CONTEXT(CLONE)
Sign in

Subject lines and ad copy narrowed by independent judges

For: Lifecycle or performance marketer preparing a campaign under brand and claims rules

Pattern: TournamentNeeds live modelsDesigned for 8 to 150 agents
Get early accessSign in to use

The pain today

Teams test the two variants someone thought of on Monday. Producing many is easy; screening them for brand voice, banned claims and sameness is the work nobody has time for.

The ask

“I attached the campaign brief, our brand voice guide, the list of claims legal has approved and past campaign copy. Write a large set of subject lines and ad headlines, remove anything off-brand or making a claim not on the approved list, and give me a varied short list to test.”

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

  • Campaign brief and audience description
  • Brand voice guide
  • Approved-claims list from legal
  • Past campaign copy

The unit of work

One worker task per one candidate line.

Why a swarm fits

Candidates cost almost nothing and do not depend on each other. Judges need only the line, the brief and the rules, and different families disagree usefully about taste.

Not for

One email to a segment you know well: write it yourself.

The decision tree

5 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. After workers, the judge checks

    Evidence checkA choice among options

    After a worker answers

    Does this line make a factual or comparative claim, and is that claim on the approved list?

    Sees only: One candidate line and the approved-claims list

    Why: Screens claims before taste, so no judge time goes to a line legal would reject.

    • No factual claimthenAccept
    • Matches an approved wordingthenAccept
    • Stretches an approved wordingthenEscalate to a strong model
    • Claim is not on the listthenSkip this unit

  2. Evidence checkA score

    After a worker answers

    How well does this line match the traits in the brand voice guide and avoid its banned phrases?

    Sees only: One candidate line and the voice guide's traits and banned phrases

    Why: Narrows a large set on stated rules, with judges from another family than the authors.

    • High: 0.70 or higherthenContinue
    • Middle: 0.40 up to 0.70thenContinue
    • Low: below 0.40thenSkip this unit

  3. Reconciler, while merging

    Conflict checkYes or no, with a probability

    While reconciling

    Do these two lines use the same angle and the same hook word, so that testing both would teach nothing?

    Sees only: Two candidate lines with their angle labels

    Why: A short list of near-twins wastes the live test.

    • Yes: 0.80 or higherthenSkip this unit
    • Unsure: 0.45 up to 0.80thenContinue
    • No: below 0.45thenContinue

  4. Run control, between rounds

    Another round?Yes or no, with a probability

    Between rounds

    Does the short list still lack a surviving line for any angle in the brief?

    Sees only: The table of angles against surviving lines

    Why: Another round is only worth paying for when it can fill an empty angle.

    • Yes: 0.60 or higherthenContinue
    • Unsure: 0.30 up to 0.60thenStop
    • No: below 0.30thenStop

  5. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does this shortlisted line mention a price, a guarantee, a comparison with a competitor or a regulated benefit?

    Sees only: One shortlisted line

    Why: Legal owns claims and the marketer picks what goes live; judges only rank text.

    Accountable: The marketer picks what goes live and legal owns claims. Judges rank text; only a live test shows what works.

    • Yes: 0.30 or higherthenAsk a person
    • Unsure: 0.10 up to 0.30thenAsk a person
    • No: below 0.10thenAccept

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 turns the brief into angles to explore and the rules into a scoring rubric.

  2. Workers

    Small fast workers from mixed open-weight families each write candidates for one angle and audience.

    Designed for 8 to 150 agents, one worker task per one candidate line. Each worker receives only its own unit.

  3. Judge, from a different model family

    Judges from a different family than the authors screen claims first, then score brand fit and clarity in rounds.

    Decisions here:1. Evidence check2. Evidence check

  4. Reconciler

    A strong reasoning model collapses near-duplicates and keeps the short list spread across distinct angles.

    Decisions here:3. Conflict check4. Another round?

  5. Accountable person

    The marketer picks what goes live and legal owns claims. Judges rank text; only a live test shows what works.

    Decisions here:5. Person decides

Checked before anything is accepted

  • Any factual or comparative claim must match the approved-claims list or the line is removed
  • Judges never score lines from their own model family
  • Near-duplicates are collapsed so the short list spans distinct angles
  • No predicted open or click rates are reported

What comes back

  • Varied short list grouped by angle
  • Removed lines with the rule they broke
  • Angles where no candidate survived

What to measure

  • Live test result of the short list against the usual process
  • Lines rejected by legal after the tournament
  • Distinct angles in the short list
  • Cost per tested variant

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 Sales and marketing

A quarter of sales calls mapped to objections

For: Sales enablement lead or revenue operations manager explaining why deals are lost

The loss reason in the pipeline is a dropdown picked by the seller.

Pattern: Map, verify, reduceNeeds scale5 decisionsDesigned for 40 to 800 agents

Competitor pricing and product page watch

For: Product marketing manager or competitive intelligence lead keeping battlecards current

Battlecards are accurate the week they are written.

Pattern: WatchtowerNeeds a connector5 decisionsDesigned for 10 to 300 agents

Tender questionnaire answered from approved sources

For: Bid manager or solutions consultant responding to a large tender or security questionnaire

A tender arrives with a very long questionnaire and a short deadline.

Pattern: Hierarchical decompositionNeeds scale6 decisionsDesigned for 20 to 500 agents