Six model families answer independently.
OpenAI, Claude, Gemini, Mistral, DeepSeek, and Grok receive the same prompt. Their original answers stay separate, so the synthesis never replaces the raw evidence.
Consensus & Differences Engine
Consensus and Difference Engines explained.
A clear look at how consens.io combines model answers, checks disagreements, and keeps the judging layer independent.
consens.io
Consensus
One question · 6 models What is the longest river in the world?
68
Strong agreement
1 minor detail · disputed: where the Amazon starts, 6 models compared
Consensus Answer
The Nile is the conventional longest river, but the Amazon has a credible claim. By convention the Nile leads at about 6,650 km Whether the Amazon overtakes it depends on where its mouth is drawn, and a disputed 2007 survey draws it further out than most references do.
Marked passages are checked key claims. Unmarked text was not compared individually.
See all differences
Is the Nile definitively the longest river?
ChatGPT, Mistral, DeepSeek
The Nile is the settled answer at about 6,650 km.
Claude, Gemini, Grok
The 2007 Amazon survey is a credible challenger.
How it works
The engine runs in four readable steps.
Technically, three roles are involved: the answer models produce independent raw responses, your selected Consensus Engine writes the synthesis, and a separate judge extracts agreement and conflict from the result.
POST /consensus
Question, included model answers, excluded models, sources, and selected engine
The backend authenticates the user, checks tier and usage rules, caps client-sent text, and requires at least two included model answers.
1
Consensus Engine
stream_consensus()
- Build synthesis prompt Drop empty or excluded answers, shuffle the remaining responses, anonymize them as Expert A/B/C, and attach compact source provenance.
- Resolve selected engine The dropdown value maps to a provider model, including standard, direct model IDs, or Pro aliases.
- Stream the synthesis The selected engine writes the user-facing answer. Provider failures are retried twice, then a fallback family is tried when a key is available.
- Gate the second pass If the final consensus is empty or an error text, the comparison is skipped so the judge never analyzes an error message.
2
Differences Engine
stream_differences()
- Build judge prompt Use the consensus answer plus included source answers, cap each answer for the judge pass, shuffle, and anonymize as Model A/B/C.
- Pick an independent judge The judge tier follows the selected Consensus Engine, but the family is preferably different to reduce self-judging bias.
- Request strict JSON The judge returns claims, agreement and dissent, difference cards, severity, verification hints, and the closest source model.
- Verify and score The server repairs JSON if safe, maps labels back, verifies quotes and anchors, computes the agreement score, and records judge metadata.
SSE final payload
consensus_response + differences_data + usage/share metadata
The UI renders one readable consensus from this payload, marks the anchored passages inline, and keeps the agreement score, the contradiction entries and the judge note attached to it.
The selected Consensus Engine writes the synthesis.
The engine you choose in the dropdown reads only the included answers, ignores excluded ones, and writes one response that preserves common ground plus necessary caveats.
A separate judge checks the consensus.
The Differences Judge extracts key claims, maps which models support or dissent, and preferably runs on a different model family than the synthesis engine.
The UI marks the evidence inside the answer.
The final payload contains the consensus text, agreement score, claim badges, contradiction entries with their anchors, and judge metadata, so each disputed passage can be marked exactly where it stands.
Model roles
Different models do different jobs.
The process is intentionally split. The model that writes the consensus is not treated as the only authority on whether the consensus is reliable.
Answer layer
OpenAI, Claude, Gemini, Mistral, DeepSeek, and Grok
These are the source answers. They are queried in parallel, shown to you directly, and then passed into the consensus step if they are not excluded.
Synthesis layer
The engine chosen in the Consensus picker
This model reads the included source answers and writes the combined response. Pick a preset (Daily, Balanced, High Quality) or choose a specific model via Custom. Standard and Pro engines can both be used.
Analysis layer
An independent Differences Judge
This judge extracts claims and contradictions from the synthesis. For independence, consens.io prefers a different model family from the Consensus Engine.
