Jersey L4 sample analyses: Charities (Jersey) Law 2014, and the Covid-19 Gathering Control Orders

[mengwong]

Two Jersey formalisations, both under mengwong/sample-analyses-2026-07/.

charities-jersey-2014/ — the Charities (Jersey) Law 2014

A complete isomorphic encoding of all 43 Articles and both Schedules of the Charities (Jersey) Law 2014 (L.41/2014), from the official consolidated version showing the law from 16 October 2025.

Twelve files, 12,762 lines, 754 assertions. Every file typechecks; every assertion passes.

Written in inert style: the verbatim statutory prose rides inline as bare strings (which evaluate to the identity of their context — TRUE under AND, FALSE under OR — so they cannot change a result), ... is AND-sugar and .. is OR-sugar, and layout is the bracketing. Only the operative atoms are lifted to booleans. Reading down the left margin of a file is reading the Law. Constitutive rules are AND/OR trees; regulative rules use the deontic layer (PARTY x MUST/MAY/SHANT ... WITHIN n) and are exercised with #TRACE.

README.md is a report written for a legislative drafter (Matt Waddington), with sidebars explaining the logic that cannot be hidden behind a low-code surface: identity elements, vacuous truth, material implication, and MUST/MAY omission semantics.

Drafting observations

About twenty interpretation calls are flagged inline with >>> markers. Four look worth an amendment or a policy decision:

• Art 2(1)(ba), inserted by R&O.56/2025 (in force 16 Oct 2025), sits between (b) and (c) — but Art 2(3)(a) still defines "Jersey entity" by the range "(a) to (h)". On the narrow reading, a Jersey-established NPO body corporate is not a Jersey entity, and so could never call itself a "Jersey charity" (Art 23(3)(b)). We encode the purposive reading and flag the gap.
• Art 11(4)/(5): "if not so satisfied" attaches to the satisfaction condition, not to the compliance gate — so an applicant who fails the gate triggers neither a duty to register nor a duty to refuse. Both readings encoded.
• Art 33(5)'s "at both times" is ambiguous as to scope: one ground subsisting at both times, or some ground at each? Both encoded; they diverge on a real scenario.
• Art 34(1) (wrong or unreasonable on merits/facts/law) narrows to Art 35(1) (unreasonable). An appeal wrong on the law but not unreasonable is competent before the tribunal and incompetent before the court.

Also: Art 18(4)(a) read with Art 2(10)(b) would, composed literally, make a governor a misfeasor for having been the victim of an outsider's fraud; an excepted foreign charity always breaches Art 21(1), because Art 22(3)(b) defines it as one that "is not a registered charity"; and Art 4(4) of the consolidation appears to be missing the word "as".

GCO/ — the Covid-19 (Gathering Control) (Jersey) Order 2020

Point-in-time encodings of R&O.166/2020 at three moments — as first enacted, as at 15 January 2021, and as last in force before repeal — on a parallel skeleton so the diffs between versions are visible, with -- DIFF: markers at the divergences. Plus the Covid-19 (Mental Health) (Jersey) Order 2020 as at 15 January 2021.

All four typecheck. GCO-first-version.l4 initially did not — it was mid-refactor between two designs — and is fixed in this branch: PubliclyAccessiblePremisesGathering was used but never declared; its fields had also been folded into Gathering (which caused the ambiguous-projection error); GatheringSituation was dead while the code referred to Article3Situation / Article7Situation; and Gathering carried eight fields the first enacted text has no concept of (private dwelling, open air, workplace, construction site, education, controlled gathering, Christmas period) — all dead, all belonging to the later amendments. Removed, so that the diff between the version files is the diff between the laws.

GCO-README.md is an after-action note on the encoding, also addressed to Matt Waddington, with a frank scorecard separating what any formal language plus an IDE already solves from what L4's legal surface actually adds — and from what stays open because it is not a tooling problem at all.

Note on the branch

smucclaw/sandbox:default and legalese/sandbox:main have diverged substantially (~1,436 / ~1,806 commits). This branch is cut from upstream/default and carries only the three new commits, so it applies cleanly and does not drag the divergence in with it.

🤖 Generated with Claude Code