rule
rule.py checks a text file against one writing rule and reports every break the way a compiler reports a syntax error: file:line:col: error:, the source line, a caret under the word, N errors generated., and exit status 1. The rule changes every day, drawn by the date from a fixed table, or is picked by name from 30. It is one file, Python 3.11 or newer, standard library only: from this folder, uv run rule.py prints today's rule and uv run rule.py draft.txt checks a file against it (python3 rule.py works the same). It takes any UTF-8 text you wrote, as a file or on standard input (-), and reads the local calendar date; a text that keeps the rule prints nothing at all.
The command
rule [--rule NAME] [--date YYYY-MM-DD] [--no-log] FILE check FILE ("-" reads stdin) rule print today's rule rule [--date YYYY-MM-DD] print the rule for a date rule --list the 30 named rules rule --explain NAME [--date YYYY-MM-DD] the exact definition of one rule rule --record print the log and the tally rule --help
Here rule stands for uv run rule.py. To type rule instead, link the file onto your PATH; it is executable and its first line runs it with uv:
ln -s "$PWD/rule.py" ~/.local/bin/rule
From this folder:
uv run rule.py today's rule uv run rule.py draft.txt check draft.txt against it uv run rule.py --rule no-the draft.txt check against a rule picked by name uv run python -m unittest -v the tests uv run python qa/accept.py the acceptance run (needs tmux and git)
For example, the sentence We walked to the market. saved as draft.txt, on a no-k day:
draft.txt:1:7: error: "walked" contains "k" [no-k] We walked to the market. ^ draft.txt:1:21: error: "market" contains "k" [no-k] We walked to the market. ^ 2 errors generated.
Errors go to standard error. The rule line, --list, --explain and --record go to standard output. Exit status: 0 the text kept the rule, 1 it broke the rule, 2 the tool could not run (unreadable file, empty file, invalid UTF-8, unknown rule, bad date). There is no fix option, no ignore option, no suppression comment and no config file. The checked file is opened read-only and never written.
Colour appears only when standard error is a terminal and NO_COLOR is unset: the word error: and the underline in brick, #EC5E52. Nothing else is coloured.
The calendar
The rule for a day is ROTATION[days since 1970-01-01 % 24], on the local calendar date. The rotation holds 24 rules: 8 graded plain (a paragraph passes without noticing) and 16 graded effort (a paragraph passes if the writer pays attention). For the first 15 days of each 24-day cycle the two grades alternate, plain on indices 0, 2, 4, 6, 8, 10, 12 and 14. The last 9 days of every cycle, indices 15 to 23, are all effort.
The 6 hard rules are the Oulipo constraints that make prose close to unwritable. The date never draws them; they are picked by name only. no-<letter> is accepted for any letter a to z. --date gives the rule for any date and is never logged. With --explain, --date stands in for today.
Every daily rule is in the rotation because a written paragraph of at least 40 words of ordinary English keeps it; test_rule.py holds those paragraphs and checks each one.
The 30 rules
This is the output of uv run rule.py --list. uv run rule.py --explain NAME prints the exact definition of each.
Rotation table, revision 1. In force from 1970-01-01. 24 daily rules, 6 by name.
R01 no-dash plain daily - No em dash, en dash, horizontal bar, doubled hyphen or spaced hyphen used as a dash. R02 no-the effort daily - The word "the" does not appear. R03 no-exclaim plain daily - No exclamation mark. R04 no-k effort daily - No word contains the letter "k". R05 short-sentences plain daily - No sentence has more than 14 words. R06 no-first-person effort daily - None of: i, me, my, mine, myself, i'm, i've, i'll, i'd. R07 no-numerals plain daily - No digits; spell numbers out. R08 no-be effort daily E-Prime, D. David Bourland Jr. No form of "to be". R09 no-contractions plain daily - No word ending in n't, 'll, 're, 've, 'm or 'd, and no "let's". R10 no-w effort daily - No word contains the letter "w". R11 no-same-opener plain daily - No two consecutive sentences open with the same word. R12 short-words effort daily - No word has more than 7 letters. R13 no-repeat-adjacent plain daily - No word immediately repeats the word before it. R14 no-alliteration effort daily - Two adjacent words in a sentence do not start with the same letter. R15 no-ly plain daily - No word of three or more letters ends in "ly". R16 no-y effort daily - No word contains the letter "y". R17 even-sentences effort daily - Every sentence has an even number of words. R18 no-and effort daily - Neither the word "and" nor "&". R19 no-v effort daily - No word contains the letter "v". R20 no-repeat-sentence effort daily - No word appears twice in one sentence. R21 no-b effort daily - No word contains the letter "b". R22 no-same-length effort daily - Two adjacent words in a sentence do not have the same number of letters. R23 no-g effort daily - No word contains the letter "g". R24 no-repeated-word effort daily - No word appears twice anywhere in the text. R25 no-e hard by name Perec, La Disparition, 1969 No word contains the letter "e". R26 no-a hard by name Oulipo lipogram No word contains the letter "a". R27 no-t hard by name Oulipo lipogram No word contains the letter "t". R28 snowball hard by name Oulipo snowball In every sentence the word in position n has exactly n letters. R29 univocalic hard by name Perec, Les Revenentes, 1972 (e only); Oulipo univocalism The whole text uses one vowel only. R30 prisoner hard by name Oulipo prisoner's constraint Only letters without ascenders or descenders: a c e m n o r s u v w x z.
no-<letter> is accepted for any letter a to z.
