } }),
+    ).toThrow('burst');
+  });
+
+  it('rejects a shape with a negative staggerDelayMs', () => {
+    expect(() =>
+      defineBenchmarkConfig({ benchmarkSlug: 's', benchmarkName: 'n', participants, shapes: { s: { slug: 'ok', staggerDelayMs: -1 } } }),
+    ).toThrow('staggerDelayMs');
+  });
 });
 
 describe('defineTask', () => {
diff --git a/packages/benchsdk-runner/src/__tests__/cli.test.ts b/packages/benchsdk-runner/src/__tests__/cli.test.ts
index 85faabab..922a9fcf 100644
--- a/packages/benchsdk-runner/src/__tests__/cli.test.ts
+++ b/packages/benchsdk-runner/src/__tests__/cli.test.ts
@@ -5,13 +5,19 @@ import { NoAvailableParticipantsError } from '../no-available-participants.js';
 const fixture = (name: string) => `src/__tests__/fixtures/${name}`;
 
 describe('runBenchmarkFile', () => {
+  it('rejects retired imperative commands (there is no `create`)', async () => {
+    await expect(runBenchmarkFile(['create', 'benchmark', 'sandbox'])).rejects.toThrow(/Usage:/);
+    await expect(runBenchmarkFile(['create', 'run'])).rejects.toThrow(/Usage:/);
+  });
+
   it('rejects when the command is not `run`', async () => {
-    await expect(runBenchmarkFile([])).rejects.toThrow(/Usage: bench run/);
-    await expect(runBenchmarkFile(['nope', fixture('good.bench.ts')])).rejects.toThrow(/Usage: bench run/);
+    await expect(runBenchmarkFile([])).rejects.toThrow(/Usage:/);
+    await expect(runBenchmarkFile(['nope', fixture('good.bench.ts')])).rejects.toThrow(/Usage:/);
   });
 
-  it('rejects when no file is given', async () => {
-    await expect(runBenchmarkFile(['run'])).rejects.toThrow(/Usage: bench run/);
+  it('rejects when no file is given, or a flag stands where the file should', async () => {
+    await expect(runBenchmarkFile(['run'])).rejects.toThrow(/Usage:/);
+    await expect(runBenchmarkFile(['run', '--shape', 'burst'])).rejects.toThrow(/Usage:/);
   });
 
   it('rejects a module that does not export a config', async () => {
diff --git a/packages/benchsdk-runner/src/__tests__/runner.test.ts b/packages/benchsdk-runner/src/__tests__/runner.test.ts
index d6c0db22..7edffce3 100644
--- a/packages/benchsdk-runner/src/__tests__/runner.test.ts
+++ b/packages/benchsdk-runner/src/__tests__/runner.test.ts
@@ -49,15 +49,23 @@ describe('parseCliArgs', () => {
     expect(parseCliArgs(['--group-by=participant'])).toEqual({ groupBy: 'participant' });
   });
 
-  it('parses --slug and --name', () => {
-    expect(parseCliArgs(['--slug', 'sandbox-burst-local'])).toEqual({ slug: 'sandbox-burst-local' });
-    expect(parseCliArgs(['--slug=sandbox-tti-local'])).toEqual({ slug: 'sandbox-tti-local' });
+  it('parses --benchmark (and its --slug alias) and --name', () => {
+    expect(parseCliArgs(['--benchmark', 'sandbox-burst-local'])).toEqual({ benchmark: 'sandbox-burst-local' });
+    expect(parseCliArgs(['--benchmark=sandbox-tti-local'])).toEqual({ benchmark: 'sandbox-tti-local' });
+    expect(parseCliArgs(['--slug', 'sandbox-burst-local'])).toEqual({ benchmark: 'sandbox-burst-local' });
     expect(parseCliArgs(['--name', 'Sandbox burst TTI'])).toEqual({ name: 'Sandbox burst TTI' });
     expect(() => parseCliArgs(['--name', ' '])).toThrow('--name');
   });
 
-  it('throws on a non-slug --slug', () => {
-    expect(() => parseCliArgs(['--slug', 'Sandbox TTI'])).toThrow('--slug');
+  it('parses --shape and --run-key', () => {
+    expect(parseCliArgs(['--shape', 'burst'])).toEqual({ shape: 'burst' });
+    expect(parseCliArgs(['--run-key=ci-123'])).toEqual({ runKey: 'ci-123' });
+    expect(() => parseCliArgs(['--shape', ''])).toThrow('--shape');
+    expect(() => parseCliArgs(['--run-key', ''])).toThrow('--run-key');
+  });
+
+  it('throws on a non-slug --benchmark', () => {
+    expect(() => parseCliArgs(['--benchmark', 'Sandbox TTI'])).toThrow('--benchmark');
     expect(() => parseCliArgs(['--slug', ''])).toThrow('--slug');
   });
 
