For Experts

About Us

Insights

Join Our Team

Book a Demo

Book a Demo

For Experts

About Us

Insights

Join Our Team

Book a Demo

Ask Synquery Connector

Connector

API v0 · Pilot phase

← All API docs

Understanding results

Synthesis, agreement and confidence

How the combined answer is built, what agreement and confidence measure, and what they don't.

When a request completes, Synquery adds a synthesis: answer, a per-field summary, agreement, clusters, disagreement_summary and confidence.

How it is built

The synthesis is generated by a model from the expert responses on that request only. It uses no outside sources and adds no figures the experts didn't give. Numeric summaries are calculated from the experts' values, which may themselves be estimates.

Incompatible answers are kept apart

Interviewers, AI or human, check that experts answer the same question with the same scope, such as the same denominator, and ask when they don't. If a mismatch still reaches the synthesis, the answers are not blended: the summary separates them, warnings names the mismatch and disagreement_summary explains it.

Agreement

agreement counts how many experts agree on the first numeric or range field in your answer_fields, or on the first choice field if there is none. It is null when there is neither.

Confidence

Confidence measures agreement within this panel and how relevant the experts' experience is to the question. It does not measure factual accuracy. The rules apply in order, and the first that matches wins:

  1. low if fewer than half the included experts agree, or more than half the answers are estimates or reused;
  2. high if at least 80% agree and at least half the answers are first-hand and from the last 2 years;
  3. otherwise medium.

A request with no numeric or choice field gets its confidence from basis and recency alone.

Where disagreement comes from

disagreement_source is stated when experts gave the reason themselves, and inferred when the model inferred it from their answers.

Synthesis fields

null while the request is pending.

answerstring#
The synthesis in one or two sentences, stating what the panel found and what kind of evidence it is.
summaryobject#
One entry per answer field, keyed by field name: low / high for numeric and range fields, top for lists, counts for choices; each with n (answers included), received (answers given) and basis where numeric.
{"agreeing": n, "total": n, "on": field}. How many experts agree on the first numeric or range field in your answer_fields, or on the first choice field if there is none; null when there is neither.
Where views group on a field: field, label, count and the sides in it.
Where experts differ and why. null if all agree.
stated (experts gave the reason) or inferred (the model inferred it).
high, medium or low. See Synthesis, agreement and confidence.
basisstring#
Most common basis across responses: first_hand, estimate or aggregate.
Anything Synquery changed or couldn't do: an unmet preference, a denominator mismatch, an unanswered field.
JSON Schema for responses[].fields, built from your answer_fields; every property is nullable.
Sample responseGET /questions/{request_id}Fictional data
Question
What share of total knee replacements at US hospitals now use robotic assistance, and what is driving adoption?
answer
Four of five experts put robotic assistance at 20–30% of total knee replacements at the hospitals they know, higher at large systems and orthopedic-focused centers. These are practitioner estimates, not a measured national share.
agreement
4 of 5
on robotic_share
confidence
medium
Agreement within this panel, not factual accuracy
basis
First-hand
Most common basis across answers
experts
5 of 5
1 reused past answer
summary · one entry per answer field
robotic_share20–30%4 of 5 included · estimate
adoption_driversSurgeon training and recruitment · Patient demandtop answers · n 5
barriersCapital cost · OR time during learningtop answers · n 5
outlook_3yrincrease 5 · flat 0 · decrease 0counts · n 5
clusters · robotic_share
20–25%
2 · Surgeon, OR director
25–30%
2 · Surgeon, Seller
35–40%
1 · Seller
disagreement_summary
source: stated
The former device rep counts only hospitals that own a robot; the other four count all hospitals. His figure is reported separately and not blended into the range.
warnings: robotic_share: exp_2c7x1 used a different denominator (hospitals with a robot); excluded from the combined range.

Next

Evidence and transcripts →

Synthesis on a complete requestjson
{
  "answer": "Four of five experts put robotic assistance at 20–30% of total knee replacements at the hospitals they know, higher at large systems and orthopedic-focused centers. These are practitioner estimates, not a measured national share.",
  "summary": {
    "robotic_share": {"low": 20, "high": 30, "basis": "estimate", "n": 4, "received": 5},
    "outlook_3yr": {"counts": {"increase": 5, "flat": 0, "decrease": 0}, "n": 5}
  },
  "agreement": {"agreeing": 4, "total": 5, "on": "robotic_share"},
  "clusters": [
    {"field": "robotic_share", "label": "20–25%", "count": 2, "sides": ["surgeon", "or_director"]},
    {"field": "robotic_share", "label": "25–30%", "count": 2, "sides": ["surgeon", "seller"]},
    {"field": "robotic_share", "label": "35–40%", "count": 1, "sides": ["seller"]}
  ],
  "disagreement_summary": "The former device rep counts only hospitals that own a robot; the other four count all hospitals. His figure is reported separately and not blended into the range.",
  "disagreement_source": "stated",
  "confidence": "medium",
  "warnings": [
    "robotic_share: exp_2c7x1 used a different denominator (hospitals with a robot); excluded from the combined range."
  ]
}