How text is read
- The file is read as strict UTF-8; invalid bytes exit 2. A leading byte order mark is dropped. Each line is normalised to NFC, split on line feeds, with a trailing carriage return dropped.
- A word is a run of Unicode letters and combining marks, with an apostrophe (straight or curly) allowed between two letters. A hyphen, an underscore or a slash ends a word, so
well-knownis two words. A word keeps its own spelling when printed. - A number is a run of decimal digits, in any script, with a single
.or,allowed between two digits. Numbers are not words, sob2bis a word, a number and a word. - The letters of a word, for the length rules, are its letters only:
don'thas 4,naïvehas 5. - Folding, for the letter rules and first letters: casefold, decompose, drop combining marks, then map ß to ss, æ to ae, œ to oe, ø to o, đ to d, ł to l and ı to i. So
écounts ase,çasc,ğasg. Only the Latin letters a to z take part in letter rules. - A sentence ends at a run of
.!?…。!?, with closing quotes or brackets allowed after it, when whitespace or the end of the text follows. A blank line also ends a sentence; a line break alone does not. A period aftermr mrs ms dr prof st vs cf e.g i.eor after a single letter (an initial) does not end one. Numbers and punctuation are not counted as words. - Columns count Unicode code points of the normalised line, starting at 1. The caret line copies the source line's tabs as tabs, so the caret lines up in a terminal.
- All word comparisons ignore capitalisation. Every message quotes the broken word exactly as it was written.
Limits
- Chinese, Japanese and Thai are written without spaces and are not segmented: a run of such text counts as one word. Right-to-left alignment and wide characters under the caret are not handled.
'sis never flagged by no-be or no-contractions, because it may be a possessive.- Sentence splitting knows a short, fixed list of abbreviations and nothing else.
- Markdown, code and commit comment lines are read as plain text.
The log
Only a check of today's rule, with no --rule and no --date, is logged. One line is appended to $RULE_LOG if set, else $XDG_STATE_HOME/rule-of-the-day/log, else ~/.local/state/rule-of-the-day/log: the date, the rule, the exit code, the word count and the break count, separated by single spaces. The file name and the text are never recorded. --no-log skips it, and a run that exits 2 is never logged. The log is append-only, so a broken run stays in the record after a later pass. rule --record prints it with a tally; a day counts as survived if any run that day exited 0.
In git
Check the last commit message:
git log -1 --format=%B | rule --rule no-dash -
As a commit-msg hook, saved as .git/hooks/commit-msg and made executable:
#!/bin/sh exec rule --rule no-dash "$1"
Git's own comment lines, the ones starting with #, are in the file the hook receives, and they are read as text like every other line.
The audit of a repository's whole past, one commit at a time so each report keeps its commit:
git rev-list HEAD | while read -r h; do git log -1 --format=%B "$h" | rule --rule no-dash --no-log - 2>&1 | sed "s/^<stdin>:/$h:/"; done
This checks every message ever committed against a rule that did not exist when they were written.
Help
The last lines of rule --help:
Rules are drawn from a fixed table. A given date always gives the same rule. There is no way to ignore a rule.
Colophon
One file of Python, standard library only, no network. The report format is borrowed unchanged from GCC and Clang. The hard rules come from Perec and the Oulipo, no-be is E-Prime after D. David Bourland Jr., and the test paragraphs retell one walk to a river under each daily rule, after Queneau's Exercices de style. Rotation table revision 1, in force from 1970-01-01. This README was written on 2026-10-13, a no-k day, and checked against all thirty rules. It keeps none of them.
normal is dead
Files
- README.md
- out/accept.txt
- out/commit-100.ans
- out/commit-100.html
- out/commit-46.ans
- out/commit-46.html
- out/explain-100.ans
- out/explain-100.html
- out/explain-46.ans
- out/explain-46.html
- out/failures-100.ans
- out/failures-100.html
- out/failures-46.ans
- out/failures-46.html
- out/help-100.ans
- out/help-100.html
- out/help-46.ans
- out/help-46.html
- out/list-100.ans
- out/list-100.html
- out/list-46.ans
- out/list-46.html
- out/nocolor-100.ans
- out/readme-100.ans
- out/ritual-100.ans
- out/ritual-100.html
- out/ritual-46.ans
- out/ritual-46.html
- out/unicode-100.ans
- out/unicode-100.html
- out/unicode-46.ans
- out/unicode-46.html
- out/unittest.txt
- rule.py
- test_rule.py