A fifth of Analogue Pocket cores are flagged for AI help, and the core list now lets authors declare it

A community checker scores 71 of the 329 openFPGA cores above zero. On 9 October the library behind Pocket updaters added a field for authors to say which AI tools they used.

AI-drafted from 6 sources. This story was written with AI from the outside reporting listed under Sources, and checked by automated tools before it went up. The originals have the full detail.

Seventy-one of the 329 cores in the Analogue Pocket's openFPGA library now carry some sign of AI involvement, according to openfpga-ai-check, a community tool that scores every core in the library and publishes one report per core. Counting the overall score in all 329 reports in the repository on 12 October gives this spread: 258 score zero, 10 score below 0.5, 42 score from 0.5 to 0.99, and 19 hit 1.0, the highest score any core has. That is 21.6% with a score above zero.

A score is a flag, not a finding. The checker looks for three things: commit messages that mention Claude, Codex, GPT or Gemini, or name one of them as author; AI tools listed among a repository's contributors; and README text about AI use. A README mention alone is worth 0.5, and a core's overall score is its single strongest signal. The README says "Currently it's unbounded", and that the aim is a score from 0 to 1, with a fully vibe-coded core scoring 1. Cores whose commits all predate the checker's cutoff are scored zero outright.

Authors can now say it themselves

On 9 October the openFPGA cores inventory, which publishes the core list that Pocket updaters read, merged pull request 1717. It lets a core's author add a declare_ai list of tool names to the core's updater file. Its author wrote that "authors who want their core flagged can only hope the heuristics catch it", and that the MorganVieira.Rally-X core would be the first to use the field. On 12 October it was the only core in the inventory's list that does. The matching change to the checker, which would let a declaration count towards the score, is still an open pull request.

Some authors already say so in public. Michael Merida-Nicolich's Atari Jaguar core, ported from Jaguar_MiSTer and released as version 0.1.1 on 6 October, came with a disclosure in the Reddit announcement: "I developed this core with assistance from Claude." Both of the repository's commits carry a Claude co-author line, and the checker gives it 1.0. Rally-X, whose README names Claude, scores 0.5.

Where players see it

Pocket Sync, the desktop manager for the Pocket, has shown each core's AI score since version 6.4 on 30 August. Version 7.0, released on 10 October, adds an Update tab that installs and updates cores in bulk, with filters. Its developer wrote on Reddit that one filter is for people who do not want "any cores that have been touched by AI".

The checker's README says core developers who think they have been scored wrongly will eventually be able to ask for a check to be zeroed by hand. That list does not exist yet.

Sources

Image: Gaming World chart, data from openfpga-ai-check

More news from Gaming World