Rewrite public guidance in plain language without changing the law

For: Content lead or service owner responsible for a large body of public guidance pages

Pattern: TournamentNeeds live modelsDesigned for 10 to 400 agents

The pain today

Hundreds of guidance pages are hard to read, and every rewrite risks changing what the rule means. Legal review of each rewrite is the bottleneck, so pages stay as they are.

The ask

I attached our guidance pages and the legislation and policy each one is based on. For each page, write several plain-language versions, throw out any that change, drop or add a rule compared with the source, and give me the best surviving version with a list of every meaning check it passed and any doubt that remains.

Plain words, as you would say it to a colleague. Edit it to fit your case before you send it.

What you attach or connect

  • Current guidance pages as text
  • Source legislation and policy per page
  • House style and plain-language standard

The unit of work

One worker task per one guidance page with several candidate rewrites.

Why a swarm fits

Pages are independent, and candidates are cheap to write but expensive to check. Many writers and independent fidelity judges per page narrow the field before a lawyer reads anything.

Not for

Publishing without review, or pages where the underlying policy is itself unsettled. A single page is better rewritten by its owner.

The decision tree

6 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

  1. Planner, while planning

    Scope checkYes or no, with a probability

    Before work starts on a unit

    Is at least one source instrument or policy document supplied and cited for this guidance page?

    Sees only: The page's source list and the supplied documents index

    Why: A page with no source cannot be checked for fidelity and is returned unchanged.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenEscalate to a strong model
    • No: below 0.30thenSkip this unit
  2. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Is each rule, condition and deadline in the planner's list for this page backed by a quoted passage from the supplied source?

    Sees only: The rule list for one page with its source quotes

    Why: The fidelity check is only as good as the rule list, so the list is verified first.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry
  3. Evidence checkA choice among options

    After a worker answers

    Does the candidate rewrite state this rule with the same condition, the same deadline and the same force, such as must rather than should?

    Sees only: One listed rule with its source quote and the matching sentence of one candidate

    Why: A candidate that alters any rule is eliminated regardless of how well it reads.

    • Same condition, deadline and forcethenAccept
    • Softened, narrowed or widenedthenSkip this unit
    • Rule missing from the candidatethenSkip this unit
    • Wording is ambiguousthenEscalate to a strong model
  4. Evidence checkA choice among options

    After a worker answers

    Does the candidate contain a requirement, entitlement or deadline that is not on the rule list?

    Sees only: One candidate and the rule list for its page

    Why: Stops helpful-sounding additions that the law does not contain.

    • Nothing addedthenAccept
    • Something addedthenSkip this unit
  5. Run control, between rounds

    Another round?A score

    Between rounds

    How closely does this surviving candidate follow the house style standard's rules on sentence length, active voice and defined terms?

    Sees only: One surviving candidate and the style standard

    Why: Runs further rewrite rounds only while survivors still fall short of the style standard.

    • High: 0.85 or higherthenStop
    • Middle: 0.50 up to 0.85thenContinue
    • Low: below 0.50thenContinue
  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Is this a recommended rewrite, a page where current guidance differs from its source, or a page where no candidate survived?

    Sees only: One page's result with its fidelity table

    Why: The policy owner and a legal adviser approve each page; nothing is published by the swarm.

    Accountable: The policy owner and a legal adviser approve each page before publication; the content lead owns the style decisions.

    • Yes: 0.40 or higherthenAsk a person
    • Unsure: 0.15 up to 0.40thenAsk a person
    • No: below 0.15thenAccept

The fleet: who does what

Model tiers by role, not brands: you choose the models. Strong reasoning models plan and reconcile, small fast models do the wide work, and the judge is a decision model from a different family, so it does not share the workers' blind spots.

  1. Planner

    A strong reasoning model lists the rules, conditions and deadlines each page must preserve, quoted from the source.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from open-weight families each write a candidate rewrite of one page to the style standard.

    Designed for 10 to 400 agents, one worker task per one guidance page with several candidate rewrites. Each worker receives only its own unit.

  3. Judge, from a different model family

    Judges from a different family eliminate candidates that alter any listed rule, then rank survivors for clarity.

    Decisions here:2. Evidence check3. Evidence check4. Evidence check

  4. Reconciler

    A strong reasoning model presents the winning version per page with its rule-by-rule fidelity check and open doubts.

    Decisions here:5. Another round?

  5. Accountable person

    The policy owner and a legal adviser approve each page before publication; the content lead owns the style decisions.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Each preserved rule is matched to a sentence in the rewrite and a quote from the source
  • A candidate that adds, drops or softens a rule is eliminated regardless of readability
  • Pages where no candidate survives are returned unchanged with the reason

What comes back

  • Recommended rewrite per page
  • Rule-by-rule fidelity table against the source
  • Pages where the current guidance already differs from the source
  • Pages returned unchanged

What to measure

  • Rewrites approved by legal review without changes
  • Meaning changes legal review caught that judges missed
  • Reviewer time per page
  • Cost per page

Names of measures only. No result is claimed for this template.

Templates open in the workspace chat with the ask filled in. Nothing runs until you send it.

Get early accessSign in to use

Code consultation responses by question and keep minority views

For: Policy analyst or consultation lead at a ministry, regulator or municipality

A public consultation brings thousands of free-text responses, many from campaigns, some from experts with one decisive point.

Pattern: Map, verify, reduceNeeds scale6 decisionsDesigned for 40 to 800 agents

Prepare tender evaluation evidence with two independent readings

For: Procurement officer or evaluation panel chair in a public contracting authority

Evaluators must score long submissions against published criteria and defend every score if challenged.

Pattern: Cross-examinationNeeds live models6 decisionsDesigned for 6 to 300 agents

Trace a changed legal definition through the statute book

For: Legislative drafter, government lawyer or policy adviser preparing an amending bill

Changing one defined term can ripple through acts, regulations and guidance that use or cross-refer to it.

Pattern: Hierarchical decompositionNeeds a connector6 decisionsDesigned for 30 to 600 agents