@@ -164,14 +172,19 @@ describe('runBenchmark', () => {
     process.env.E2B_API_KEY = 'x';
     process.env.MODAL_TOKEN = 'y';
     process.env.BENCHMARKS_PLATFORM_API_KEY = 'test-key';
-    calls = { upsertBenchmark: [], createRun: [], planWorkers: [], runWorker: [], taskData: [] };
+    calls = { upsertBenchmark: [], createRun: [], planWorkers: [], upsertParticipant: [], getRun: [], runWorker: [], taskData: [] };
     fakeClient = {
       upsertBenchmark: vi.fn(async (...a: any[]) => { calls.upsertBenchmark.push(a); return {}; }),
       createRun: vi.fn(async (...a: any[]) => { calls.createRun.push(a); return { run: { id: 'run-1' }, participants: [] }; }),
       planWorkers: vi.fn(async (...a: any[]) => { calls.planWorkers.push(a); return []; }),
+      upsertParticipant: vi.fn(async (...a: any[]) => { calls.upsertParticipant.push(a); return {}; }),
+      getRun: vi.fn(async (slug: string, runId: string) => {
+        calls.getRun.push([slug, runId]);
+        return { id: runId, totalTasks: 3, participantSized: runId === 'run-open' };
+      }),
       runWorker: vi.fn(async (opts: any) => {
         calls.runWorker.push(opts);
-        const total = calls.createRun[0]?.[1]?.totalTasks ?? 1;
+        const total = calls.createRun[0]?.[1]?.totalTasks ?? calls.upsertParticipant[0]?.[3]?.totalTasks ?? 1;
         // The platform hands out globally-indexed task ranges; `taskRangeStart`
         // lets a test exercise a worker whose range doesn't start at 0.
         const start = taskRangeStart;
@@ -294,6 +307,84 @@ describe('runBenchmark', () => {
     expect(calls.createRun[0][0]).toBe('sandbox-burst-local');
   });
 
