Where: llms.txt line 9, llms-full.txt line 7 (generated by scripts/build_explainers.py's build_llms_full(), around lines 627-628).
The gap: both files tell any AI crawler/answer engine reading them that the paper freeze is still active and new audits are paused:
Repro:
$ grep -n "peer review\|paused" llms.txt llms-full.txt scripts/build_explainers.py
llms.txt:9:> Status: the benchmark results are frozen for a research paper in peer review (tag `v1.0-paper`). New audits are paused until publication...
llms-full.txt:7:...The benchmark results are frozen for a paper in peer review (tag v1.0-paper); new audits are paused, explainers continue.
scripts/build_explainers.py:628: "review (tag v1.0-paper); new audits are paused, explainers continue.",
vs. CLAUDE.md's current policy: "Paper freeze - lifted... New audits merge to main like anything else" - and this repo's own history shows real audits have merged since (e.g. Tenant Screening's related work, several post-freeze fixes).
Why it matters: these two files are the public-facing index served to AI crawlers/answer engines at thefaircode.xyz/llms.txt specifically so they get an accurate, current picture of the project - and they're actively telling any LLM reading them the opposite of current, repo-documented policy. Because the stale sentence is hardcoded inside the generator itself (not just the checked-in output), scripts/check_generated_files_current.py's "regenerate fresh, then diff against HEAD" check can never catch this - regenerating just reproduces the same wrong text byte-for-byte.
Suggested fix: update the hardcoded string in build_llms_full() (scripts/build_explainers.py lines ~627-628) and the hand-written llms.txt line 9 to reflect CLAUDE.md's current status, then run make build-explainers to regenerate both files.
Where:
llms.txtline 9,llms-full.txtline 7 (generated byscripts/build_explainers.py'sbuild_llms_full(), around lines 627-628).The gap: both files tell any AI crawler/answer engine reading them that the paper freeze is still active and new audits are paused:
Repro:
vs.
CLAUDE.md's current policy: "Paper freeze - lifted... New audits merge tomainlike anything else" - and this repo's own history shows real audits have merged since (e.g. Tenant Screening's related work, several post-freeze fixes).Why it matters: these two files are the public-facing index served to AI crawlers/answer engines at
thefaircode.xyz/llms.txtspecifically so they get an accurate, current picture of the project - and they're actively telling any LLM reading them the opposite of current, repo-documented policy. Because the stale sentence is hardcoded inside the generator itself (not just the checked-in output),scripts/check_generated_files_current.py's "regenerate fresh, then diff against HEAD" check can never catch this - regenerating just reproduces the same wrong text byte-for-byte.Suggested fix: update the hardcoded string in
build_llms_full()(scripts/build_explainers.pylines ~627-628) and the hand-writtenllms.txtline 9 to reflect CLAUDE.md's current status, then runmake build-explainersto regenerate both files.