Differences Engine
The second pass turns disagreement into structured evidence.
After the consensus text is written, consens.io runs a separate analysis pass. Its job is not to rewrite the answer, but to explain how the source models support, qualify, or contradict it.
Inputs are filtered and anonymized.
Only non-empty, non-excluded model answers are used. Each answer is capped before analysis and then renamed to neutral labels like Model A, Model B, and Model C.
Why: anonymization and shuffled order reduce provider-name and position bias in the judge pass.
The judge must return strict JSON.
The prompt asks for central claims, agreeing models, dissenting quotes, difference cards, contradiction severity, verification hints, and the model closest to the consensus.
Types: contradiction for incompatible facts or conclusions, emphasis for different focus or missing context.
The judge is chosen independently.
The judge tier follows the selected Consensus Engine. Standard consensus uses a standard judge; Pro consensus tries a Pro judge from another model family.
Fallback: primary judge, retry, then the next available family; Pro attempts can fail open to a standard judge.
The server verifies and scores the result.
JSON is repaired if safely possible, model labels are mapped back, quotes are checked against the original answers, and unverifiable quotes are removed.
Output: agreement score, level, major/minor contradiction counts, emphasis counts, and judge metadata.
Structured payload
What comes out of the Differences Engine
claimsConsensus anchors, agreeing models, dissenting model quotes.
differencesContradiction or emphasis entries with positions and verification hints, each with a
consensus_anchor: the verbatim passage in the consensus answer that the disagreement is about.agreementA 0-100 score computed from claim support and weighted disagreement penalties.
judgesThe provider, model, and tier that actually delivered the analysis.
64
Partial agreement
1 critical · disputed: the condition attached to it, 6 models compared
Analysis by Gemini (Pro)
independent of the consensus engine
Differences
Does the caveat change the recommendation?
ChatGPT, Mistral, DeepSeek
The recommendation holds as stated.
Claude, Gemini, Grok
It only holds under a specific condition.
Worth verifying: check whether the condition applies to your case before acting on the answer.
What you see
A useful answer without pretending everything is settled.
There is one answer to read. Where the judge found a verified disagreement, the passage that carries it is marked in place, and the full analysis stays one click below.
82
Strong agreement
1 minor detail · disputed: how strong the evidence is, 6 models compared
Analysis by Mistral
independent of the consensus engine
Consensus Answer
Most models agree on the main recommendation.
The answer combines the repeated points and removes duplicated wording How strong the underlying evidence is turns out to be the open question, because the sources rate the measurement quality differently.
Marked passages are checked key claims. Unmarked text was not compared individually.
See all differences
How strong is the evidence?
ChatGPT, Gemini, DeepSeek
The evidence is settled and supports the recommendation.
Claude, Mistral, Grok
A caveat about measurement or source quality applies.
Analysis by a judge model, independent of the consensus engine where keys allow it.
Why this helps
Consensus is a reading aid, not a magic truth label.
Less blind trust
One confident answer is not enough
Seeing several model families makes weak spots easier to notice before you act on an answer.
More signal
Agreement becomes visible
Repeated claims are separated from one-off claims, so you can scan the strongest common ground first.
Clear caveats
Disagreement stays in view
Contradictions are not averaged away. They are shown as specific points to inspect or verify.
FAQ
Common questions.
Is the consensus just the majority answer?
No. The engine writes a synthesis from the included model responses. Agreement matters, but the output is written as a readable answer with caveats, not as a simple vote.
Can I choose the Consensus Engine?
Yes. In the app you pick a preset (Daily, Balanced, or High Quality) or select a specific consensus model via Custom. If the resulting engine is a Pro engine, the differences analysis tries to use a Pro judge from another family.
Why use a different judge family?
If the same model family writes and judges the answer, it may be less likely to challenge its own framing. consens.io therefore prefers an independent family when available.
What happens if the Differences JSON is malformed?
The backend extracts or repairs a complete JSON object when it can do so safely. If the output still cannot be parsed, raw JSON is not shown to the user.
Try the flow