Skip to content

fix: resolve Built-in eval scorers: faithfulness, answer releva (#5) #7

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
boumarate wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: resolve Built-in eval scorers: faithfulness, answer releva (#5) #7

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 27 additions & 0 deletions typescript/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,3 +2,30 @@

The TypeScript implementation of threads. It must pass everything in `../spec/conformance/`.
Rules: `../AGENTS.md` (Zod 4, strict tsconfig, Bun, Biome).

## Built-in Eval Scorers

`threads` provides three built-in evaluation scorers to grade your agent's runs without writing custom judges from scratch. These scorers can be used directly in a case's `expect` block or programmatically via `runEvals()`.

### 1. Faithfulness (`faithfulness`)
- **Type**: Judge-based (uses the case's configured LLM judge model).
- **Score Range**: `0.0` to `1.0`
- **Description**: Evaluates whether the generated response is grounded in and supported strictly by the provided context/retrieved documents. It detects hallucinations or claims not supported by the source text.

### 2. Answer Relevancy (`answerRelevancy`)
- **Type**: Judge-based (uses the case's configured LLM judge model).
- **Score Range**: `0.0` to `1.0`
- **Description**: Evaluates how relevant the generated response is to the user's input query. It penalizes redundant, incomplete, or off-topic answers.

### 3. Tool-Call Correctness (`toolCallCorrectness`)
- **Type**: Deterministic (no LLM calls).
- **Score Range**: `0.0` to `1.0`
- **Description**: Compares the tool calls executed during the run against the expected tool calls. It checks for tool name matches, argument correctness, and execution order.

---

## Usage Examples

### Using Scorers in a Case's `expect` Block

You can define these scorers directly inside your test cases. The judge-based scorers will automatically use the case's configured judge model.