+  it('shares a run by --run-key: get-or-creates it keyed, then registers only its own participant', async () => {
+    const config: BenchmarkConfig<typeof participants[number]> = {
+      benchmarkSlug: 'sandbox-tti-local',
+      benchmarkName: 'Sandbox TTI',
+      iterations: 2,
+      participants: [participants[0]],
+    };
+
+    const outcome = await runBenchmark(config, defineTask(async () => ({})), ['--run-key', 'ci-123', '--provider', 'e2b']);
+
+    // The benchmark is still materialized from the file identity.
+    expect(calls.upsertBenchmark[0][0]).toBe('sandbox-tti-local');
+    // One keyed get-or-create, carrying the key and no size/participant list.
+    expect(calls.createRun).toHaveLength(1);
+    expect(calls.createRun[0][1]).toMatchObject({ runKey: 'ci-123' });
+    expect(calls.createRun[0][1].totalTasks).toBeUndefined();
+    expect(calls.createRun[0][1].participants).toBeUndefined();
+    // Registers only the provider it runs, sized to its own iteration count.
+    expect(calls.upsertParticipant[0].slice(0, 3)).toEqual(['sandbox-tti-local', 'run-1', 'e2b']);
+    expect(calls.upsertParticipant[0][3]).toMatchObject({ totalTasks: 2 });
+    expect(calls.runWorker[0]).toMatchObject({ runId: 'run-1' });
+    expect(outcome.runId).toBe('run-1');
+    expect(outcome.participants[0].records).toHaveLength(2);
+  });
+
+  it('does not rename a benchmark it was merely retargeted at', async () => {
+    const config: BenchmarkConfig<typeof participants[number]> = {
+      benchmarkSlug: 'sandbox-tti-local',
+      benchmarkName: 'Sandbox TTI',
+      iterations: 1,
+      participants: [participants[0]],
+    };
+
+    await runBenchmark(config, defineTask(async () => ({})), ['--benchmark', 'sandbox-burst-local']);
+    expect(calls.upsertBenchmark).toEqual([]);
+
+    await runBenchmark(config, defineTask(async () => ({})), [
+      '--benchmark',
+      'sandbox-burst-local',
+      '--name',
+      'Sandbox burst TTI',
+    ]);
+    expect(calls.upsertBenchmark[0]).toEqual(['sandbox-burst-local', { name: 'Sandbox burst TTI' }]);
+  });
+
+  it('selects a declared shape by --shape, reporting under its slug and name', async () => {
+    const config: BenchmarkConfig<typeof participants[number]> = {
+      benchmarkSlug: 'sandbox-tti-local',
+      benchmarkName: 'Sandbox TTI',
+      iterations: 1,
+      participants: [participants[0]],
+      shapes: {
+        staggered: { slug: 'sandbox-staggered-local', name: 'Sandbox staggered TTI', staggerDelayMs: 200 },
+      },
+    };
+
+    const outcome = await runBenchmark(config, defineTask(async () => ({})), ['--shape', 'staggered']);
+
+    expect(calls.upsertBenchmark[0][0]).toBe('sandbox-staggered-local');
+    expect(calls.upsertBenchmark[0][1]).toMatchObject({ name: 'Sandbox staggered TTI' });
+    expect(calls.createRun[0][0]).toBe('sandbox-staggered-local');
+    // The shape's stable knob applies; scale knobs stay defaulted/overridable.
+    expect(outcome.config.staggerDelayMs).toBe(200);
+  });
+
+  it('rejects an unknown --shape, listing the declared ones', async () => {
+    const config: BenchmarkConfig<typeof participants[number]> = {
+      benchmarkSlug: 'sandbox-tti-local',
+      benchmarkName: 'Sandbox TTI',
+      participants: [participants[0]],
+      shapes: { burst: { slug: 'sandbox-burst-local' } },
+    };
+
+    await expect(
+      runBenchmark(config, defineTask(async () => ({})), ['--shape', 'nope']),
+    ).rejects.toThrow('Known shapes: burst');
+  });
+
   it('throws NoAvailableParticipantsError, listing the skips, when no participant has its env vars set', async () => {
     delete process.env.E2B_API_KEY;
     delete process.env.MODAL_TOKEN;
diff --git a/packages/benchsdk-runner/src/bench-config.ts b/packages/benchsdk-runner/src/bench-config.ts
index b27558ec..c1952ab2 100644
--- a/packages/benchsdk-runner/src/bench-config.ts
+++ b/packages/benchsdk-runner/src/bench-config.ts
@@ -19,6 +19,10 @@
  *   burst       { iterations: N, concurrency: N }
  *   staggered   { iterations: N, concurrency: N, staggerDelayMs: 200 }
  *
+ * A benchmark can name these variants up front via `shapes`, so one file backs
+ * several platform benchmarks (`bench run <file> --shape burst`) without
+ * restating each one's slug/name in scripts and CI.
+ *
  * A task is comprised of steps, declared via `ctx.step` inside a task function
  * — it supports closures, conditionals and try/finally, so values (a created
  * sandbox, say) flow naturally between steps. A task that declares no steps is
@@ -37,6 +41,25 @@ import type {
 /** How tasks are ordered across participants. */
 export type GroupBy = 'participant' | 'round';
 
