bitcoin-dust-utxo-lab

Reproducible measurements of the relay policy changes proposed by Bitcoin DOG Mode, run against Bitcoin Core v31.1.

This repository does not argue for or against the proposal. It answers questions that were being asserted on both sides without measurement, and publishes the harness so the numbers can be checked.

1. The 1-sat dust threshold needs no code change

   • -dustrelayfee=0 yields a 1-sat threshold on every standard output type in unmodified v31.1, on any network.

2. Raising the max standard tx weight is not a one-constant change

   • It does not compile. Core has a static_assert chain that forces three coupled constants to move together.

3. What a 1-sat output costs, and who pays it

   • 42.65 bytes of chainstate per output, permanently. A block of dust costs its creator 0.01 BTC and every node operator 1.37 MB forever.

4. The work is the tests, not the constants

   • Against a three line diff: five unit test cases and 21 functional tests. Eleven of the functional failures are the node refusing to start at all, because the enforced -maxmempool minimum moves to 39 MB.

5. Preferential peering works, and it is not free

   • Reserving 4 outbound slots costs exactly 4 inbound slots: 114 to 110 at default -maxconnections. At 25% adoption the reserved slots fill in 15 seconds. But the reserved branch is reached only after 8 full-relay and 2 block-relay slots are satisfied, so a node short of peers never engages the mechanism while still paying the inbound cost. Starting from gossip rather than a pre-filled addrman, it did not engage at all in 25 minutes.

6. What changes at the -maxmempool floor

   • At the floor, one maximum-size transaction is 19% of the mempool instead of 2%. Four transactions raise the minimum feerate 13x, where thirty leave stock's untouched. A control run separates this from the multiplier change itself.

7. The orphan pool ceiling does not move, and a claim I got wrong

   • I argued the weight change loosens an orphan-pool DoS bound. Measured on both builds, the effective ceiling is ~400,000 WU either way, because the per-peer memory reservation binds before the check I was pointing at. Correction, with the numbers.

Raw JSON for every run is in results/.

Requires a Bitcoin Core v31.1 bitcoind binary. Standard library only, no Python dependencies.

git clone https://github.com/bitcoin/bitcoin.git core && cd core
git checkout v31.1
cmake -B build -DCMAKE_BUILD_TYPE=Release -DBUILD_GUI=OFF -DENABLE_IPC=OFF
cmake --build build -j"$(nproc)"
cd ..

python3 scripts/exp1_dust_threshold.py core/build/bin/bitcoind
python3 scripts/exp2_weight_and_cluster.py core/build/bin/bitcoind
python3 scripts/exp3_utxo_growth.py core/build/bin/bitcoind

Each script starts and tears down its own isolated regtest node and writes JSON to results/.

.github/workflows/experiments.yml builds Bitcoin Core v31.1 and reruns the fast experiments on every push, then asserts that they still reproduce the figures quoted in the findings. A harness change that quietly stops measuring anything fails there rather than going unnoticed. The long-running experiments (UTXO growth, peer discovery) are reproduced locally instead; see METHODOLOGY.md.

METHODOLOGY.md states the full procedure, the metrics, and the known limitations, including which experiments were run before the method was written down.

Tarık İsmet ALKAN GitHub: @tialkan · X: @tialkan

MIT licensed.