> ## Documentation Index
> Fetch the complete documentation index at: https://docs.plomo.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# How Plomo Processes, Classifies, and Scores Deal Documents
> Learn how Plomo parses, classifies, and scores every document in your data room — from raw upload to a reviewed, evidence-grounded taxonomy.
When you upload a file to a Plomo deal, it passes through a five-stage pipeline that turns raw documents into structured, searchable, evidence-backed classifications — without any manual effort on your part. Understanding this pipeline helps you interpret confidence scores, act on review flags, and trust the results you see in the workspace.
## The Processing Pipeline
```mermaid theme={null}
flowchart TD
Upload["Upload documents"] --> Parse["Parse & extract text"]
Parse --> Classify["Multimodal classification"]
Classify --> Recover["Recovery voting"]
Recover --> Ground["Evidence grounding"]
Ground --> Score["Confidence scoring & triage"]
```
***
## Stage 1 — Parse & Extract
As soon as a file arrives, Plomo extracts its text and structure. For documents that are image-based or scanned — such as photographed contracts or legacy PDFs — Plomo's OCR layer converts the visual content into machine-readable text before any AI analysis begins.
Plain-text extraction happens first so that the document becomes searchable and available to Copilot immediately. Classification runs as an enrichment step after indexing, which means large batches don't block your workflow while processing is in progress.
***
## Stage 2 — Multimodal Classification
With the extracted content in hand, Plomo's AI analyses the document against your deal's taxonomy and assigns it to the most appropriate category and subcategory.
Classification operates in two modes simultaneously:
* **Text mode** — the extracted text is matched against the semantic meaning of each taxonomy item, drawing on the language of your information request list.
* **Vision mode** — for visually structured documents such as financial statements, org charts, and presentations, Plomo's multimodal AI interprets layout, tables, and visual cues that pure text analysis could miss.
Combining both signals produces a more accurate result, particularly for documents that are ambiguous when read as plain text alone.
***
## Stage 3 — Recovery Voting
If the initial classification returns a response that is structurally invalid or flagged for low reliability, Plomo runs an automatic recovery pass. Multiple independent assessments are collected and the strongest, most consistent result is selected.
This voting step is reserved for genuinely ambiguous cases — it is not part of every document's journey. When it does run, it ensures that the classification you see reflects the most defensible interpretation of the document, rather than a single unreliable call.
***
## Stage 4 — Evidence Grounding
Every classification is anchored to a specific passage from the document. Plomo identifies the sentence or paragraph that most directly supports the category assignment and stores it as the **evidence excerpt** you see in the document view.
This grounding step does two things for you:
1. It lets you verify a classification in seconds — you can read the evidence and judge whether you agree, without opening the full document.
2. It prevents the AI from assigning a category based on vague or tangential content. A classification without locatable evidence is not accepted.
***
## Stage 5 — Confidence Scoring & Triage
Each classified document receives a confidence score between 0 and 1. Plomo uses this score to route the document into one of three review states:
| Confidence | Status | What it means |
| --------------- | --------------- | --------------------------------------------------------------------------------------------- |
| **≥ 0.50** | ✅ Accepted | Plomo is confident in the classification. No action needed unless you disagree. |
| **0.30 – 0.49** | ⚠️ Needs Review | The classification is plausible but uncertain. Review the evidence and confirm or correct it. |
| **\< 0.30** | ❓ Uncategorized | Plomo could not reliably classify the document. Assign it manually or investigate the file. |
Reviewing **Needs Review** and **Uncategorized** documents early improves the accuracy of Coverage tracking and Copilot responses across the deal.
***
## Supported File Formats
Plomo processes the following file types. Files in unsupported formats are stored but not classified.
| Format | Notes |
| -------------- | ---------------------------------------- |
| **PDF** | Includes scanned/image-only PDFs via OCR |
| **DOCX / DOC** | Microsoft Word documents |
| **XLSX / XLS** | Microsoft Excel spreadsheets |
| **PPTX** | Microsoft PowerPoint presentations |
| **CSV** | Comma-separated value files |
| **TXT** | Plain-text files |
***
## Real-Time Processing
Plomo processes documents in parallel — multiple files are classified at the same time rather than in a queue. As each document finishes, it appears immediately in the taxonomy tree and document list without waiting for the rest of the batch to complete.
This means you can start reviewing, querying with Copilot, and checking Coverage as soon as the first documents are ready, even while the remainder of a large upload is still processing. The status indicator on each document shows whether it is processing, needs review, or fully accepted.
Your documents are never used to train Plomo's AI or any underlying models. All document content is encrypted in transit and at rest, and is used solely to produce results within your own deal workspace.
***
## What's Next
Follow the step-by-step guide to create your first deal and upload documents in minutes.
Explore how the taxonomy and confidence scoring work, and how to correct classifications.
Learn how Plomo maps classified documents to your request list and surfaces gaps.
Ask plain-language questions about your deal and get evidence-grounded answers.