CONTEXT(CLONE)
Sign in

Literature review where every claim cites a passage

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

Pattern: Map, verify, reduceNeeds scaleDesigned for 30 to 800 agents
Get early accessSign in to use

The pain today

A review means a very long reading list. Summaries written from abstracts overstate findings, and a chatbot asked for a review invents citations. Checking each claim against the paper is the part that gets skipped.

The ask

“I attached the full text of the papers from my search and my review questions. For each question, tell me what the papers actually found, with the passage quoted for every claim, note the sample and method, and show where studies contradict each other.”

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

  • Full text of the papers
  • Review questions and inclusion criteria
  • Extraction form, if one exists

The unit of work

One worker task per one paper against the review questions.

Why a swarm fits

Each paper is read alone against the same short list of questions. Synthesis then works on small quoted extracts, so no model has to hold the literature.

Not for

Understanding one key paper deeply: read it, or use one strong model with the paper open.

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

    Scope checkA choice among options

    Before work starts on a unit

    Does this paper meet the inclusion criteria on population, study design and outcome?

    Sees only: The paper's abstract and methods summary with the inclusion criteria

    Why: Screening first keeps reading costs on the papers that belong, and unread papers are listed.

    • Meets all criteriathenAccept
    • Fails a criterionthenSkip this unit
    • Cannot be judged from abstract and methodsthenEscalate to a strong model
    • Text could not be parsedthenMark unresolved

  2. Before workers, before a task runs

    Small worker or strong modelYes or no, with a probability

    Before a task runs

    Is this a single-study paper with a conventional methods and results layout, rather than a review, a multi-study paper or one dominated by tables?

    Sees only: The paper's section headings and length

    Why: Unusual structures are where small workers quote the wrong section.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenEscalate to a strong model
    • No: below 0.30thenEscalate to a strong model

  3. After workers, the judge checks

    Evidence checkA choice among options

    After a worker answers

    Does the quoted passage support the finding at the strength the worker worded it?

    Sees only: The worker's claim, the quoted passage and the name of the section it came from

    Why: Overstatement copied from abstracts is the usual way a review goes wrong.

    • Supports it at the stated strengththenAccept
    • Supports only a weaker or narrower findingthenReject and retry
    • Abstract only, and the results differthenReject and retry
    • Ambiguous about direction or subgroupthenEscalate to a strong model
    • Does not address itthenReject and retry

  4. Reconciler, while merging

    Conflict checkA choice among options

    While reconciling

    Do these two studies disagree about the same outcome in comparable populations?

    Sees only: Two verified extracts with their methods and samples

    Why: Real contradictions are set side by side and never averaged.

    • Opposite findings in comparable studiesthenMark unresolved
    • They differ because population or measure differsthenAccept
    • Same findingthenAccept

  5. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does this synthesis statement weigh studies against each other, or conclude more than any single quoted passage says?

    Sees only: One synthesis statement with the extracts under it

    Why: Weighing study quality and drawing conclusions is the researcher's work.

    Accountable: The researcher owns inclusion criteria, the weighing of study quality and every conclusion drawn.

    • 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 review questions into an extraction form and screens papers against the criteria.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from an open-weight family each read one paper and fill the form with quoted passages.

    Designed for 30 to 800 agents, one worker task per one paper against the review questions. 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 each quoted passage supports the claim as worded, no stronger.

    Decisions here:3. Evidence check

  4. Reconciler

    A strong reasoning model synthesises per question and sets contradicting studies side by side with their methods.

    Decisions here:4. Conflict check

  5. Accountable person

    The researcher owns inclusion criteria, the weighing of study quality and every conclusion drawn.

    Decisions here:5. Person decides

Checked before anything is accepted

  • Every claim quotes the passage and names paper and section
  • Claims taken from an abstract are checked against the results section
  • Contradicting studies are reported side by side, never averaged
  • Papers that could not be parsed are listed as unread

What comes back

  • Answer per review question with quoted passages
  • Evidence table: paper, method, sample, finding
  • Contradictions between studies
  • Questions the literature does not answer
  • Excluded and unread papers with the reason

What to measure

  • Quoted passages that support their claim, from a sampled check
  • Relevant findings a human reviewer found that the swarm missed
  • Hours to a first defensible draft
  • Cost per paper read

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 Research and intelligence

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