> ## Documentation Index
> Fetch the complete documentation index at: https://help.polygon-one.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Evidence Documents

> Upload documents, understand the AI analysis, and accept values into components

## Overview

Test reports, technical documentation, and declarations of conformity are the evidence behind your PPWR data. In Polygon One you upload these documents directly on the packaging unit or component; your suppliers also deliver documents via the [supplier portal](/en/ppwr/datenerfassung/lieferantenportal). An AI reads extractable documents and proposes the values they contain for your components — while you stay in control.

## Uploading evidence

<Steps>
  <Step title="Start the upload">
    Open the packaging unit or component and click **Upload evidence**.

    <Frame caption="Upload dialog for evidence documents with document-type selection">
      <img src="https://mintlify.s3.us-west-1.amazonaws.com/polygonone/images/ppwr/datenerfassung/nachweis-upload-dialog.png" />
    </Frame>
  </Step>

  <Step title="Choose the document type">
    Select the **document type**: heavy metals, PFAS, recycled content, technical documentation, technical drawing, declaration of conformity, marking evidence, or other. For extractable types, the dialog shows the hint: "The AI will read this document and propose the values it contains."
  </Step>

  <Step title="Select the file">
    Supported are **PDF, JPG, and PNG up to 10 MB** (one file per upload).
  </Step>

  <Step title="Upload & analyze">
    Click **Upload & analyze** (or **Upload** when no extraction applies to the type). The analysis runs in the background; you can keep working.
  </Step>
</Steps>

<Info>
  Every document belongs to exactly one packaging unit **or** one component — so it is always unambiguous which data it substantiates.
</Info>

## What the AI does

The AI analysis has three building blocks. It is consistently **assistive**: the app points this out on every result — "AI assessment, not a compliance verdict."

### 1. Classification

The AI checks whether the chosen document type matches the content. If it recognizes a different type with high certainty, it corrects it and marks the document **Type corrected** ("You selected: {type}") — so a heavy-metals test report uploaded as "Other", for example, is still evaluated correctly.

### 2. Extraction

For extractable types (heavy metals, PFAS, recycled content, technical documentation, technical drawing), the AI reads out the regulated values: heavy metals (mg/kg), PFAS compliance, recycled content with source and method, materials, polymer, and mass. It also summarizes the document and shows which aspects (materials, heavy metals, PFAS, weight, recyclability) it covers and which are missing.

Declarations of conformity, marking evidence, and "other" documents only go through a light check and are marked **AI-checked** — no values are extracted from them.

### 3. Proposals for your components

From the extracted values, the AI generates **proposals**: new components or updates to existing components, each with the field, the current value, the proposed value, and a **confidence percentage**. Values the AI is not sure about carry the **Uncertain** marker ("check it against the document before accepting").

## Reviewing proposals: auto-applied vs. held for review

Polygon One distinguishes two paths:

* **Automatically applied (AI):** values the AI is confident about are written directly to the component. These writes are fully transparent: the **"Automatically applied by AI"** card lists every automatically written value with its date, and any applied value can be removed again via **Revert value** — the field then returns to "not provided".
* **Held for review:** uncertain values are held back. They appear in the review panel **"AI analysis: proposed data"**, where you decide each proposal via **Accept** or **Reject** — you can also reject individual fields of a proposal. Only your acceptance writes the value to the component. One particularly important safety net: if the AI reads exactly 100 mg/kg for heavy metals (the legal limit), the value is **always** held for manual review — the likely explanation is that it accidentally read the limit instead of the measurement.

<Frame caption="Review panel with AI proposals: current value, proposed value, confidence">
  <img src="https://mintlify.s3.us-west-1.amazonaws.com/polygonone/images/ppwr/datenerfassung/nachweis-vorschlaege-pruefen.png" />
</Frame>

<Info>
  Accepting a value only stores the **measured value** on the component's record — whether the duty is met is derived separately from that value by the [obligation checklist](/en/ppwr/rollen-pflichten/pflichtenkatalog). An accepted heavy-metals value of 120 mg/kg therefore correctly results in **Outstanding**, not "Met".
</Info>

Via the **Accept all AI suggestions ({count})** button in the documents view, you accept all open proposals across all documents and units in one pass; the confirmation dialog states how many of them are flagged as uncertain. Proposals missing required details stay open for review.

### Reanalyze

Via **Reanalyze** in the document viewer you re-run the analysis of an already-analyzed document — for instance after a model improvement. Open proposals are replaced; values already applied to components are kept (still undoable via **Revert value**).

## The document library

Under **Documents** (`/packaging/documents`) you find all packaging evidence in one place: "All packaging evidence — technical documentation, declarations of conformity, and substance attestations."

<Frame caption="Documents view: cards grouped by document type with analysis badges">
  <img src="https://mintlify.s3.us-west-1.amazonaws.com/polygonone/images/ppwr/datenerfassung/dokumente-bibliothek.png" />
</Frame>

* Documents are shown as **cards grouped by document type**; the search ("Search documents…") filters the list.
* Every card shows the file name, upload date, the linked unit or component, and analysis badges: **{n} value(s) extracted**, **{n} to review**, **{n} uncertain**, **No usable data**, or **AI-checked**.
* A click opens the **document viewer**: the PDF preview on the left; type, review status, linkage, the AI analysis with extracted values or open proposals, and download on the right.

### Approving or rejecting documents

Independently of the AI proposals, you can judge each document as a whole: **Approve** or **Reject** (with a reason). This matters especially for supplier-provided documents — a rejected document no longer satisfies any duty, and the corresponding [data request](/en/ppwr/datenerfassung/datenanfragen) automatically goes back to the supplier with a "Documents to replace" note. Until you decide, a document shows as **Pending review**.

## Retention

The PPWR requires conformity records to be kept for **5 years** — for reusable packaging **10 years** (likewise the traceability records under Art. 22). Your documents uploaded to Polygon One remain stored centrally and can be retrieved and exported at any time, so you can meet these periods towards market surveillance.

## Next steps

<CardGroup cols={2}>
  <Card title="Technical documentation" icon="file-lines" href="/en/ppwr/dokumentation/technische-dokumentation">
    How data and evidence become the Annex VII documentation.
  </Card>

  <Card title="Declaration of Conformity" icon="file-signature" href="/en/ppwr/dokumentation/konformitaetserklaerung">
    Generate the EU Declaration of Conformity or collect and verify it.
  </Card>
</CardGroup>