Guidelines › Bill Drafting Auditor
Bill & Act Drafting Compliance Auditor
Upload a drafted bill or statutory clauses and benchmark them against national and supranational legislative drafting manuals. Produces a 1–10 compliance score, dimensional sub-scores, a rule-by-rule checklist with official manual citations, and concrete redraft suggestions.
How the auditor works
The auditor is a rule engine that runs entirely in your browser (assets/auditor.js). Your text is not uploaded anywhere unless you add an AI provider key below. The same engine is available offline as scripts/evaluate_bill.py.
Steps
- Read the text. XML or HTML tags (Akoma Ntoso, USLM) are stripped, the text is split into sentences, and any recitals (lines from Whereas or Having regard to up to the enacting formula or the first section) are separated from the operative provisions.
- Load a rule profile. The ten featured guidelines have hand-written profiles: legal tradition, policy on shall, target average sentence length (26–32 words), recitals policy, and the manual sections cited in results. Every other catalogued guideline gets a profile derived from its country code and jurisdiction level (common law, civil law, bijural, supranational or international).
- Run the checks. Each check produces
PASS,WARNorFAILand deducts points from one or more sub-scores, each starting at 10. - Score. Sub-scores are clamped to 1–10 and combined into the overall score.
Checks
| Check | What it looks for | Sub-score |
|---|---|---|
| Normative verbs | Depends on the profile. UK, Commonwealth, Australia, NZ and similar: any shall (use must or the present tense). EU: shall in operative articles, and no shall/must in recitals. US HOLC: shall with an inanimate subject, or declarative shall mean / shall be deemed. Civil law: shall instead of the normative present tense. | Normative verbs |
| Definitions hygiene | Definitions that impose duties (a shall or must inside a definition) and the phrase means and includes. | Definitions |
| Plain language | A list of archaic terms and doublets such as herein, thereof, aforesaid, forthwith, whosoever, null and void, and/or, deemed to be. | Clarity |
| Sentence length | Average sentence length against the profile's target, and any sentence over 42 words not broken into (a), (b), (c) paragraphs. | Clarity |
| Active voice | Passive constructions with no named actor (shall be imposed, must be paid without by …). | Clarity, Normative verbs |
| Gender-neutral drafting | Masculine-generic pronouns, paired he or she, and gendered titles such as Chairman. | Conventions |
| Structure | Numbered sections or articles, a commencement provision in longer texts, the EU citations–recitals–enacting formula sequence, and no recitals inside a civil law statute. | Structure, Conventions |
| Cross-references | Positional references such as above, below, the preceding section, hereinbefore. | Definitions, Structure |
Scoring
Overall score = Clarity × 25% + Normative verbs × 25% + Definitions × 20% + Conventions × 15% + Structure × 15%. The exact index is shown to one decimal place and rounded for the headline score. Bands: 8.0 and above is high compliance, 5.5–7.9 moderate, below 5.5 low. Each flagged provision comes with a suggested redraft generated from the same rule (for example, shall rewritten as must).
Comparison and AI critique
Ticking guidelines under Multi-Jurisdiction Comparison runs the same text through each profile, so the same draft can score differently under the UK OPC (which discourages shall) and the EU Joint Practical Guide (which expects it in articles). The optional AI critique sends the first 12,000 characters of the text and the guideline name to Gemini (gemini-2.5-flash) or OpenAI (gpt-4o-mini) with your key and shows the response separately; it does not change the score.
Limitations
The checks are pattern matches on English text. They do not assess legal substance, policy or constitutional validity, and they can miss problems or flag acceptable usage. Drafts in other languages are scored, but most checks look for English wording. Treat the score as a drafting-style indicator for review by counsel.