+/**
+ * A named variant of a benchmark, selected with `--shape <name>`. A shape
+ * carries only the parts that make it a distinct *benchmark* — its platform
+ * identity plus any stable distinguishing knob (e.g. staggered's delay). The
+ * scale knobs that vary per environment (`--iterations`, `--concurrency`) stay
+ * on the invocation, so a shape never sets a value only to have the CLI
+ * override it.
+ */
+export interface BenchmarkShape {
+  /** Platform slug this shape reports under (e.g. 'sandbox-burst-local'). */
+  slug: string;
+  /** Display name shown on the platform; defaults to the slug. */
+  name?: string;
+  /** Benchmark kind; defaults to the config's `benchmarkKind`. */
+  kind?: string;
+  /** Default stagger delay (ms) for this shape; overridable with `--stagger-delay-ms`. */
+  staggerDelayMs?: number;
+}
+
 /**
  * What a task returns: whatever it measured itself. This replaces the
  * assumption that the framework owns all timing. A plain data payload is
@@ -132,6 +155,7 @@ export interface ResolvedRunConfig {
  */
 export interface BenchmarkRunOutcome {
   runId: string;
+  /** Link to this run on the platform dashboard. */
   dashboardUrl: string;
   participants: ParticipantRecords[];
   config: ResolvedRunConfig;
@@ -146,14 +170,22 @@ export interface BenchmarkRunOutcome {
 export interface BenchmarkConfig<T extends BaseParticipant = BaseParticipant> {
   /**
    * Stable platform slug for this benchmark (e.g. 'sandbox-tti-local').
-   * Overridable per run with `--slug`, so one entrypoint can report under
-   * several benchmarks.
+   * Selectable per run with `--shape` (or overridable with `--benchmark`), so
+   * one entrypoint can report under several benchmarks.
    */
   benchmarkSlug: string;
   /** Human-readable name shown on the platform. Overridable with `--name`. */
   benchmarkName: string;
   /** Optional platform benchmark kind (e.g. 'sandbox'). */
   benchmarkKind?: string;
+  /**
+   * Named variants of this benchmark, selected with `--shape <name>`. Each
+   * shape swaps in its own platform identity (and optional stable knob) while
+   * reusing the same task and participants, so one bench file can back several
+   * platform benchmarks without duplicating the slug/name triple across
+   * package scripts and CI.
+   */
+  shapes?: Record<string, BenchmarkShape>;
   /**
    * Total tasks to run per participant. Default: 1. Mutually exclusive with
    * `phases` — when `phases` is set, total iterations = sum of phase iterations.
@@ -234,6 +266,19 @@ export function defineBenchmarkConfig<T extends BaseParticipant = BaseParticipan
   if (config.groupBy !== undefined && config.groupBy !== 'participant' && config.groupBy !== 'round') {
     throw new Error(`groupBy must be 'participant' or 'round' (got ${config.groupBy})`);
   }
+  if (config.shapes !== undefined) {
+    for (const [shapeName, shape] of Object.entries(config.shapes)) {
+      if (!shape.slug || !/^[a-z0-9][a-z0-9-]*$/.test(shape.slug)) {
+        throw new Error(`shape '${shapeName}' needs a lowercase slug (got ${JSON.stringify(shape.slug)})`);
+      }
+      if (shape.name !== undefined && (typeof shape.name !== 'string' || shape.name.trim() === '')) {
+        throw new Error(`shape '${shapeName}' name must be a non-empty string`);
+      }
+      if (shape.staggerDelayMs !== undefined && (!Number.isFinite(shape.staggerDelayMs) || shape.staggerDelayMs < 0)) {
+        throw new Error(`shape '${shapeName}' staggerDelayMs must be a number >= 0 (got ${shape.staggerDelayMs})`);
+      }
+    }
+  }
   return config;
 }
 
diff --git a/packages/benchsdk-runner/src/cli.ts b/packages/benchsdk-runner/src/cli.ts
index c76b4587..71089261 100644
--- a/packages/benchsdk-runner/src/cli.ts
+++ b/packages/benchsdk-runner/src/cli.ts
@@ -1,23 +1,31 @@
 /**
- * `bench run <file> [--flags]` — the author-facing entrypoint. Imports a
- * benchmark module, reads its `config` and `task` exports, and drives the run
- * via the internal `runBenchmark`. CLI flags override the config's knobs, and
- * `config.onComplete` (if any) fires once the run finishes.
+ * The author-facing entrypoint. `bench` is verbs-only — the benchmark and its
+ * runs are implicit, never nouns you type:
+ *
+ *   bench run <file.bench.ts> [--flags]     execute a benchmark
+ *
+ * `run` imports a benchmark module, reads its `config` and `task` exports and
+ * drives `runBenchmark`; CLI flags override the config's knobs and
+ * `config.onComplete` (if any) fires once the run finishes. The benchmark is
+ * declared in the file (`--shape` picks a named variant) and materialized on
+ * run; a run is opened as a side effect, shared across sibling processes when
+ * they pass the same `--run-key`. There are no imperative `create` commands.
  *
  * The executable wrapper lives in `bin.ts`; this module has no side effects so
  * it can be unit-tested by calling `runBenchmarkFile` directly.
  */
 import { resolve } from 'node:path';
 import { pathToFileURL } from 'node:url';
-import { runBenchmark } from './runner.js';
+import { parseCliArgs, runBenchmark } from './runner.js';
 import { NoAvailableParticipantsError } from './no-available-participants.js';
 import type { BaseParticipant } from '@benchsdk/client';
 import type { BenchmarkConfig, BenchmarkTask } from './bench-config.js';
 
 const USAGE =
-  'Usage: bench run <file.bench.ts> [--iterations N] [--concurrency N] ' +
-  '[--stagger-delay-ms N] [--group-by participant|round] [--provider a,b] ' +
-  "[--slug my-benchmark] [--name 'My benchmark']";
+  'Usage:\n' +
+  '  bench run <file.bench.ts> [--shape name] [--provider a,b] [--run-key key]\n' +
+  '      [--benchmark slug] [--name "My benchmark"]\n' +
+  '      [--iterations N] [--concurrency N] [--stagger-delay-ms N] [--group-by participant|round]';
 
 /** A benchmark module is expected to export `config` and `task`. */
 interface BenchmarkModule {
@@ -33,19 +41,17 @@ function isBenchmarkConfig(value: unknown): value is BenchmarkConfig {
 }
 
