Statistics of bitcoin input (script) lengths from august 2015 to august 2017-ish

2017-08-17

Note: this is here just for historical reason. It was written some time in august 2017, in the height of dumb Segwit debates. In 2021, these stats are very much out of date. The script could still work, but segwit added more complexity to already needlessly complex protocol.

I wanted to know, how long inputs are in bitcoin transactions.

The script is here.

It is in node, and it is using bitcoinjs-lib libraries.

Complete results are here (just the final ones), sorted by counts.

tl;dr:

• 84% of inputs are from "normal" 1-addresses
• 16% is from P2SH
• from all the P2SH, almost 100% is multisig
• most popular multisigs are 2-of-3 (67%), followed by 2-of-2 (32%)

Now for some graphs and tables.

All the lengths are lengths of the *input script*, not the whole input. The input itself is always 41 bytes larger. Also note: I skipped the "coinbase inputs". They make 110.000 inputs; there was 110.000 blocks I looked at.

pkh vs sh

This is classified by bitcoinjs-lib, see the code

+-------------+-------------+
|    type     |    count    |
+-------------+-------------+
| pubkeyhash  | 344 968 239 |
| scripthash  |  63 713 612 |
| pubkey      |     354 869 |
| nonstandard |     110 409 |
| multisig    |      11 284 |
+-------------+-------------+

Multisig is not very popular.

edit: as was noted to me on IRC, this is actually native multisig, not the normally used, p2sh multisig. Native multisig is not used for actual multisig, only for adding more data to transactions. That explains the small numbers.

Graph, made in google charts

Pkh input script lengths

Most lengths in PKH are either 106/107 (with scraps around the edges), and a bit 138/139 (again with scraps around edges)

pkh_time.png

pkh_pie.png

See this file for just PKH counts.

Looking deeper into PKH lengths

Most often, the script consists of (1) signature, which is ~70-72 bytes long, encoded using BIP66, and (2) public key

BIP66

The public key can be either compressed or uncompressed, both cases are valid; compressed is 33 bytes long, uncompressed is 65 long.

With the length encoduing (1 byte for sig length, 1 byte for pubkey length), this adds up to the numbers above

• 1 + 72 + 1 + 65 = 139
• 1 + 72 + 1 + 33 = 107

So, 91% of inputs are using compressed private keys, 8% are using uncompressed ones, the rest is doing something strange.

Sh input script lengths

Scripthash input script lengths can be more varied by definition, but surprisingly less than I thought.

Most are in the area around 253 and around 218, +- 2 bytes

However, there is also a very very long tail. Cutting the tail decreases average length around 1 byte. Not much, but not zero.

sh_time.png

sh_pie.png

See SH counts in CSV.

Looking deeper into SH types

I have looked deeper into P2SH statistics.

The script is here.

Results here.

Basically all P2SH inputs are multisig. Those are the most popular types with > 0.01% (note: it's percentage of p2sh inputs, not all inputs.)

+-------------+------------+
|    type     | percentage |
+-------------+------------+
| 2-of-3      | 66.84%     |
| 2-of-2      | 31.36%     |
| 2-of-6      | 0.73%      |
| 3-of-4      | 0.22%      |
| 2-of-4      | 0.21%      |
| 1-of-1      | 0.20%      |
| 3-of-5      | 0.16%      |
| nonstandard | 0.13%      |
| 1-of-2      | 0.06%      |
| 1-of-3      | 0.02%      |
| 4-of-5      | 0.01%      |
| 3-of-3      | 0.01%      |
| 3-of-6      | 0.01%      |
| 1-of-6      | 0.01%      |
| pubkey      | 0.01%      |
| 2-of-5      | 0.01%      |
+-------------+------------+

"nonstandard" are all the weird p2shs that bitcoinjs-lib cannot categorize - like the one below - lumped together.

nonstandard tx

The data check out with the lengths before - the inputs with script lengths "around" 218 are 2-of-2 multisigs, the inputs with script lengths "around" 253 are 2-of-3 multisigs.

msig.png