Answer a cross-border policy question, both readings shown

For: HR business partner or people operations adviser answering policy questions

Pattern: Cross-examinationRuns todayDesigned for 3 to 24 agents

The pain today

A question about leave or benefits for someone moving between countries touches several policies that do not agree. Advisers answer from memory, and two advisers give two answers.

The ask

I attached our leave, mobility and benefits policies for both countries. What do they say about parental leave when an employee transfers between the two countries mid-year? Show me the passages, and tell me where the policies disagree or say nothing.

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

  • Leave policies for both countries
  • International mobility policy
  • Benefits policies
  • Group policy on transfers

The unit of work

One worker task per one policy document per reader.

Why a swarm fits

Each reader needs one policy, not the set. Two model families answer independently so a shared misreading is less likely, and the judge keeps a disagreement open instead of averaging it away.

Not for

A one-policy question with a plain answer. One model reading one document is cheaper and just as good.

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

    Does this policy's scope and effective date mean it applies to a transfer between these two countries at the time asked about?

    Sees only: One policy's scope, definitions and effective date, and the question

    Why: Keeps superseded or out-of-scope policies from colouring the answer.

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

    Evidence checkYes or no, with a probability

    After a worker answers

    Does the quoted policy passage state the leave entitlement or transfer condition that the reader attributes to it, without the reader adding assumptions?

    Sees only: One sub-question, one reader's statement and the policy passage quoted for it

    Why: An adviser will repeat this to an employee, so nothing inferred may pass as quoted.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry
  3. Reconciler, while merging

    Conflict checkA choice among options

    While reconciling

    Do the two independent readings of this policy give different answers to the same sub-question?

    Sees only: Two readers' statements and quotes for one sub-question on one policy

    Why: Disagreement between model families is the cheapest signal that a passage is ambiguous.

    • Same answerthenAccept
    • Different answersthenEscalate to a strong model
    • One found nothingthenReject and retry
  4. Run control, between rounds

    Retry or stopYes or no, with a probability

    After a rejection or low confidence

    Does the policy text itself settle the disagreement between the two readings, without needing to know what the authors intended?

    Sees only: Both readings and the full section they quote from

    Why: A strong model is only worth paying for when the text can decide; otherwise the conflict stays open.

    • Yes: 0.85 or higherthenEscalate to a strong model
    • Unsure: 0.50 up to 0.85thenMark unresolved
    • No: below 0.50thenMark unresolved
  5. Reconciler, while merging

    Conflict checkYes or no, with a probability

    While reconciling

    Do the home and host country policies state entitlements for the transfer year that cannot both apply?

    Sees only: The accepted passages from the two countries' policies on one sub-question

    Why: The conflict between policies is often the real answer the adviser needs.

    • Yes: 0.70 or higherthenMark unresolved
    • Unsure: 0.30 up to 0.70thenMark unresolved
    • No: below 0.30thenAccept
  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does this answer contain an open conflict, a silence or a statement about entitlement that an adviser might pass on to an employee?

    Sees only: The drafted answer with its quotes, conflicts and silences

    Why: The HR adviser owns what an employee is told; the run never speaks to the employee's case.

    Accountable: The HR adviser owns the answer given to the employee and takes open conflicts to the policy owner.

    • Yes: 0.20 or higherthenAsk a person
    • Unsure: 0.05 up to 0.20thenAsk a person
    • No: below 0.05thenAccept

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 splits the question into what each policy must be asked.

    Decisions here:1. Scope check

  2. Workers

    Two sets of small fast workers from different families read each policy independently and quote what it says.

    Designed for 3 to 24 agents, one worker task per one policy document per reader. Each worker receives only its own unit.

  3. Judge, from a different model family

    A decision model from a third family settles disagreements only where the text is clear, and keeps the rest open.

    Decisions here:2. Evidence check

  4. Reconciler

    A strong reasoning model writes one answer with the passages, the conflicts and the silences.

    Decisions here:3. Conflict check4. Retry or stop5. Conflict check

  5. Accountable person

    The HR adviser owns the answer given to the employee and takes open conflicts to the policy owner.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Every statement in the answer links to a quoted policy passage
  • The two families' readings are compared before either is accepted
  • Where a policy is silent, the answer says silent rather than inferring
  • Policy version and effective date are shown beside each quote

What comes back

  • A plain answer with quoted passages
  • Where the policies conflict, both passages side by side
  • Points no policy covers
  • Which policy owner to ask about each open point

What to measure

  • Share of answers the policy owner accepts unchanged
  • Adviser time per policy question
  • Conflicts surfaced that were previously unknown
  • Cost per question compared with one strong model alone

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

Check country handbooks against group people policy

For: Head of people policy or HR compliance lead in a company with staff in many countries

Each country has its own handbook and local policies, written at different times.

Pattern: Map, verify, reduceNeeds scale6 decisionsDesigned for 40 to 600 agents

Check job descriptions against the levelling framework

For: Reward or organisation design lead maintaining a job architecture across functions

Hundreds of job descriptions drift from the levelling framework.

Pattern: Hierarchical decompositionNeeds scale6 decisionsDesigned for 30 to 500 agents

Watch employment law changes against local policies

For: HR compliance lead or employment counsel covering many jurisdictions

Statutory leave, notice, working time and whistleblowing rules change country by country.

Pattern: WatchtowerNeeds a connector5 decisionsDesigned for 20 to 400 agents