 /**
- * Loads a benchmark file and runs it. Throws on bad usage / invalid exports and
- * lets `NoAvailableParticipantsError` propagate so the caller can map it to a
- * clean exit. Does not call `process.exit`.
+ * Dispatches one CLI invocation. Throws on bad usage / invalid exports and lets
+ * `NoAvailableParticipantsError` propagate so the caller can map it to a clean
+ * exit. Does not call `process.exit`.
  */
 export async function runBenchmarkFile(argv: string[]): Promise<void> {
-  const [command, file, ...rest] = argv;
-  if (command !== 'run' || !file) {
-    throw new Error(USAGE);
-  }
+  const [command, ...rest] = argv;
 
-  const moduleUrl = pathToFileURL(resolve(process.cwd(), file)).href;
-  const mod = (await import(moduleUrl)) as BenchmarkModule;
+  const [file, ...flags] = rest;
+  if (command !== 'run' || !file || file.startsWith('-')) throw new Error(USAGE);
 
+  const mod = (await import(pathToFileURL(resolve(process.cwd(), file)).href)) as BenchmarkModule;
   const config = mod.config;
   const task = mod.task ?? mod.default;
 
@@ -56,7 +62,7 @@ export async function runBenchmarkFile(argv: string[]): Promise<void> {
     throw new Error(`${file} must export a \`task\` created with defineTask.`);
   }
 
-  await runBenchmark(config as BenchmarkConfig<BaseParticipant>, task as BenchmarkTask<BaseParticipant>, rest);
+  await runBenchmark(config as BenchmarkConfig<BaseParticipant>, task as BenchmarkTask<BaseParticipant>, flags);
 }
 
 /** Executable entry: runs the file and maps outcomes to process exit codes. */
diff --git a/packages/benchsdk-runner/src/runner.ts b/packages/benchsdk-runner/src/runner.ts
index b07477ad..950188cf 100644
--- a/packages/benchsdk-runner/src/runner.ts
+++ b/packages/benchsdk-runner/src/runner.ts
@@ -24,7 +24,6 @@ import { NoAvailableParticipantsError } from './no-available-participants.js';
 import type {
   BaseParticipant,
   BenchmarkClient,
-  BenchmarkRun,
   JsonObject,
   RunWorkerContext,
   TaskResultRecord,
@@ -34,6 +33,7 @@ import { TaskError } from './bench-config.js';
 import type {
   BenchmarkConfig,
   BenchmarkRunOutcome,
+  BenchmarkShape,
   BenchmarkTask,
   GroupBy,
   ParticipantRecords,
@@ -44,8 +44,16 @@ import type {
 import { LogBuffer } from './log-buffer.js';
 
 export interface CliArgs {
-  slug?: string;
+  /** Which platform benchmark to report as (`--benchmark`, aka the benchmark slug). */
+  benchmark?: string;
   name?: string;
+  /** Named variant from the bench file's `shapes` (`--shape`), swapping in its identity. */
+  shape?: string;
+  /**
+   * Idempotency key (`--run-key`): sibling processes passing the same key share
+   * one run (get-or-created), instead of each opening its own.
+   */
+  runKey?: string;
   iterations?: number;
   concurrency?: number;
   staggerDelayMs?: number;
@@ -98,12 +106,14 @@ export function parseCliArgs(argv: string[]): CliArgs {
     const arg = argv[i];
     const name = arg.includes('=') ? arg.slice(0, arg.indexOf('=')) : arg;
     switch (name) {
-      case '--slug': {
+      // `--slug` is the pre-`--benchmark` spelling, kept working for existing scripts.
+      case '--slug':
+      case '--benchmark': {
         const { value, nextIndex } = readValue(arg, i);
         if (!/^[a-z0-9][a-z0-9-]*$/.test(value)) {
-          throw new Error(`--slug expects a lowercase slug (got "${value}")`);
+          throw new Error(`${name} expects a lowercase benchmark slug (got "${value}")`);
         }
-        args.slug = value;
+        args.benchmark = value;
         i = nextIndex;
         break;
       }
@@ -114,6 +124,20 @@ export function parseCliArgs(argv: string[]): CliArgs {
         i = nextIndex;
         break;
       }
+      case '--shape': {
+        const { value, nextIndex } = readValue(arg, i);
+        if (value.trim() === '') throw new Error('--shape expects a value');
+        args.shape = value;
+        i = nextIndex;
+        break;
+      }
+      case '--run-key': {
+        const { value, nextIndex } = readValue(arg, i);
+        if (value.trim() === '') throw new Error('--run-key expects a value');
+        args.runKey = value;
+        i = nextIndex;
+        break;
+      }
       case '--iterations': {
         const { value, nextIndex } = readValue(arg, i);
         args.iterations = intFlag(value, '--iterations');
@@ -234,40 +258,99 @@ function resolvePlatform(): { baseUrl: string; apiKey: string } {
 }
 
