> ## Documentation Index > Fetch the complete documentation index at: https://docs.coverbase.com/llms.txt > Use this file to discover all available pages before exploring further. # Generate a clause set from your own contract > Upload your template MSA or DPA and let Coverbase extract a clause set from it: what it pulls out, why it drops language it cannot quote, and what you must add before the set is usable.
For AI agents: a documentation index is available at [https://docs.coverbase.com/llms.txt](https://docs.coverbase.com/llms.txt). This page is also available in markdown by appending .md to the URL.
This page is part of the [User Guides](/user-guides/overview) collection and goes with the [Contract Guardian guide](/user-guides/contract-guardian), which covers clause sets end to end. Read this one when your standard is a document rather than a spreadsheet. Most companies already have their standard written down as a contract: the MSA you send first, the DPA your privacy team maintains, the security addendum you attach to anything touching customer data. **Upload Baseline Contract** reads one of those and turns it into a clause set, so your playbook starts from language you already use rather than a blank page. This is one of four ways to build a clause set. The others are copying the packaged library, [importing a spreadsheet](/user-guides/clause-set-spreadsheet-import), and adding clauses by hand. They mix freely: generate from your paper, then paste in the two clauses your template forgot. ## Which import do I want? The two upload paths solve different problems. **You have a contract.** Coverbase reads the PDF and proposes the clauses, with the exact source text quoted. Fastest way from "we have a template" to a working set, but every clause arrives with one tier, so you finish the tiering by hand. **You have a playbook.** Your standards already exist as structured rows, probably with fallback positions. You get exactly what you typed, including full multi-tier variant language, and nothing is inferred. A common sequence: generate from your paper to get the clause inventory, then hand-write the fallback tiers in the UI. If your team already maintains a negotiation matrix, skip extraction and import the spreadsheet. ## Run an extraction On **Configuration → Clause Sets**, open **Add Clause Set** and choose **Upload Baseline Contract**. Submit your template as a PDF. Use a **clean template**, not an executed agreement. A signed contract carries one counterparty's negotiated edits, and you would be baking their redlines into your standard. Extraction runs in the background; you can leave the page and come back. The clause set appears immediately with a processing indicator and fills in as the run completes. Each proposed clause arrives with a name, category, severity, applicable component types, and the **exact source text** it came from, with a page citation you can check against your document. This is the step people skip, and the set is not usable without it. See [What you have to finish](#what-you-have-to-finish) below. Generated clause sets are created **inactive**, so a half-finished playbook cannot be run against a live contract. Flip it to active once the tiers are in. ## What you get back Each extracted clause is a normal reference clause, so you can edit, retier, or archive any of it. | Field | How it is filled | | ------------------------------ | ------------------------------------------------------------------------------------------------------------------- | | **Identifier** | Auto-assigned as `BASE-001`, `BASE-002`, ... Rename them to your own scheme if you have one. | | **Name** | Proposed from the clause heading. | | **Category** | Proposed from the clause subject. | | **Severity** | Proposed from the clause subject. Check these: the model does not know which clauses would block *your* signature. | | **Applicable component types** | Proposed from the document type it read. | | **Source evidence** | The exact quoted text plus its page, kept on the clause so a reviewer can trace any standard back to your template. | | **Variants** | Exactly one: **Tier 1, labeled "Baseline"**, your template's own language. | **Language it cannot quote verbatim is dropped, not paraphrased.** Every extracted clause and every piece of source evidence is checked back against the text of the document it cites; anything that does not appear there contiguously is discarded. You may get fewer clauses than you expected, but what you get is your own paper, not a model's summary of it. ## What you have to finish **A single Tier 1 variant is a floor, not a playbook.** It marks anything that is not your exact template language as needing review, which on a real vendor's paper is almost everything. A set run in that state flags nearly every clause. Before you point a generated set at a live contract: 1. **Add the tiers you would accept.** For each clause that matters, write the Tier 2 ("acceptable") and Tier 3 ("fallback") language your legal team would sign without escalating. 2. **Fill in the Guidance field.** Extraction does not write the aliases and extraction signals that help the matcher find a clause under a different heading. The [Contract Guardian guide](/user-guides/contract-guardian) shows the four-section format. 3. **Retune severities.** Reserve Critical for clauses that would block a signature. 4. **Archive what you do not care about.** A template contains boilerplate you have no opinion on. Every clause you keep is a clause every review has to triage. 5. **Test against a contract you already know**, then activate. ## Troubleshooting | What you see | What it means | | ----------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Extraction returned far fewer clauses than the document has | Expected on documents where the text layer is poor. Language that cannot be quoted verbatim is dropped rather than guessed at. Check the PDF is text-based, not a scan of a printout. | | The clause set is stuck processing | Extraction runs in the background and larger documents take longer. If it stays pending well beyond a few minutes, the run failed. The clause set surfaces the failure, and you can archive it and retry. | | Clauses came back with your counterparty's edits | You uploaded an executed agreement instead of your template. Start again from clean paper. | | Everything is marked as a deviation on the first review | The set still has one tier per clause. Add fallback tiers before running it against a live contract. | | The clause set will not run against a contract | Generated sets start inactive. Activate it once the tiers are in. | ## Related Clause sets end to end: the reference clause anatomy, writing risk-tier variants, and running a review. The 12 packaged clause standards you can copy instead of, or alongside, extracting your own. Bring a playbook in from a workbook when your standards already live in structured rows.