Task: calibrate allocator parity and freeze the matched 0-150k baseline

[metaphorics]

Parent

#16

What to build

Establish the canonical processing-bound comparison before any new optimization lands. Measure otherwise-identical bitcoin-rs replay binaries with the current replay allocator and production-matched mimalloc, interleaved with fresh Bitcoin Core controls. Freeze the winning allocator posture, source SHA, binaries, compiler, flags, corpus, validation posture, and correctness digests for every dependent ticket.

This is calibration, not a silent baseline reset. Preserve both allocator panels. Use the same local block corpus for both nodes so REST fetch work and a second competing process cannot distort the result.

Blocked by

None. This is the execution frontier.

Acceptance criteria

• Both bitcoin-rs arms use the same source, release profile, default fjall+kernel feature posture, compiler, flags, and taskset -c 0-31; allocator is the only changed mechanism.
• Run three interleaved control/candidate pairs and report median wall, user CPU, system CPU, mean-core use, and peak RSS.
• Run fresh Bitcoin Core 31.0 controls in the same idle sessions against the same 0→150,000 local block corpus with full verification.
• Record exact source and binary hashes, rustc/linker versions, block corpus identity, stop hash, UTXO digest, and CoinStats MuHash.
• Reject every run that overlaps a build, another replay/IBD, or abnormal host load.
• Select the canonical allocator posture explicitly. If mimalloc changes either axis materially, keep both panels and use production-matched mimalloc for all later controls.
• Publish a machine-readable baseline packet that dependent tickets can consume without reconstructing hidden state.

[coderabbitai]

🔗 Related PRs

#7 - Parallelize block input script checks for performance improvement [merged]
#14 - feat: header timestamp consensus rules, index reorg rollback, mempool policy, fuzz targets [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[metaphorics]

Completed in stacked PR #29.

Three interleaved rounds, local blocks 0→150,000, full verification, fjall, taskset -c 0-31, direct-child wait4:

Arm	Wall median	CPU median (user+system)	Peak RSS median
system allocator	63.43s	399.63s	573,440,000 B
production-matched mimalloc	56.16s	396.50s	664,252,416 B
Core 31.0	64.74s	477.82s	680,341,504 B

Mimalloc is the canonical control. It clears the wall gate at 1.129x, changes CPU only 1.008x, and raises RSS 15.8%. Against the fresh Core median, the canonical bitcoin-rs panel leads 1.153x wall and 1.205x CPU.

Custody:

• source ff2615a2946cfdf980ed80ed14c0ad5986631d8a, Rust 1.95.0;
• corpus SHA-256 c0a2b9aa35498dacdf1aec792a1b68d9ec68e372fba88ef0e62bbccb1732fe2c;
• stop hash 0000000000000a3290f20e75860d505ce0e948a1d1d846bec7e39015d242884b in every arm;
• system and mimalloc validation artifacts byte-identical;
• both match Core at height 150,000 for MuHash, UTXO count, total amount, and stop hash;
• workspace tests: 1,334 passed, 15 ignored, 0 failed;
• clippy -D warnings passed for both feature variants;
• independent review: no findings.

Machine packet: docs/benchmarks/data/end-to-end-sync/allocator-custody-v1.json in #29.