 /**
- * Runs `config`'s `task` against `participants`. Selects participants by
- * `--provider` (if given), env-gates them, then drives them per the resolved
- * `groupBy`. `--slug`/`--name` retarget the whole run at a different platform
- * benchmark, so one entrypoint can report under several slugs.
+ * Resolves `--shape <name>` against the config's declared `shapes`. Throws with
+ * the known names if the shape is unknown, so a typo fails loudly instead of
+ * silently running the base benchmark.
  */
-export async function runBenchmark<T extends BaseParticipant>(
+function resolveShape<T extends BaseParticipant>(
+  config: BenchmarkConfig<T>,
+  shapeName: string | undefined,
+): BenchmarkShape | undefined {
+  if (!shapeName) return undefined;
+  const shape = config.shapes?.[shapeName];
+  if (!shape) {
+    const known = Object.keys(config.shapes ?? {});
+    throw new Error(
+      known.length > 0
+        ? `Unknown --shape "${shapeName}". Known shapes: ${known.join(', ')}.`
+        : `Unknown --shape "${shapeName}": this benchmark declares no shapes.`,
+    );
+  }
+  return shape;
+}
+
+/**
+ * Swaps a shape's identity (and its stable knob) into the config. Only the
+ * parts that make it a distinct benchmark move here; scale knobs stay on the
+ * CLI, so `mergeConfig` still lets `--concurrency`/`--iterations` win.
+ */
+function applyShape<T extends BaseParticipant>(
+  config: BenchmarkConfig<T>,
+  shape: BenchmarkShape | undefined,
+): BenchmarkConfig<T> {
+  if (!shape) return config;
+  const kind = shape.kind ?? config.benchmarkKind;
+  return {
+    ...config,
+    benchmarkSlug: shape.slug,
+    benchmarkName: shape.name ?? shape.slug,
+    ...(kind ? { benchmarkKind: kind } : {}),
+    ...(shape.staggerDelayMs !== undefined ? { staggerDelayMs: shape.staggerDelayMs } : {}),
+  };
+}
+
+/** Applies the `--benchmark`/`--name` overrides, so one entrypoint can report under several benchmarks. */
+function applyIdentityOverrides<T extends BaseParticipant>(
   fileConfig: BenchmarkConfig<T>,
-  task: BenchmarkTask<T>,
-  argv: string[] = [],
-): Promise<BenchmarkRunOutcome> {
-  const args = parseCliArgs(argv);
-  const config = {
+  args: CliArgs,
+): BenchmarkConfig<T> {
+  return {
     ...fileConfig,
-    ...(args.slug ? { benchmarkSlug: args.slug } : {}),
+    ...(args.benchmark ? { benchmarkSlug: args.benchmark } : {}),
     ...(args.name ? { benchmarkName: args.name } : {}),
   };
-  const resolved = mergeConfig(config, args);
-  const schedule = buildSchedule(config, resolved.iterations, task);
-  const totalTasks = schedule.length;
+}
 
-  const selected = selectParticipants(config.participants, resolved.providers);
-  const { available, skipped } = filterParticipantsByEnv(selected);
+function dashboardUrlFor(baseUrl: string, organizationSlug: string, benchmarkSlug: string, runId: string): string {
+  return `${baseUrl.replace(/\/api\/v1\/?$/, '')}/${organizationSlug}/benchmarks/${benchmarkSlug}/runs/${runId}`;
+}
 
+/** The participants a run covers: `--provider` selection, minus any whose env vars are unset. */
+function resolveParticipants<T extends BaseParticipant>(config: BenchmarkConfig<T>, resolved: ResolvedRunConfig): T[] {
+  const { available, skipped } = filterParticipantsByEnv(selectParticipants(config.participants, resolved.providers));
   for (const s of skipped) {
     console.log(`Skipping ${s.name}: missing ${s.missing.join(', ')}`);
   }
+  if (available.length === 0) throw new NoAvailableParticipantsError(skipped);
+  return available;
+}
 
