The reviewer’s copy — what it is, and how to check it
model_act_v3_4_reviewers_copy.pdf is the statute set for print and annotation. It is not a typeset edition and makes no design claims: monospaced text, one printed line per source line, numbers in the left margin, and a wide right margin to write in.
Why line numbers. Legislative drafting is discussed by line — “page 4, line 17, strike.” The numbers here are the source file’s own line numbers, so p. 6, l. 236 in a marked-up printout and model_act_v3_4.txt#L236 name the same text. A disposition written on paper travels back into the repository without translation.
Overflow. Source lines longer than the printed measure continue on the next row, indented and unnumbered. Nothing is truncated; the numbered row always begins its source line.
Reproducible, which is the point. The predecessor typeset edition was withdrawn because nobody could prove it matched the statute (ERRATA E7). This one proves it by construction: it is generated by archive/build_reviewers_copy.py from the authoritative text, with dates pinned to the v3.4 tag and the PDF’s document ID derived from the source’s own hash, so nothing depends on the build clock or the machine.
Check it yourself:
python3 archive/build_reviewers_copy.py model_act_v3_4.txt /tmp/check.pdf
sha256sum /tmp/check.pdf archive/model_act_v3_4_reviewers_copy.pdf
Two identical hashes mean the PDF in this folder is the statute in the root, and nothing else.
| Source | model_act_v3_4.txt, sha256 399c725adcd117aa7736a63b716328226eb24f33a48695115d941b68caace1bf |
sha256 b355a0241c8126effe2ac9967fc58b6555f3cf192d11d207bb91284d0bd1d570 | |
| Extent | 612 source lines · 17 pages · US Letter, landscape |
| Generated | 20 August 2026 against the v3.4 tag |
If a rebuild ever disagrees with these hashes, that is an erratum and we want to hear about it: FrontierAIAccountabilityProject@proton.me.
One known stale word is preserved in this text on purpose — the register’s E10 explains why correcting it would cost more than it fixes.