> ## Documentation Index > Fetch the complete documentation index at: https://suprmind.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Consensus Score: How Suprmind Measures Agreement and Confidence # Consensus Score: How Suprmind Measures Agreement and Confidence > **Quick Answer:** Suprmind does not reduce agreement to a single number. Three signals do the work, and each measures something different: the **DCI turn score** (0 to 100) shows how much the AIs diverged on a turn, the **Decision Brief confidence level** (HIGH, MEDIUM, LOW) shows how much disagreement is still unresolved, and the **agreement indicators on Scribe notes** show which AIs backed each captured point. All three live in the right sidebar. ## What You'll Learn * What "consensus" means in Suprmind and how it is measured * How to read the DCI turn score and its colours * What the Decision Brief confidence level means, and why LOW is not a failure * What the agreement indicators on Scribe notes tell you * Which signal to use for which question *** ## What "consensus" means in Suprmind When two or more AIs answer the same turn, Suprmind compares their answers against each other. It looks for three things: * **Contradictions** - one AI says something another AI disputes * **Corrections** - one AI fixes a factual claim another AI made * **Unique insights** - a point only one AI raised Consensus is what remains after that comparison: the claims no AI contradicted or corrected. The comparison is done by an AI reading the answers, not by counting votes, so treat every signal on this page as a strong indication rather than a precise measurement. It is very good at pointing you to the places where the answer is not settled. It is not a grade for how good the answer is. *** ## The DCI turn score (0 to 100) The DCI turn score is the number badge on each turn in the **Adjudicator** tab of the right sidebar. Each turn is listed as "Turn 14, 9 items" with a score beside it, and the top of the tab shows the running totals for the session: how many contradictions, corrections, and unique insights the DCI (Disagreement/Correction Index) has found so far. **What the number means:** how much the AIs genuinely diverged on that turn, and how much that divergence was worth. The score adds up every contradiction, correction, and unique insight found in the turn, weighted by how serious each one is. Corrections count the most, contradictions next, unique insights least. The total is capped at 100. **The colours:** | Score | Colour | What it usually means | | - | - | - | | 50 to 100 | Purple | The AIs disagreed or corrected each other on points that matter. Read this turn closely. | | 20 to 49 | Amber | Some divergence, mostly moderate. Worth a scan. | | 0 to 19 | Grey | The AIs largely agreed, or the question was straightforward. | **A high score is not a good answer, and a low score is not a bad one.** A high score means the AIs caught each other on things. That is the turn where your own judgment is needed most. A low score means they landed in the same place, which is useful to know too. The score only exists when two or more AIs answered the turn. A turn answered by one AI has nothing to compare and gets no score. You will also see the same counts on the small DCI card that appears under an AI response in the chat, for example "2 contradictions, 1 correction, 3 insights". Click it to open the details in the Adjudicator tab. *** ## The Decision Brief confidence level (HIGH, MEDIUM, LOW) The confidence level is the chip next to the words **Decision Brief** in the Adjudicator tab, after you click **Generate Decision Brief**. It is the closest thing Suprmind has to a single consensus score for a whole conversation. **What it measures:** how much unresolved disagreement is left after the Adjudicator has read the DCI ledger and the Scribe's notes. It is a measure of deviation between the AIs, not a quality grade for the recommendation. | Level | What it means | | - | - | | **HIGH** | The AIs broadly agree, and no serious disagreement is still open. | | **MEDIUM** | The main direction holds, but a significant disagreement is still unresolved or a factual correction had to be made along the way. | | **LOW** | Several serious disagreements are still open, or the recommended direction goes against where most of the AIs landed. | **LOW is not a failure.** It means the question is genuinely hard and the AIs did not converge. The brief tells you exactly where they split and what would settle each dispute, which is often the most valuable output of the whole conversation. If the conversation has produced only one or two DCI items, the brief says so and leans toward next steps that would give you more signal rather than claiming a confidence it cannot justify. **What the brief contains:** * **Recommended direction** - one headline recommendation and the reasoning behind it * **Why this direction** - the council consensus, the baseline points from the Scribe that support it, and the DCI items that shaped or changed it * **Unresolved disagreements** - each open dispute with a decision rule for how you can settle it * **Uncontested risks** - risks a single AI raised that nobody argued against * **Correction ledger** - factual corrections made during the conversation, with severity and a suggested action * **Next action** - one concrete, verb-first next step **Copy** puts the brief on your clipboard. **Export** downloads it as a document with an audit trail showing exactly which Scribe notes and DCI items fed into it. **Regenerate** after a few more turns and compare the confidence level. **Availability:** generating a Decision Brief is available on Spark Enhanced, during Pro Preview on a trial, and on Pro, Frontier, Power, and Enterprise. *** ## The agreement indicators on Scribe notes Each note the Scribe captures in the **Scribe** tab can carry a coloured edge and a short line underneath naming which AIs agreed and which dissented, for example "GPT and Claude agree, Grok dissents on timeline". | Edge colour | What it means | | - | - | | Green | The AIs that answered broadly agree on this point | | Amber | The AIs are split on this point | | Red | This point was directly contradicted | **Why some notes have no indicator:** the Scribe only marks agreement on points where more than one AI expressed a view. A note captured from a turn answered by a single AI has nothing to compare. Notes that are not opinions, such as action items or context about your question, are never marked. **Turning the indicators on or off:** go to **Settings → Profile** and use the **Agreement indicators** toggle. It is on by default. Turning it off hides the coloured edges and the agreement lines. It does not change what the Scribe captures. These indicators are per note. They are handy while you read, but the DCI turn score and the Decision Brief are the measures built for making a decision. *** ## Which signal should I use? | You want to know | Look at | | - | - | | Which turn is worth re-reading because the AIs clashed | DCI turn score in the Adjudicator tab | | Whether the conversation as a whole reached a settled answer | Decision Brief confidence level | | Which specific points are still in dispute and how to settle them | Unresolved disagreements in the Decision Brief | | Whether one captured insight was widely backed or contested | Agreement indicator on the Scribe note | *** ## Tips * Generate a Decision Brief once, ask follow-up questions on the unresolved disagreements it lists, then Regenerate. Watching the confidence move from MEDIUM to HIGH is a clear sign the open questions were real and are now settled. * A run of purple turn scores in Debate mode is normal. Debate is designed to make the AIs clash. Look at the Correction Ledger in the brief to see which arguments actually survived. * If every turn scores grey, ask a harder question or ask the AIs to challenge each other. Consensus that arrives too easily is worth testing. * Use a red-edged Scribe note as a follow-up prompt: "You disagreed on the timeline. Which assumption drives the difference?" ## Related Articles * [The Adjudicator: Get a Structured Decision Brief](/docs/decision-support/adjudicator.md) * [Disagreement Index: Where Do the AIs Disagree?](/docs/decision-support/disagreement-index.md) * [Live Synthesis](/docs/scribe-panel/live-synthesis.md) ## Still Need Help? Reach out to us at [support@suprmind.ai](mailto:support@suprmind.ai) or use the feedback button in the app. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.