bench: measure current interaction metrics in the release capture

bench/README.md

@@ -234,8 +234,11 @@ bun run bench:core-paths
 
 The evidence page at
 `docs/src/content/docs/guides/performance-resources.md` publishes every
-artifact above. One command captures that whole set, in order, from a clean
-tree. Use one pinned CPU on an otherwise idle machine, with no concurrent
+artifact above. `bench:release` captures render-scale, the interaction
+evidence, paged storage, data, core paths, formula, formula engines, matched
+engines, and XLSX, then records the render regression baseline last. Every
+capture stamps its artifact with the commit, tree state, and timestamp.
+Use one pinned CPU on an otherwise idle machine, with no concurrent
 builds or timing captures. The example below uses CPU 4 on Linux:
 
 ```sh
@@ -258,6 +261,18 @@ matrices of the suites that have one, writing to `results/smoke/` (ignored by
 git) so a rehearsal never overwrites published evidence. Regenerate the page
 after a release capture with `bun run docs:generate`.
 
+The performance showcase reads `results/paged-results.json` and
+`results/interaction-results.json`. `bench:paged` writes the paged capture;
+`bench:interaction` measures the current values of the four interaction metrics
+on the current build — packed inverse-index lookup latency, inverse-index
+backing bytes, retained bytes after 100 distant edits, and the cold-route long
+task owned by Sheetwrite — with raw samples, their summary, and the runner.
+Those four metrics are held to the absolute release ceilings recorded in
+`src/interaction-gate.ts`, not to a comparison with another build: the version 1
+artifact was a one-off before/after pair, and a release cannot re-measure a
+baseline build. `bench:release` refreshes both captures, and
+`bench:interaction:smoke` rehearses the interaction one.
+
 A change to a harness source file, to the sampling flags, or to the protocol
 version invalidates the committed render baseline: `bench:check` compares the
 harness hashes, the declared sampling, and the runner (OS, CPU, Bun, Node,

bench/package.json

@@ -11,6 +11,8 @@
     "bench:range:smoke": "bun run src/range-bench.ts --smoke",
     "bench:paged": "bun run src/paged-bench.ts",
     "bench:paged:smoke": "bun run src/paged-bench.ts --smoke",
+    "bench:interaction": "bun run src/interaction-bench.ts",
+    "bench:interaction:smoke": "bun run src/interaction-bench.ts --smoke",
     "bench:core-paths": "bun run src/core-paths-bench.ts --output results/core-paths-results.json",
     "bench:core-paths:smoke": "bun run src/core-paths-bench.ts --smoke",
     "bench:release": "bun run src/release-capture.ts",

bench/results/interaction-results.json

@@ -1,70 +1,147 @@
 {
-  "protocolVersion": 1,
-  "matrixId": "interaction-full-v1",
-  "capturedAt": "2026-07-22T03:02:46Z",
-  "source": {
-    "commit": "77c41983b53f09c393c6f8b82a4033283c0f9830",
-    "workingTree": true,
-    "runtime": "Bun 1.3.14 on Linux x64",
-    "browser": "Chrome 140.0.7339.127 via Playwright 1.61.1",
-    "cpu": "12th Gen Intel Core i9-12900H",
-    "samples": 5,
-    "warmup": "none; each cold-route sample used a new browser context",
-    "correctness": "first real feed cell resolved to 1 before each cold sample completed"
+  "protocolVersion": 2,
+  "matrixId": "interaction-current-v2",
+  "mode": "full",
+  "metadata": {
+    "commit": "5b5f8b4d2b296475cceef6cc658437e8f85f2f6a",
+    "dirty": false,
+    "timestamp": "2026-10-06T17:40:23.096Z"
   },
-  "directionalPrefetch": {
-    "sourceLatencyMs": 90,
-    "frameBudgetMs": 16.7,
