> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getclaro.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Analyse

> Operations that read the catalogue and produce assessments, suggestions, or transformations.

Analyse is where you understand the state of a catalogue and trigger the transformations that improve it. The module exposes two surfaces: **Analyse** (data-quality reporting) and **Generative Engine** (model-driven content operations).

***

## Analyse

Runs a data-quality assessment on a catalogue or a filtered record set. Output includes:

* **Headline metrics** — total records, unique entries, duplicate estimates, columns analysed.
* **Strengths and attention areas** — narrative summary of what's good and what's weak.
* **Prioritized recommendation** — the single highest-leverage next operation to run.
* **Per-field scores** — fill rate, uniqueness, length variance, with an `excellent / good / fair / poor` rating.
* **Distribution stats** — mean, median, standard deviation, min, max — for numeric columns.
* **Critical issues** — issues with one-click follow-on actions:
  * *Trigger Enrichment* on a low-fill column
  * *Review and Validate* on suspicious distributions
  * *Find Duplicates* on suspected entity collisions
  * *Generate Taxonomy* when category coverage is poor
* **Overall Data Quality Score** — out of 100, computed from per-field scores weighted by the importance configured in the catalogue's Config tab.

Analyse is non-destructive — it reads records and writes a report, never modifies data.

### When to run

* Before kicking off a new pipeline, to set a baseline.
* After a large batch of new records, to spot regressions.
* On a schedule (e.g. weekly), to track quality over time.

Reports are versioned; you can diff two reports to see what improved or regressed.

***

## Generative Engine

Model-driven generation operations that produce content from your records.

| Capability                          | Output                                                                                      |
| ----------------------------------- | ------------------------------------------------------------------------------------------- |
| **Descriptions**                    | Long-form product descriptions, optimized for SEO and brand voice.                          |
| **Marketing copy**                  | Short copy variants — headlines, taglines, social posts.                                    |
| **Structured attribute extraction** | Pull typed fields out of free-text descriptions or supplier blurbs.                         |
| **Image alt-text**                  | Accessibility-compliant alt text from product images.                                       |
| **Translations**                    | Multi-language variants of any text attribute, with brand-voice consistency across locales. |

### Configuration

For every Generative Engine job you choose:

* **Target attribute(s)** — where the output is written.
* **Prompt** — the instruction, with `@attribute` references to other fields on the same record.
* **Knowledge Bases** — domain documents (brand guidelines, datasheets) the model can cite.
* **Model** — Claro default (with confidence scoring and citations) or another available model.
* **Confidence thresholds** — auto-apply, review, reject.
* **Length and format constraints** — character limits, schema, allowed tones.

Outputs ship with confidence scores and citations to the sources used, so each generated value is auditable. Below-threshold outputs queue for review in Notifications.

### Tips

* Reference other attributes by name with `@` to ground generation in the record. *"Write a 100-word description for `@product_name` based on `@spec_summary`, in the voice of `@brand`."*
* Include a Knowledge Base of brand and category guidelines for consistency across thousands of generations.
* Generate in batches and review confidence distributions before raising the auto-apply threshold.