bench(core): add memory receipts

docs/bench/core-memory-2026-09-05.json

@@ -0,0 +1,28 @@
+{
+  "command": "cargo run --manifest-path engine/core/Cargo.toml --example membench --quiet",
+  "toolchain": {
+    "rustc": "rustc 1.97.1 (8bab26f4f 2026-07-14)",
+    "cargo": "cargo 1.97.1 (c980f4866 2026-06-30)"
+  },
+  "git_revision": "b7dd46f",
+  "origin_main_revision": "2d20ddad228db52fa29ad7d7d06f44e8672164af",
+  "benchmark_implementation_revision": "0821ddf",
+  "workload": {
+    "shape": "99 nodes; 24 steady ticks; 8 rounds of 4-node subtree churn; 16 text ticks; 12 burst ticks",
+    "text_mode": "atlas",
+    "texture_mode": "atlas"
+  },
+  "receipt": {
+    "peak_requested_bytes": "29591",
+    "final_requested_bytes": "10793",
+    "allocation_count": "11682",
+    "total_allocated_bytes": "6225569",
+    "avg_layout_us": "1210",
+    "max_layout_us": "5163",
+    "nodes": "99",
+    "structural_relayouts": "16",
+    "text_mode": "atlas",
+    "texture_mode": "atlas",
+    "drawlist_checksum": "cc6a0b00efdba151"
+  }
+}

docs/bench/core-memory-2026-09-05.md

@@ -0,0 +1,39 @@
+# Core Memory Benchmark Baseline
+
+The benchmark harness was based on the latest `origin/main` revision
+`2d20ddad228db52fa29ad7d7d06f44e8672164af`; the first receipt was measured at
+harness commit `b7dd46f`. The benchmark implementation was introduced
+separately at `0821ddf`. `git_revision` identifies the revision where the
+receipt was measured, while `origin_main_revision` identifies the source
+revision used as the harness base. This receipt was not measured from
+`origin/main` itself.
+
+The byte values below are **allocator-independent requested-byte metrics** from
+the counting allocator. They are not process RSS, committed pages, or PSP arena
+high-water measurements. Timing values are host measurements and are excluded
+from deterministic receipt comparisons.
+
+## Reproduction
+
+```text
+cargo run --manifest-path engine/core/Cargo.toml --example membench --quiet
+```
+
+Toolchain: `rustc 1.97.1 (8bab26f4f 2026-07-14)`, `cargo 1.97.1 (c980f4866 2026-06-30)`.
+
+Workload shape: 99 nodes; 24 steady ticks; 8 rounds of 4-node subtree churn;
+16 text ticks; 12 burst ticks. Text and texture modes are both `atlas`.
+
+## Receipt
+
+| Field | Value | Meaning |
+| --- | ---: | --- |
+| `peak_requested_bytes` | 29591 | Peak requested bytes |
+| `final_requested_bytes` | 10793 | Final live requested bytes |
+| `allocation_count` | 11682 | Count of measured allocation events |
+| `total_allocated_bytes` | 6225569 | Total requested bytes across measured allocation events |
+| `avg_layout_us` | 1210 | Host timing, excluded from canonical comparison |
+| `max_layout_us` | 5163 | Host timing, excluded from canonical comparison |
+| `nodes` | 99 | Workload node count |
+| `structural_relayouts` | 16 | Structural relayout count |
+| `drawlist_checksum` | `cc6a0b00efdba151` | Deterministic drawlist checksum |

