Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/aer-consistency
Use when auditing a finished or near-finished AER, AER:Insights, or AEJ manuscript for internal consistency: headline numbers across abstract, introduction, results, and tables; sample sizes; log-point and percentage-point conversions; cross-references; and in-text-citation/bibliography matching. Apply after the body and exhibits exist, before aer-referee-sim and aer-submission.
Passed 15 of 21 test cases with the skill, and 9 of 21 without it.
and 6 more in the full report
7 fixed9 passed either way6 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
The bibliography file main.bib contains 45 bibtex entries. Processing main.tex reveals that 43 references are cited in text, leaving 2 entries without in-text citations. What is the Audit 6 result?
| Checked | Without skill | With skill |
|---|---|---|
| The audit marks Audit 6 as FAIL. | Passed | Passed |
| The report specifies that every bibliography entry must be cited at least once in text. | Missed | Passed |
The instructions your model reads when the skill loads.
Referees and editors run cheap integrity checks before engaging with ideas: does the abstract's number appear in the tables? Do the Ns add up? Does "Table 4" exist? Does every citation resolve? A single mismatch reframes the entire report from "is this right?" to "what else is wrong?" — and for AI-assisted manuscripts these mismatches are the modal failure, because text and tables are often generated in separate passes.
This skill is the full-manuscript integrity audit. It is mechanical by design: every check below has a yes/no answer obtained by comparing two artifacts, not by judgment. Run it after every revision round, not only before first submission.
aer-referee-sim (so the simulated referees attack substance, not
typos) and before aer-submissionBuild a register of every number that appears more than once in the manuscript, then verify each row against its single source of truth (the table or the replication output):
NUMBER SOURCE ABSTRACT INTRO RESULTS CONCL MATCH
4.2 log points Tab 3 col 4 yes yes yes yes OK
s.e. 1.1 Tab 3 col 4 yes no yes no OK
$84 billion cited source no yes no yes OK
N = 37,824 Tab 1 no no yes no OK
Rules:
test cases passed, vs 9 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.