# Research Quality Score: how Praxiom AI scores a synthesis

The Research Quality Score (RQS) is a 0 to 100 score Praxiom AI gives every synthesis. It weights five components: citation coverage (30%), cross-source triangulation (25%), severity distribution (20%), insight density (15%) and theme diversity (10%). A score below 60 makes the agent re-run the synthesis, up to two more times.

Updated14 November 2026

## What does the Research Quality Score measure?

RQS measures how well a set of insights is grounded in the research behind it: whether each claim has evidence, whether evidence comes from more than one source, and whether the synthesis covers the material instead of repeating one theme. It is computed deterministically from the saved insights and the number of sources analyzed, with no AI model involved, so the same input always gets the same score.

The five RQS components and their weights

| Component | Weight | How it is calculated (each scored 0 to 100) |
| --- | --- | --- |
| Citation coverage | 30% | Share of insights that include at least one direct supporting quote |
| Cross-source triangulation | 25% | Share of insights that cite two or more distinct sources |
| Severity distribution | 20% | How evenly severity ratings spread across critical, high, medium and low (Shannon entropy over the four levels, normalized). 0 when every insight has the same severity |
| Insight density | 15% | Insights per source, capped at 5. Five or more insights per source scores 100 |
| Theme diversity | 10% | How many of the five insight types appear: pain point, opportunity, user behavior, feature request and feedback theme |

The total is the weighted sum of the five component scores, rounded to one decimal place. A synthesis with no insights, or no sources, scores 0.

## What do the RQS grades mean?

RQS grade bands

| Score | Grade | Label | What Praxiom AI does |
| --- | --- | --- | --- |
| 80 to 100 | A | Excellent | Accepts the result and shows the score |
| 60 to 79.9 | B | Good | Accepts the result and points to the weakest component |
| 40 to 59.9 | C | Fair | Lists every component under 60 and re-runs the synthesis |
| 0 to 39.9 | D | Weak | Lists every component under 60 and re-runs the synthesis |

## What happens when a synthesis scores below 60?

- Before scoring, Praxiom AI checks every insight for at least one supporting quote and a valid severity (critical, high, medium or low). Problems are sent back to the agent to fix.
- Praxiom AI computes the RQS and gives the agent the score with guidance for each component under 60, for example "triangulate insights across multiple sources".
- If the score is below 60, the agent is told to re-analyze the sources and run the synthesis again, focusing on the weak components, before it saves.
- This repeats up to **two times**, so a synthesis gets at most three attempts.
- If the score is still below 60 after the second retry, Praxiom AI keeps the result and records that it was accepted below the threshold, along with the first attempt's score.

The score you see is computed from the final saved set of insights, not from an earlier attempt.

## How is an RQS calculated in practice?

A synthesis of 10 interviews saves 20 insights. 18 have supporting quotes, 12 cite two or more interviews, severities are 2 critical, 6 high, 8 medium and 4 low, and four of the five insight types appear.

Worked RQS example: 10 sources, 20 insights

| Component | Score | Weight | Contribution |
| --- | --- | --- | --- |
| Citation coverage | 90.0 (18 of 20) | 30% | 27.0 |
| Cross-source triangulation | 60.0 (12 of 20) | 25% | 15.0 |
| Severity distribution | 92.3 | 20% | 18.5 |
| Insight density | 40.0 (2 per source, cap 5) | 15% | 6.0 |
| Theme diversity | 80.0 (4 of 5 types) | 10% | 8.0 |
| **Total** | | | **74.5, grade B (Good)** |

Here the weakest component is insight density, so Praxiom AI would suggest extracting more distinct insights per interview. The result is above 60, so no retry runs.

## Where do you see the score?

- On the result of each synthesis, as a badge with the grade, score and label. Click it to see the five component scores.
- On the dashboard's **Research Quality** card, which charts your recent scores, their average and whether quality is trending up or down.

Recommendations and documents have their own scores on the same badge: RecQS (actionability, specificity, priority alignment, feasibility, evidence basis) and DocQS (structure and clarity, completeness, consistency, audience fit, reference quality).

## What does RQS not measure?

RQS scores the structure of the evidence, not whether a conclusion is right. A well-cited insight can still be a poor reading of what customers meant. That is why every insight keeps its quotes and links back to the source text: the score tells you where to look hardest, and the quotes let you check.

## Frequently asked questions

### What is a good Research Quality Score?

60 or above. Scores of 80 or more are graded A (Excellent), 60 to 79.9 B (Good), 40 to 59.9 C (Fair) and below 40 D (Weak). Praxiom AI re-runs a synthesis that scores below 60.

### How do I raise my RQS?

Analyze several sources on the same topic so insights can cite more than one, and use full transcripts rather than summaries so the agent can quote customers directly. Citation coverage and cross-source triangulation together make up 55% of the score.

### Does an AI model grade the synthesis?

No. RQS is a fixed formula over the saved insights and the number of sources. The same insights always produce the same score.

### Do re-runs below the threshold cost extra credits?

Retries happen inside the same synthesis run, and a run is charged by the compute it uses, so a run that needs retries uses more credits than one that passes first time. See [credits and billing](https://www.praxiomai.xyz/docs/credits-and-billing).

### What is the highest score a single-source synthesis can get?

75. With one source no insight can cite two sources, so cross-source triangulation, worth 25% of the score, is always 0. Every other component can still reach 100.

Run your product discovery on autopilot.

Bring your interviews, tickets and surveys. Get scored insights, a PRD and tickets.

[Sign up](https://app.praxiomai.xyz/sign-up)[See pricing](https://www.praxiomai.xyz/pricing)

Source: https://www.praxiomai.xyz/docs/research-quality-score