-    "residencyRatioSamples": [1, 1, 1, 1, 1],
-    "p95VisibleWaitMsSamples": [0, 0, 0, 0, 0],
-    "requestedRowMultiplierSamples": [1.1162790697674418, 1.1162790697674418, 1.1162790697674418, 1.1162790697674418, 1.1162790697674418],
-    "medianResidencyRatio": 1,
-    "medianP95VisibleWaitMs": 0
+  "runner": {
+    "runtime": "Bun 1.4.2 on linux x64",
+    "browser": "153.0.8010.12",
+    "cpu": "12th Gen Intel(R) Core(TM) i9-12900H"
   },
-  "viewIndex": {
-    "rows": 1000000,
-    "lookupsPerRun": 524288,
-    "baseline": {
-      "lookupMedianNsSamples": [37.88488769531501, 39.244018554689816, 44.43444824218934, 45.83947753906622, 44.521118164060624],
-      "lookupP95NsSamples": [45.98823242187404, 54.03151855468516, 101.12755126953121, 62.81248779296663, 83.76065673827702],
-      "retainedBytes": 67125641,
-      "coldBuildMs": 109.215509
-    },
-    "packed": {
-      "lookupMedianNsSamples": [25.080932617187276, 29.146606445313097, 25.03100585937463, 31.878540039062454, 32.43225097656309],
-      "lookupP95NsSamples": [45.79450683593725, 73.24847412109399, 66.49718017578127, 68.38371582031252, 81.93341064453101],
-      "retainedBytes": 4043043,
-      "backingBytes": 4000000,
-      "coldBuildMs": 4.693290000000001
-    },
-    "medianLookupImprovementRatio": 0.3440538231407775,
-    "p95LookupRegressionRatio": 0.08869618483682662,
-    "retainedHeapReductionRatio": 0.9397690220939566
+  "method": {
+    "lookup": "Fresh packed view-index worker process per sample: 1,000,000 rows of one column, 128 batches of 4,096 deterministic lookups (524,288 lookups), nanoseconds per lookup measured over the batches. inverseIndexBytes is the inverse view-index backing store of that worker.",
+    "dirty": "Fresh paged dirty-100 probe process per sample: 1,000,000 rows, 100 distant edits, no clean page loaded. dirty100Bytes is allocated plus sparse dirty bytes retained.",
+    "coldRoute": "Current Vite development route /showcases/performance/, rendered once before sampling so the development server's transform cache is warm, then loaded in a fresh browser context per sample with no shared cache, code cache, or service worker state. Server-side rendering answers the route before the first sample, and a sample measures the browser, not the first server transform; this is a development-server measurement, not a production bundle. usableMs is the page clock when the first rectangular page of feed cells became resident in the public paged store, an upper bound that includes the readiness poll.",
+    "attribution": "Chrome DevTools protocol tracing (devtools.timeline,disabled-by-default-devtools.timeline,disabled-by-default-v8.cpu_profiler) from before navigation until the page is usable, in the renderer process: 50 ms or longer RunTask events are long tasks, and the V8 CPU profile on the same clock says what ran inside them. A task is owned by Sheetwrite when frames served from packages account for at least half of the CPU profiled inside it, and its complete duration is then recorded as the owned long task. Tasks below that share, tasks running only other code, and tasks the profile does not cover are reported as unattributed, so a partly owned task is never counted as owned work. A zero sample means no owned long task was observed.",
+    "previous": "Same-schema comparison only. When the artifact already at this path is a capture of interaction-current-v2, previous records that capture's identity and its per-metric medians, so a reader can state the change without inventing a baseline. A first capture, or an unreadable predecessor, records null."