-  if (available.length === 0) {
-    throw new NoAvailableParticipantsError(skipped);
-  }
+/**
+ * Runs `config`'s `task` against its participants. Selects participants by
+ * `--provider` (if given), env-gates them, then drives them per the resolved
+ * `groupBy`. `--shape` swaps in a declared variant's identity; `--benchmark`/
+ * `--name` retarget the run at a different platform benchmark, so one entrypoint
+ * can report under several slugs. With `--run-key`, sibling processes (e.g. one
+ * CI job per provider) get-or-create one shared run and each registers only its
+ * own participants.
+ */
+export async function runBenchmark<T extends BaseParticipant>(
+  fileConfig: BenchmarkConfig<T>,
+  task: BenchmarkTask<T>,
+  argv: string[] = [],
+): Promise<BenchmarkRunOutcome> {
+  const args = parseCliArgs(argv);
+  const shaped = applyShape(fileConfig, resolveShape(fileConfig, args.shape));
+  const config = applyIdentityOverrides(shaped, args);
+  const resolved = mergeConfig(config, args);
+  const available = resolveParticipants(config, resolved);
 
   const { baseUrl, apiKey } = resolvePlatform();
   const client = createBenchmarkClient({ baseUrl, apiKey });
 
+  const schedule = buildSchedule(config, resolved.iterations, task);
+  const totalTasks = schedule.length;
+
   const concurrencyLabel = resolved.groupBy === 'round' ? 'n/a (round mode)' : String(resolved.concurrency);
   console.log(`${config.benchmarkName} (self-contained)`);
   console.log(`Date: ${new Date().toISOString()}`);
@@ -276,34 +359,65 @@ export async function runBenchmark<T extends BaseParticipant>(
       `staggerDelayMs=${resolved.staggerDelayMs}, groupBy=${resolved.groupBy}\n`,
   );
 
-  await client.upsertBenchmark(config.benchmarkSlug, {
-    name: config.benchmarkName,
-    ...(config.benchmarkKind ? { kind: config.benchmarkKind } : {}),
-  });
-
-  const { run, organizationSlug } = await client.createRun(config.benchmarkSlug, {
-    name: `${config.benchmarkSlug} — ${totalTasks} iterations, concurrency ${resolved.concurrency}`,
-    totalTasks,
-    workerCount: 1,
-    participants: available.map((p) => p.name),
-  });
+  // Declaratively materialize the benchmark from the file/shape identity, which
+  // is authoritative (its name lives in the file). A bare `--benchmark X` only
+  // *retargets* reporting at a benchmark this file doesn't name, so we don't
+  // upsert it — that would rename it to the file's own name.
+  const identityIsOurs =
+    args.shape !== undefined ||
+    args.name !== undefined ||
+    !args.benchmark ||
+    args.benchmark === fileConfig.benchmarkSlug;
+  if (identityIsOurs) {
+    await client.upsertBenchmark(config.benchmarkSlug, {
+      name: config.benchmarkName,
+      ...(config.benchmarkKind ? { kind: config.benchmarkKind } : {}),
+    });
+  }
 
-  const dashboardUrl = `${baseUrl.replace(/\/api\/v1\/?$/, '')}/${organizationSlug}/benchmarks/${config.benchmarkSlug}/runs/${run.id}`;
-  console.log(`Run created: ${run.id}`);
-  console.log(`View at: ${dashboardUrl}\n`);
+  let runId: string;
+  let dashboardUrl: string;
+  if (args.runKey) {
+    // Shared run: get-or-created by key, so sibling processes (one per provider)
+    // converge on one run. Opened participant-sized — register only the
+    // providers this process runs and let each sibling register its own, so the
+    // run lists exactly who's benchmarked and each brings its own task count.
+    const { run, organizationSlug } = await client.createRun(config.benchmarkSlug, {
+      name: config.benchmarkName,
+      runKey: args.runKey,
+    });
+    runId = run.id;
+    dashboardUrl = dashboardUrlFor(baseUrl, organizationSlug, config.benchmarkSlug, run.id);
+    for (const participant of available) {
+      await client.upsertParticipant(config.benchmarkSlug, runId, participant.name, { totalTasks });
+    }
+    console.log(`Shared run (key "${args.runKey}"): ${runId}`);
+    console.log(`View at: ${dashboardUrl}\n`);
+  } else {
+    const { run, organizationSlug } = await client.createRun(config.benchmarkSlug, {
+      name: `${config.benchmarkSlug} — ${totalTasks} iterations, concurrency ${resolved.concurrency}`,
+      totalTasks,
+      workerCount: 1,
+      participants: available.map((p) => p.name),
+    });
+    runId = run.id;
+    dashboardUrl = dashboardUrlFor(baseUrl, organizationSlug, config.benchmarkSlug, run.id);
+    console.log(`Run created: ${runId}`);
+    console.log(`View at: ${dashboardUrl}\n`);
+  }
 
   const onResult = defaultOnResult;
 
