> ## Documentation Index > Fetch the complete documentation index at: https://docs.getzell.com/llms.txt > Use this file to discover all available pages before exploring further. # Scorecards & Signals > Learn how to create scorecards, manage signals, and configure evaluation criteria for your team ## Overview Scorecards evaluate calls, emails, and interactions based on **signals** — reusable evaluation criteria that can be shared across multiple scorecards. Each signal defines *what* to detect and *how* to score it. A scorecard is simply a weighted collection of signals assigned to your team's interactions. ## Key Concepts A single evaluation criterion with a detection method and scoring config. Signals are reusable across scorecards. A named collection of signals, each with a weight and order. Assigned to agents or teams. How the signal is evaluated: LLM Rubric, Keyword match, Metric, or Semantic similarity. How much each signal contributes to the overall scorecard score (weights sum to 100%). *** ## Signal Detection Methods Signals support four detection methods, each suited to different evaluation needs: ### LLM Rubric The AI judges the transcript against a rubric you define. Best for nuanced, qualitative criteria. | Field | Description | | -------------------- | ------------------------------------------------------------------ | | **Description** | What the signal checks (e.g., "Did the rep handle the objection?") | | **Evaluation Guide** | Instructions for reviewers on what good/bad performance looks like | | **Rubric** | Structured scoring criteria with examples | | **Mode** | `YES_NO` (binary pass/fail) or `RANGE` (0–10 scale) | ### Keyword Deterministic phrase matching over the transcript — no LLM needed. Fast and consistent. | Field | Description | | -------------------- | ------------------------------------------------------ | | **Phrases** | List of phrases to detect (at least one required) | | **Speaker** | Who to listen for: `rep`, `customer`, or `any` | | **Match Word Forms** | Whether to match conjugations and plurals | | **Bands** | Maps hit count → score (e.g., ≥1 hit → 10, 0 hits → 0) | ### Metric A numeric measure computed from call timing data. No LLM. | Field | Description | | ---------- | --------------------------------------------------------------------------- | | **Metric** | One of: `talk_ratio`, `longest_monologue`, `question_rate`, `speaking_time` | | **Bands** | Maps measured value → score | ### Semantic Embedding similarity match against example phrases. Useful for detecting intent without exact keyword matches. | Field | Description | | -------------------- | ------------------------------------------------ | | **Prompt** | What you're looking for semantically | | **Examples** | Optional example phrases to improve matching | | **Threshold** | Similarity threshold (0–1) to trigger a match | | **Pass/Fail scores** | Score when above/below threshold (default: 10/0) | *** ## Creating Signals Go to **Team Management** → **Signals**. Click **Create New** and choose a detection method based on your evaluation need: * Use **LLM Rubric** for subjective or complex criteria * Use **Keyword** for specific phrases you want to detect * Use **Metric** for call timing measurements * Use **Semantic** for intent-based detection Fill in the configuration fields for your chosen method. Be specific in descriptions and provide examples where possible. For LLM Rubric signals, writing a clear evaluation guide significantly improves scoring consistency. Signals have three statuses: * **Published** — active and available for use in scorecards * **Draft** — work in progress, not yet used for evaluation * **Archived** — retired, no longer used You cannot delete a signal that is currently in use by a scorecard. Archive it instead. Click **Save**. The signal is now available to add to any scorecard. *** ## Creating a Scorecard Go to **Team Management** → **Scorecards**. Click **Create New** to start from scratch, or choose **Duplicate** to clone an existing scorecard. Duplicating an existing scorecard is a quick way to create variations for different teams or use cases. * Set a descriptive name (e.g., "Sales Discovery Call" or "Support Quality Check") * Optionally add a description and difficulty level * Select applicable call types Add signals to your scorecard from the signal library: * Browse or search available signals * Click **Add** to include a signal * Reorder signals by dragging them into the desired sequence Each signal can only appear once per scorecard. Assign a weight to each signal to control its contribution to the overall score. Weights are distributed as percentages that sum to 100%. If you want equal weighting, leave the defaults — the system automatically distributes weights evenly across all signals. Save your scorecard, then assign it to agents or teams. Run a test evaluation to verify your signals and weights produce meaningful results. *** ## Cloning & Reusing Signals Signals are designed to be reusable: * **Shared across scorecards** — One signal (e.g., "Proper greeting") can be used in multiple scorecards with different weights. * **Signal library** — Global library signals (maintained by Zell) are available to all tenants. Clone them to customize. * **Clone to edit** — Use **Clone** to create an editable copy of any signal (library or your own) that you can modify independently. *** ## Importing from Gong When importing scorecards from Gong, each question is automatically converted into an **LLM Rubric** signal. The answer guide becomes the signal description, and all signals are assigned equal weights. *** ## Best Practices Each signal should evaluate one specific behavior or criterion. Avoid combining multiple checks into a single signal. Before creating a new signal, check if one already exists in the library. Reusing signals ensures consistent evaluation across teams. LLM Rubric is powerful but slower. Use Keyword or Metric signals for simple, deterministic checks — they're faster and more predictable. Run test evaluations after setting up a scorecard. Adjust weights and signal configs based on the results.