CONTEXT(CLONE)
Sign in

Code consultation responses by question and keep minority views

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

Pattern: Map, verify, reduceNeeds scaleDesigned for 40 to 800 agents
Get early accessSign in to use

The pain today

A public consultation brings thousands of free-text responses, many from campaigns, some from experts with one decisive point. Summaries drift toward the loudest themes, and the team cannot show how each response was counted.

The ask

“I attached the consultation document with its questions and all the responses we received. For each response, tell me which questions it answers, what position it takes and what reasons it gives, with quotes. Then summarise by question, keep campaign responses separate from individual ones, and make sure minority and expert points stay visible.”

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

  • Consultation document with questions
  • Responses as text with respondent type
  • Coding frame if one exists

The unit of work

One worker task per one consultation response.

Why a swarm fits

Each response is coded alone against the question list. Responses are independent and numerous, and every coded point must stay traceable to its own words for the published analysis.

Not for

Deciding what the policy should be or weighing views against each other. A consultation with a few dozen responses should simply be read.

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 checkA choice among options

    Before work starts on a unit

    Does this submission contain a response to the consultation, rather than a blank form, an acknowledgement or a duplicate upload?

    Sees only: One submission's text

    Why: Keeps the count honest and workers off empty submissions.

    • Substantive responsethenAccept
    • Blank or administrativethenSkip this unit
    • Exact duplicate of an earlier submissionthenSkip this unit

  2. Before workers, before a task runs

    Small worker or strong modelA choice among options

    Before a task runs

    Does the response answer under the numbered questions, or is it a free-form letter that must be mapped to questions?

    Sees only: The first part of one response

    Why: Long expert letters get a stronger reader; form answers stay on the cheap floor.

    • Follows the question numbersthenAccept
    • Free-form letter or reportthenEscalate to a strong model

  3. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Do the respondent's quoted words state the position the worker coded for this question, rather than describing someone else's view?

    Sees only: One coded position, the question and the quoted words

    Why: No one is counted for or against on words they did not write.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry

  4. Reconciler, while merging

    Conflict checkA choice among options

    While reconciling

    Is this response's wording substantially the same as the campaign template text shown?

    Sees only: One response and one detected template text

    Why: Groups campaign responses while keeping any personal additions as separate points.

    • Same textthenAccept
    • Template text with added personal pointsthenMark unresolved
    • Different textthenContinue

  5. After workers, the judge checks

    Evidence checkA choice among options

    After a worker answers

    Does this point offer evidence, a legal argument or an affected group that none of the listed themes for the question covers?

    Sees only: One coded point and the theme list for its question

    Why: Protects minority and expert points from being dropped for low frequency.

    • Covered by an existing themethenAccept
    • New point or new evidencethenMark unresolved

  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does the item change the coding frame, concern a statutory consultee or raise a legal or equality point?

    Sees only: One flagged item with its quotes

    Why: The consultation lead owns the frame and the published analysis; policy owners decide what the responses mean.

    Accountable: The consultation lead owns the coding frame and the published analysis; the policy owner decides how responses shape the policy.

    • 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 drafts the coding frame from a sample, has it confirmed, and freezes it for the wide pass.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from an open-weight family code one response per question with position, reasons and quotes.

    Designed for 40 to 800 agents, one worker task per one consultation response. Each worker receives only its own unit.

    Decisions here:2. Small worker or strong model

  3. Judge, from a different model family

    A decision model from a different family checks that each code is supported by the quoted words of the respondent.

    Decisions here:3. Evidence check5. Evidence check

  4. Reconciler

    A strong reasoning model summarises per question, separates campaign text and lists rare but substantive points.

    Decisions here:4. Conflict check

  5. Accountable person

    The consultation lead owns the coding frame and the published analysis; the policy owner decides how responses shape the policy.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Every coded position quotes the response
  • Near-identical campaign responses are detected and grouped, not counted as independent views
  • Points raised by few respondents are listed, not dropped for low frequency

What comes back

  • Coded response table with quotes
  • Summary per question with positions and reasons
  • Campaign groups with their template text
  • Rare substantive points and new evidence offered

What to measure

  • Coding agreement with analysts on a sample
  • Substantive points analysts found that the swarm missed
  • Analyst time per response
  • Cost per response

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

More in Public sector and policy

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