   let participantRecords: ParticipantRecords[];
   if (resolved.groupBy === 'round') {
-    participantRecords = await runGroupedByRound(config, schedule, available, resolved, client, run, baseUrl, apiKey, onResult);
+    participantRecords = await runGroupedByRound(config, schedule, available, resolved, client, runId, baseUrl, apiKey, onResult);
   } else {
-    participantRecords = await runGroupedByParticipant(config, schedule, available, resolved, client, run, onResult);
+    participantRecords = await runGroupedByParticipant(config, schedule, available, resolved, client, runId, onResult);
   }
 
   console.log(`All done. View at: ${dashboardUrl}`);
   const outcome: BenchmarkRunOutcome = {
-    runId: run.id,
+    runId,
     dashboardUrl,
     participants: participantRecords,
     config: resolved,
@@ -325,7 +439,7 @@ async function runGroupedByParticipant<T extends BaseParticipant>(
   available: T[],
   resolved: ResolvedRunConfig,
   client: BenchmarkClient,
-  run: BenchmarkRun,
+  runId: string,
   onResult: OnResult,
 ): Promise<ParticipantRecords[]> {
   const participantRecords: ParticipantRecords[] = [];
@@ -338,11 +452,11 @@ async function runGroupedByParticipant<T extends BaseParticipant>(
     // count can't inflate launch offsets: a task whose slot frees after its
     // scheduled launch time starts immediately instead of sleeping index*delay.
     let rampStartMs: number | undefined;
-    await client.planWorkers(config.benchmarkSlug, run.id, participant.name);
+    await client.planWorkers(config.benchmarkSlug, runId, participant.name);
 
     const result = await client.runWorker({
       benchmarkSlug: config.benchmarkSlug,
-      runId: run.id,
+      runId: runId,
       participantSlug: participant.name,
       concurrency: resolved.concurrency,
       task: async (ctx: RunWorkerContext) => {
@@ -372,7 +486,7 @@ async function runGroupedByParticipant<T extends BaseParticipant>(
     });
 
     if (!result.assignment) {
-      console.error(`  No pending worker to claim for run ${run.id} — it may already be fully claimed.`);
+      console.error(`  No pending worker to claim for run ${runId} — it may already be fully claimed.`);
       participantRecords.push({ participant: participant.name, records: result.records ?? [] });
       continue;
     }
@@ -397,7 +511,7 @@ async function runGroupedByRound<T extends BaseParticipant>(
   available: T[],
   resolved: ResolvedRunConfig,
   client: BenchmarkClient,
-  run: BenchmarkRun,
+  runId: string,
   baseUrl: string,
   apiKey: string,
   onResult: OnResult,
@@ -414,7 +528,7 @@ async function runGroupedByRound<T extends BaseParticipant>(
     // reads `targetConcurrency` as tasks-per-worker, so it must be the full
     // schedule length — otherwise only one task is planned and every record
     // past the first falls outside the worker's task range.
-    await client.planWorkers(config.benchmarkSlug, run.id, participant.name, {
+    await client.planWorkers(config.benchmarkSlug, runId, participant.name, {
       workerCount: 1,
       targetConcurrency: schedule.length,
     });
@@ -424,7 +538,7 @@ async function runGroupedByRound<T extends BaseParticipant>(
         baseUrl,
         apiKey,
         benchmarkSlug: config.benchmarkSlug,
-        runId: run.id,
+        runId: runId,
         participantSlug: participant.name,
         processKind: 'process',
         processKey: process.env.HOSTNAME ?? 'local',
diff --git a/packages/benchsdk/src/types.ts b/packages/benchsdk/src/types.ts
index 644400c5..df9b639c 100644
--- a/packages/benchsdk/src/types.ts
+++ b/packages/benchsdk/src/types.ts
@@ -28,7 +28,11 @@ export interface BenchmarkRun {
   benchmarkId: string;
   name?: string | null;
   status: BenchmarkRunStatus | string;
+  /** Idempotency key: runs created with the same key (per org + benchmark) are the same run. */
+  runKey?: string | null;
   totalTasks: number;
+  /** The run declared no size: `totalTasks` is the sum of what its participants declare. */
+  participantSized?: boolean;
   workerCount: number;
   config?: JsonObject;
   createdAt?: string;
@@ -116,8 +120,14 @@ export interface UpdateBenchmarkInput {
 
 export interface CreateRunInput {
   name?: string;
-  totalTasks: number;
-  workerCount: number;
+  /**
+   * Idempotency key for get-or-create: sibling callers passing the same key
+   * (per org + benchmark) converge on one run instead of each opening its own.
+   */
+  runKey?: string;
+  /** Omit to open a participant-sized run: each participant declares its own size when it registers. */
+  totalTasks?: number;
+  workerCount?: number;
   participants?: string[];
   config?: JsonObject;
 }