docs/bench/core-memory-ppsspp-2026-09-05.json

@@ -0,0 +1,46 @@
+{
+  "schema_version": 1,
+  "status": "PASS",
+  "selected_app": "stats",
+  "workload_role": "representative workload for the shared workload profile; not a byte-identical event sequence",
+  "framework": "solid",
+  "samples": 3,
+  "report_path": "dist/bench/ppsspp-bench-2026-09-05T05-07-18-714Z.json",
+  "checksum": "c88e7bcedc5d42a5",
+  "commands": {
+    "baseline": {
+      "command": "PSP_SDK=\"$PSP_SDK\" BENCH_PPSSPP_TIMEOUT=60 bun tools/bench-ppsspp.ts --apps=stats --samples=3",
+      "status": "PASS",
+      "exit_status": 0
+    },
+    "memory_scan": {
+      "command": "PSP_SDK=\"$PSP_SDK\" BENCH_PPSSPP_TIMEOUT=60 bun tools/bench-ppsspp.ts --apps=stats --samples=3 --memory-scan",
+      "status": "PASS",
+      "exit_status": 0
+    }
+  },
+  "execution_context": {
+    "checkout": "bench/core-memory-receipt",
+    "git_revision_used_for_both_attempts": "3e0b772",
+    "later_metadata_commits_benchmarked": false
+  },
+  "revisions": {
+    "git_revision": "3e0b772",
+    "ppsspp_revision": "f929a74"
+  },
+  "metrics": {
+    "arena_bump_bytes": 2654944,
+    "arena_tail_free_bytes": 15232288,
+    "avg_tick_us": 516,
+    "avg_work_us": 4686,
+    "max_work_us": 62816,
+    "uncapped_arena_bump_bytes": 2654944,
+    "min_pass_arena_bytes": 2883584,
+    "safe_arena_bytes": 3670016
+  },
+  "field_definitions": {
+    "report_path": "Expected generated JSON report path under dist/bench/ (ppsspp-bench-<timestamp>.json); null means no report was generated.",
+    "checksum": "Checksum of the representative workload drawlist used by the run; null means no benchmark ran. Compare with the Rust profile and report differences explicitly."
+  },
+  "caveat": "PPSSPP measurements are emulator evidence, not real PSP hardware proof. Arena high-water is a practical allocator capacity requirement, not a precise live-object heap profile. PSP checksum c88e7bcedc5d42a5 differs from the Rust baseline checksum cc6a0b00efdba151 because the PSP journey is not byte-identical to the synthetic Rust workload."
+}

docs/bench/core-memory-ppsspp-2026-09-05.md

@@ -0,0 +1,77 @@
+# Core Memory PSP Receipt
+
+Status: **PASS**
+
+Receipt schema version: **1**
+
+Both benchmark commands reached the `stats` PSP app under PPSSPPHeadless.
+
+The PSP target is the **`stats` representative workload**, using the existing
+fixed input script and PSP JSONL schema. It corresponds to phases in the Rust
+synthetic workload profile; the two journeys are not byte-identical event
+sequences. No second demo or Taffy A/B run was added.
+
+## Commands
+
+Both commands ran on revision `3e0b772`. The PPSSPP source revision was
+`f929a74`.
+
+```bash
+git status --short --branch
+```
+
+Baseline:
+
+```bash
+PSP_SDK="$PSP_SDK" BENCH_PPSSPP_TIMEOUT=60 bun tools/bench-ppsspp.ts --apps=stats --samples=3
+```
+
+Status: **PASS**; exit status: `0`.
+
+Memory scan:
+
+```bash
+PSP_SDK="$PSP_SDK" BENCH_PPSSPP_TIMEOUT=60 bun tools/bench-ppsspp.ts --apps=stats --samples=3 --memory-scan
+```
+
+Status: **PASS**; exit status: `0`.
+
+## Revisions
+
+| item | value |
+|---|---|
+| git revision | `3e0b772` |
+| PPSSPP revision | `f929a74` |
+| selected app | `stats` |
+| framework | `solid` |
+| samples | `3` |
+| report path | `dist/bench/ppsspp-bench-2026-09-05T05-07-18-714Z.json` |
+| checksum | `c88e7bcedc5d42a5` |
+
+`report_path` points to the authoritative generated JSON report under
+`dist/bench/`. `checksum` is the representative workload drawlist checksum.
+The PSP checksum `c88e7bcedc5d42a5` differs from the Rust baseline checksum
+`cc6a0b00efdba151`; the workloads are not byte-identical.
+
+## Metrics
+
+| metric | value |
+|---|---:|
+| arena bump | `2,654,944` bytes (`2.53 MiB`) |
+| arena tail free | `15,232,288` bytes |
+| avg tick | `516us` |
+| avg work | `4,686us` |
+| max work | `62,816us` |
+| uncapped arena bump | `2,654,944` bytes |
+| minimum passing arena | `2,883,584` bytes (`2.75 MiB`) |
+| safe arena | `3,670,016` bytes (`3.50 MiB`) |
+
+## Environment
+
+The fixed PSP SDK was loaded from `/Users/quake/.cache/pocket-stack/psp/env.sh`.
+`PPSSPPHeadless` was built at `/Users/quake/ppsspp-src/build/PPSSPPHeadless`.
+
+**PPSSPP results, when available, are emulator evidence rather than real PSP
+hardware proof.** `arena_bump_bytes` is a practical allocator capacity
+requirement and includes size-class fragmentation; it is not a precise
+live-object heap profile.