A contested claim settled against the supplied sources

For: Analyst, policy adviser or editor who must state a fact and defend it

Pattern: Cross-examinationRuns todayDesigned for 3 to 30 agents

The pain today

Sources disagree, or seem to. One model reads them and returns a confident blend. The person signing the memo needs to know what each source actually says and whether the disagreement is real.

The ask

I attached the reports and articles I am relying on, and my question is in the notes file. Have two independent readers answer it from these sources only, show me where they disagree, and tell me which statements the sources really support.

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

  • Reports, articles and dataset descriptions, as text
  • The question and the statements to test
  • Notes on which sources are primary

The unit of work

One worker task per one statement against the supplied sources.

Why a swarm fits

The question breaks into a few testable statements. Two families answering independently show whether a disagreement sits in the sources or in the reader; a third settles only what the text settles.

Not for

A settled fact with one authoritative source: look it up.

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. Planner, while planning

    Split or keep serialYes or no, with a probability

    While planning

    Does the question contain more than one separately testable statement, such as a quantity, a trend and a cause?

    Sees only: The question as written by the author

    Why: A blended question gets a blended answer; separate statements can each be checked.

    • Yes: 0.60 or higherthenSplit the work
    • Unsure: 0.30 up to 0.60thenKeep serial
    • No: below 0.30thenKeep serial
  2. After workers, the judge checks

    Evidence checkA choice among options

    After a worker answers

    What does the quoted source passage do to this statement?

    Sees only: One statement and one quoted passage with its document name

    Why: Silence is not support, and a contradiction is kept in view rather than outvoted.

    • Supports itthenAccept
    • Contradicts itthenMark unresolved
    • Is silent on itthenSkip this unit
    • Addresses a different period or placethenMark unresolved
  3. Reconciler, while merging

    Conflict checkA choice among options

    While reconciling

    Do the two readers' verdicts on this statement differ, and if so why?

    Sees only: Both readers' verdicts and quotes for one statement

    Why: Shows whether a disagreement sits in the sources or in the reader.

    • Same verdictthenAccept
    • They read the same passage differentlythenEscalate to a strong model
    • They relied on different sourcesthenMark unresolved
  4. Run control, between rounds

    Retry or stopA choice among options

    After a rejection or low confidence

    Do the two conflicting sources use the same definition, period and geography for the figure?

    Sees only: The two conflicting passages with their stated definitions

    Why: Many source conflicts are two definitions under one name, and the memo should say so.

    • Same basis, so the conflict is realthenMark unresolved
    • Different basis, so the conflict is only apparentthenAccept
    • A source does not state its basisthenMark unresolved
  5. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Will this statement be presented as fact in the memo while a source contradicts it or the readers' disagreement stayed open?

    Sees only: One statement's reconciled status

    Why: The author signs the memo and decides how much weight each source deserves.

    Accountable: The author decides what goes into the memo and how much weight each source deserves.

    • 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 breaks the question into statements and points each at the relevant source sections.

    Decisions here:1. Split or keep serial

  2. Workers

    Two answerers from different families test each statement independently, quoting the sources.

    Designed for 3 to 30 agents, one worker task per one statement against the supplied sources. Each worker receives only its own unit.

  3. Judge, from a different model family

    A decision model from a third family settles disagreements from the quotes, or leaves them open.

    Decisions here:2. Evidence check

  4. Reconciler

    A strong reasoning model writes supported, contradicted and unsettled statements, with definitional differences noted.

    Decisions here:3. Conflict check4. Retry or stop

  5. Accountable person

    The author decides what goes into the memo and how much weight each source deserves.

    Decisions here:5. Person decides

Checked before anything is accepted

  • Every verdict quotes the source passage and names the document
  • Apparent conflicts are checked for different definitions, dates or scope
  • Statements the sources do not address are marked not covered

What comes back

  • Each statement: supported, contradicted or unsettled, with quotes
  • Real conflicts between sources
  • Apparent conflicts explained by definitions or dates
  • What the supplied sources cannot answer

What to measure

  • Verdicts upheld by the author's own check
  • Disagreement rate between the two answerers
  • Time to a defensible statement

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

Literature review where every claim cites a passage

For: Research lead, analyst or doctoral researcher scoping a field before committing to a direction

A review means a very long reading list.

Pattern: Map, verify, reduceNeeds scale5 decisionsDesigned for 30 to 800 agents

Competitor teardown from public sources with citations

For: Strategy analyst, product leader or founder preparing a market entry or board discussion

A teardown is stitched from a few web pages and hearsay.

Pattern: Hierarchical decompositionNeeds a connector6 decisionsDesigned for 20 to 500 agents