Nockchain has shipped five forks since July 25. Notice ran from 2.7 to 77.1 hours.

At 15:47 UTC on July 25, the @nockchain account made a public promise: “From here on out, every consensus-critical upgrade takes the same path: public testnet, multiple-entity review, published audit, then an activation height announced with real notice.”¹ This means every consensus-critical upgrade should get a public testnet, review by multiple entities and a published audit. Then its activation height should be announced with real notice.

Since then, five consensus upgrades have activated, each at its own height. This piece measures one part of that promise: how much notice miners and node operators got. Notice here runs from the first public signal that names the final activation height to the timestamp of the activation block on Nockscan. A signal counts if it is a pull request opened, a commit pushed to the default branch, a tag, or an official post naming the height.

Fork	Height	First public signal (UTC)	Activated (UTC)	Notice	Emergency
Block 112,500 fix	112,500	Jul 25, 21:04	Jul 26, 02:25	5.3 h	Yes
Zoe	119,400	Aug 3, 23:00	Aug 4, 01:39	2.7 h	Yes
Logos	126,000	Aug 10, 20:58	Aug 14, 02:05	77.1 h	No
Anthropos	147,500	Sep 14, 20:36	Sep 15, 07:20	10.7 h	No
154,500 reset	154,500	Sep 25, 21:43	Sep 26, 12:29	14.8 h	No

The two emergency fixes

The first fix came the same evening. At 21:04 UTC on July 25, about five hours after the promise, a commit went straight to the default branch with no pull request.² It activates a fix at block 112,500.² Block 112,500 arrived 5.3 hours later.¹³

The fix closed a mining channel built on zero-padded composition polynomials and re-anchored the ZK difficulty target.² The repository’s protocol changelog has no specification file for it. The @nockchain account later referred to it as “the coefficient-padding channel closed at 112,500”.³

Zoe followed on August 4. Its pull request opened at 12:26 UTC on August 3 with a proposed height of 120,400.⁴ It merged into the default branch at 23:00 UTC naming block 119,400, and the block arrived at 01:39 UTC, 2.7 hours later.⁴¹³

A commit naming 119,400 had been pushed to the open pull request at 18:24 UTC.⁴ Under the definition above, the merge is the first signal that counts.⁴

Zoe fixed a trace-reuse mining exploit.⁵ The @nockchain account wrote afterward: “Every proof produced this way was a genuinely valid proof.” It added: “Impact was limited to block rewards captured at an inflated effective proof rate.”³

The three planned upgrades

Logos set the high mark for notice. The @nockchain account named block 126,000 at 20:58 UTC on August 10, 77.1 hours before it arrived.⁶¹³ Work on the code continued inside that window.

Eleven commits went to the default branch without pull requests in the final 72 hours. Two were AI proof-of-work changes, one of them a soundness fix.⁷ Its commit says the old code had allowed “a full mining proof whose matmul ran on a matrix the block never committed to”.⁷ The production CUDA miner for the new puzzle merged 11 minutes before block 126,000.⁸

Anthropos opened as pull request #187 at 20:36 UTC on September 14 with height 147,500, and it merged at 23:10 UTC.⁹ It activated at 07:20 UTC on September 15, which gave 10.7 hours of notice.⁹¹³ It moved ZK mining to proof version 5.¹⁰ It also changed the block split “from Logos’s 30% AI / 70% ZK target to 70% AI / 30% ZK”.¹⁰

The 154,500 reset opened as pull request #209 at 21:43 UTC on September 25 and merged 26 minutes later.¹¹ It activated at 12:29 UTC on September 26, which gave 14.8 hours of notice.¹¹¹³ It re-anchored both difficulty lanes and revised how AI proofs are verified.¹² Its cutover notes add: “Bridge withdrawal processing must remain disabled.”¹²

Nocktane found no @nockchain post naming 112,500, 147,500 or 154,500 before those blocks arrived. If one exists and came earlier than the GitHub signal, those notice figures grow, and we will correct the log.

Releases and review

None of the five has a git tag.¹⁴ The newest tag in the repository points to a commit from May 15, and the repository has never published a GitHub Release.¹⁴²²

From July 20 to September 30, 58 commits reached the default branch, and one account authored 49 of them.¹⁵ Thirty-seven reached the branch without a pull request, and the same account pushed all 37, including the 112,500 activation.¹⁵ The last commit on the default branch is the 154,500 release, merged September 25.¹⁵

Pull request #203, titled “Security advisory bundle”, merged 16 minutes after it opened on September 22.¹⁶ The files it changes cite 15 GitHub security advisory identifiers. The repository lists no published security advisories as of October 10.¹⁷

How the split played out

Each of the three planned upgrades moved the share of blocks each puzzle wins. Here is how each one compared with its target. Counting blocks by puzzle type on Nockscan, AI puzzles won 29.5% of blocks under Logos (target 30%), 75.9% under Anthropos (target 70%) and 70.4% from block 154,500 to block 162,773 (target 70%).¹⁸

Over the seven days to 12:09 UTC on October 10, one payout address won most ZK blocks. The top ZK prover won 60.6% of 1,209 ZK blocks, and the top five won 98.1%.¹⁹ In the AI lane, the top miner won 45.7% of 2,762 blocks and the top two won 78.2%.¹⁹ Method: a prover is identified by its first payout address, and the protocol fund’s fixed share inside every block is excluded.

The top ZK address’s daily share has fallen. It ran from 69% to 82% from October 2 to October 6.¹⁸ From October 7 to October 10, it ran from 41% to 54% (October 10 counted to 12:09 UTC).¹⁸

The docs have some catching up to do. The Compute Networks page still describes a target of about 30% AI and 70% ZK blocks, and no docs page mentions Anthropos.²⁰ The repository’s PROTOCOL.md still lists Zoe and Anthropos as draft, although both have activated.²¹

The full table, with every source, lives in the Fork Notice Log. We will add the next fork when its height is first named. That fork has room to match the 77.1 hours of notice Logos gave.