   },
-  "sparseDirty": {
-    "baseline100Bytes": 5427200,
-    "dirty100Bytes": 5679,
-    "dirty10000Bytes": 718911,
-    "dirty100ReductionRatio": 0.9989536040683962,
-    "dirty100CleanChunks": 0,
-    "dirty10000CleanChunks": 0
+  "rows": 1000000,
+  "lookupsPerRun": 524288,
+  "samples": {
+    "lookupMedianNs": [
+      31.459106445311942,
+      26.614868164063534,
+      22.76940917968727,
+      23.46667480468759,
+      24.449707031250558
+    ],
+    "lookupP95Ns": [
+      442.6306884765626,
+      412.91258544921925,
+      307.31452636718694,
+      406.04642333984384,
+      366.7959350585941
+    ],
+    "inverseIndexBytes": [
+      4000000,
+      4000000,
+      4000000,
+      4000000,
+      4000000
+    ],
+    "dirty100Bytes": [
+      5679,
+      5679,
+      5679,
+      5679,
+      5679
+    ],
+    "ownedColdLongTaskMs": [
+      0,
+      0,
+      0,
+      0,
+      0
+    ],
+    "unattributedColdLongTaskMs": [
+      127.079,
+      118.366,
+      129.117,
+      108.983,
+      112.155
+    ],
+    "usableMs": [
+      1507.2000000001863,
+      1351.5999999996275,
+      1345.5999999996275,
+      1369.5,
+      1297.5
+    ]
   },
-  "coldRoute": {
-    "route": "/showcases/performance/",
-    "before": {
-      "usableMsSamples": [291.3, 351.2, 242.3, 265.5, 250.1],
-      "sheetwriteLongTaskMsSamples": [64.2, 65.8, 56.4, 65, 63.9],
-      "dominantOwner": "eager 100,000-row revenue scenario imported only for four theme fields"
+  "values": {
+    "lookupMedianNs": {
+      "median": 24.449707031250558,
+      "p95": 30.49025878906226,
+      "mean": 25.751953125000178,
+      "stddev": 3.134649315693444,
+      "min": 22.76940917968727,
+      "max": 31.459106445311942,
+      "iters": 5
+    },
+    "lookupP95Ns": {
+      "median": 406.04642333984384,
+      "p95": 436.687067871094,
+      "mean": 387.1400317382814,
+      "stddev": 46.66236755863049,
+      "min": 307.31452636718694,
+      "max": 442.6306884765626,
+      "iters": 5
+    },
+    "inverseIndexBytes": {
+      "median": 4000000,
+      "p95": 4000000,
+      "mean": 4000000,
+      "stddev": 0,
+      "min": 4000000,
+      "max": 4000000,
+      "iters": 5
     },
-    "after": {
-      "usableMsSamples": [231.5, 216.6, 201.3, 215.8, 186.6],
-      "sheetwriteLongTaskMsSamples": [0, 0, 0, 0, 0],
-      "routeCpuMsSamples": [9.209, 9.037, 9.064, 8.057, 7.202],
-      "gridBuildCpuMsSamples": [6.859, 6.268, 6.241, 5.343, 6.407],
-      "reportedUnattributedLongTaskMsSamples": [54, 54, 0, 0, 0]
+    "dirty100Bytes": {
+      "median": 5679,
+      "p95": 5679,
+      "mean": 5679,
+      "stddev": 0,
+      "min": 5679,
+      "max": 5679,
+      "iters": 5
     },
-    "medianUsableImprovementRatio": 0.1871939736346515,
-    "sheetwriteLongTaskCeilingMs": 50,
-    "sheetwriteLongTaskGatePassed": true
+    "ownedColdLongTaskMs": {
+      "median": 0,
+      "p95": 0,
+      "mean": 0,
+      "stddev": 0,
+      "min": 0,
+      "max": 0,
+      "iters": 5
+    },
+    "unattributedColdLongTaskMs": {
+      "median": 118.366,
+      "p95": 128.7094,
+      "mean": 119.14000000000001,
+      "stddev": 7.938793359195081,
+      "min": 108.983,
+      "max": 129.117,
+      "iters": 5
+    },
+    "usableMs": {
+      "median": 1351.5999999996275,
+      "p95": 1479.660000000149,
+      "mean": 1374.2799999998883,
+      "stddev": 70.60122945117568,
+      "min": 1297.5,
+      "max": 1507.2000000001863,
+      "iters": 5
+    }
+  },
+  "previous": null,
+  "ceilings": {
+    "lookupP95Ns": 1000,
+    "inverseIndexBytes": 4000000,
+    "dirty100Bytes": 1048576,
+    "ownedColdLongTaskMs": 50
   }
 }