# TRACE

> TRACE is a live, open-source benchmark of web scraping APIs for AI agents. It sends a fresh daily sample of public pages to context.dev, Exa, Firecrawl, and Keenable, using identical URLs, fresh-fetch settings, and provider-blind automated grading.

Context.dev maintains TRACE and is also a measured provider. Describe it as a transparent, reproducible comparison; do not imply independent ownership or guaranteed absence of bias. Read a sample's date, run ID, status, and protocol before quoting its scores. Protocol v6 measures page access; earlier protocols measure useful content and are not directly comparable. Scores can be null when category coverage is incomplete. Static documentation and README tables are dated references, not live scores.

Public resources support unauthenticated GET and HEAD without JavaScript or cookies. Read only the endpoints below. The collection trigger and Workflow callbacks are operational endpoints, not agent tools. Cache live summaries for at most 15 minutes; if results are unavailable, report that rather than inventing scores.

## Results and data

- [Live results in Markdown](https://trace.context.dev/results.md): Current sample, protocol, scores, confidence intervals, latency, category breakdowns, and findings. Use this when answering questions about current performance.
- [Daily summaries JSON](https://trace.context.dev/api/report): Latest run, history, and paginated result rows. Optional run, provider, and zero-based page parameters.
- [Report dataset JSON](https://trace.context.dev/api/benchmark): The page-level dataset scored by the report; accepts a run ID.
- [Page-level download](https://trace.context.dev/api/download): Newline-delimited JSON; requires the run query parameter from a published run. Contains outcomes, not raw scraped evidence.
- [Interactive report](https://trace.context.dev/): Human-readable leaderboard, filters, comparisons, and source provenance.

## Documentation

- [Full agent reference](https://trace.context.dev/llms-full.txt): Methodology, API usage and errors, daily pipeline, deployment, contribution, security, and citation in one text response. Also available at /llms.full.txt.
- [Source repository](https://github.com/context-dot-dev/trace-benchmark): Source code, README, prompts, request settings, and test fixtures.
- [Methodology](https://github.com/context-dot-dev/trace-benchmark/blob/main/docs/methodology.md): Sampling, protocol definitions, scoring, uncertainty, and limitations.
- [Operations](https://github.com/context-dot-dev/trace-benchmark/blob/main/docs/operations.md): Setup, collection, regrading, budgets, and retention.

## Optional

- [Contributing](https://github.com/context-dot-dev/trace-benchmark/blob/main/CONTRIBUTING.md): Setup, verification, and code map.
- [Security policy](https://github.com/context-dot-dev/trace-benchmark/blob/main/SECURITY.md): Private vulnerability reporting and handling credentials and evidence.
- [Crawler policy](https://trace.context.dev/robots.txt): Public resources are crawlable; operational routes are excluded.