Standards & Alignment

Overview

Standards and playbooks are reference documents that define how your contracts should look. Upload your preferred templates and annotate key sections, then use them to align and evaluate new documents. ContractRabbit generates clause-by-clause recommendations with importance scores, tracked changes, and category labels.

Standards vs. playbooks

• Standards — Reference documents representing your ideal contract language for a specific document type
• Playbooks — Guided instruction documents that describe how to review or negotiate specific contract types

Both are managed from the same interface and work the same way — the distinction is organizational.

Creating a standard or playbook

1. Navigate to Standards in the left sidebar under Data
2. Switch to the Standards or Playbooks tab
3. Drag and drop a PDF or Word document onto the upload area
4. ContractRabbit creates the standard and opens the detail page

Standards take an abbreviated processing path — they are clause-structured and versioned but skip party extraction and scoring, since they are reference documents, not documents to analyze.

Annotating sections

On the standard detail page, you'll see the full document in a viewer. Click on any section to create an annotation — a marker that tells ContractRabbit this section is important for alignment.

Annotations appear in a sidebar list, making it easy to see which parts of the document you've highlighted as key reference points.

How alignment works

When you align a document against a standard, ContractRabbit runs a sophisticated comparison pipeline:

Matching

Each section of the standard is matched to the best corresponding section in the target document using a hybrid score that combines body text similarity (60% weight) and section title similarity (40% weight). This BM25-based matching handles synonym drift better than pure embedding similarity.

Two-pass analysis

1. Substantive pass — All non-definition sections are analyzed concurrently, generating UPDATE, INSERT, or DELETE recommendations
2. Definition pass — Only definitions that are actually referenced by substantive recommendations are analyzed, avoiding unnecessary changes to unused definitions

Party substitution

ContractRabbit extracts party names from both the standard and the target document, builds a substitution table (e.g., "Vendor" → "Supplier"), and applies it to all suggested text — preventing "wrong party" errors in recommendations.

Recommendation types

Each recommendation includes:

• Importance score (0–1) calibrated by match strength, change type, and content overlap
• Category label from a 26-category taxonomy (Confidentiality, Indemnification, Termination, Data Protection, etc.)
• Suggested text with word-level diffs highlighting exactly what changed
• Reference sources showing which standard or corpus clauses influenced the recommendation

Quality guardrails

Recommendations pass through a 10-guard post-processing pipeline that catches systematic LLM errors:

• Discards DELETE recommendations without strong text match evidence
• Removes recommendations with suspiciously short suggested text
• Deduplicates repeated clauses within suggestions
• Strips leaked internal formatting from suggested text
• Replaces placeholder party names with actual names
• Penalizes recommendations where suggested text has low overlap with the original (likely about a different subject)
• Consolidates multiple child-level changes into a single parent-level update when appropriate

Score calibration

Raw LLM importance scores are adjusted based on evidence quality:

• No BM25 match → 30% penalty
• Weak BM25 match → 15% penalty
• INSERT type → additional 10% penalty (more uncertain than UPDATE)
• DELETE type → additional 25% penalty (requires strongest evidence)

Reviewing recommendations

After alignment completes, recommendations appear in the Review Panel alongside the document viewer. Each recommendation shows:

• Inline diff with additions in blue and deletions in red strikethrough
• Importance score as a color-coded badge (red for high, orange for medium, gray for low)
• Category tag (e.g., "Indemnification", "Data Protection")
• Source badges showing origin: Standard (indigo), Corpus (green), Playbook (purple), or User Edit (blue)

Actions

For each recommendation you can:

• Accept — Creates a tracked change in the document that can be exported as a redline
• Reject — Dismisses the recommendation, with three feedback levels:
  • Wrong — The recommendation is incorrect
  • Dismiss — Not relevant to this document
  • Optional — Valid but not needed for this agreement
• Set priority — Mark as CRITICAL, IMPORTANT, or OPTIONAL for future reference

Use the importance slider to filter out low-scoring recommendations, and the status tabs to switch between Pending, Accepted, and Rejected views.

Hovering a recommendation scrolls the document viewer to the corresponding clause and highlights it, and vice versa.

Preferences

Over time, as you review documents aligned against a standard, ContractRabbit learns from your feedback. The Preferences page shows these learned adjustments:

• Weight adjustments — How much emphasis to place on specific clause types (scale of -1 to +1)
• Permanent dismissals — Issues you've told ContractRabbit to stop flagging
• Permanent flags — Issues you always want highlighted

You can:

• Edit individual preference weights
• Bulk-delete selected preferences
• Reset all preferences to start fresh

Each preference shows its category, type, weight, notes, and creation date.

Exporting results

After accepting recommendations, you can export the document in two formats:

• Redline — A .docx file with Word-compatible tracked changes showing all accepted modifications
• Clean final — A .docx file with all accepted changes applied (no tracked changes visible)

Using standards with the Agent

Once you've set up a standard, you can ask the Agent to "align this document with [standard name]" and it will run the full alignment pipeline, streaming recommendations as they're generated.