DeepEval

# Fix list: DeepEval

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/deepeval, the October 2026 research run, assessed 9 October 2026. Grade B, 64.7 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on DeepEval: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 47 out of 100, up to 9.3 more on the total

Why it scored 47: Local runs need no credential. Provider keys and `CONFIDENT_API_KEY` are read from the environment or a dotenv file, and storing keys in the legacy `.deepeval/.deepeval` file is deprecated in the source. Confident AI project keys take a name and an expiry and can be deactivated, rotated with a grace period or deleted. No per-permission scopes were found (20). The framework has no read-only mode. Once logged in, test runs and traces upload to Confident AI (8). Metrics pass application output to an LLM judge, and no guidance on untrusted content reaching the judge was found in the pages read (5). Each run is recorded in local JSON or SQLite. Confident AI documents audit log exports, which we did not read (7). GitHub shows no security policy and no published advisories, and security.txt answered 404 on both domains. No bug bounty was found. The Confident AI site states SOC 2 Type II and HIPAA, and its trust centre is drawn by script and was not read (7).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Reliability, 59 out of 100, up to 8.2 more on the total

Why it scored 59: Scored on the local-software lines, since DeepEval runs where the owner installs it. Official `deepeval` packages on PyPI and npm, with Python 3.9 or later stated (20). Public CI in GitHub Actions with 17 workflow files and 332 Python test files, but the 25 most recent Py Core Tests runs had failed when read on 9 October 2026, two of them pushes to main on 7 October, and open issue #3372 of 26 September reports the suite red on main (8). 322 open issues and 387 open pull requests on a repository with 18.7k stars. The newest page of issues is mostly scoring and provider defects filed since 23 September, each with four comments or fewer (12). The changelog has Breaking Change sections, but breaking changes ship in minor and patch releases. 4.2.0 reversed the score direction of four metrics and 3.9.8 and 3.9.9 removed options (7). Python is at 4.2.8. The TypeScript package is at 0.9.21 (12).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 3. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: No x402, MPP or L402 (0). DeepEval is free, and the paid option is the hosted Confident AI platform, whose prices are public without a login. Starter is $200 a month, Team $2,000 a month, and trace spans beyond the plan are $1 per GB-month (20). The framework is free and Confident AI's Free plan needs no card (20). `pip install -U deepeval` and a local run need no account, so an agent can start without a person signing up. Confident AI itself needs a browser sign-up, and its auth.md says there is no self-serve registration endpoint for API keys. Scoring this line on the local framework is a judgement call (20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 4. Agent ergonomics, 72 out of 100, up to 4.6 more on the total

Why it scored 72: Read on the framework lines. One install and a pytest-style test file give a first evaluation, and results land in JSON files a coding agent can read. `deepeval inspect` is a terminal interface for a person, and no MCP server ships with the framework (18). `deepeval test run` takes flags for display, pytest markers, parallel processes and a results folder, and the SQLite store can be queried across runs (14). `--ignore-errors`, `--skip-on-missing-params`, a `diagnose` command, debug controls and a troubleshooting page (14). `--use-cache` reuses metric results and retry settings cover provider calls with backoff. Scores from LLM judges can differ between runs (12). Python and TypeScript packages with few required parameters. The default judge needs `OPENAI_API_KEY` (14).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Schema & documentation, 78 out of 100, up to 3.6 more on the total

Why it scored 78: Read on the framework lines. The Python package ships `py.typed` and pydantic models, the TypeScript package ships type declarations, and the CLI is built on typer. A library has no OpenAPI file to publish (18). llms.txt on deepeval.com indexes the docs, and pages carry a Copy Markdown control (10). Each metric has its own page with required parameters and usage, and three agent skills in the repository tell a coding agent how to add evaluations and tracing (16). Typed test case and metric classes, with enums for evaluation parameters and CLI choices (12). Many runnable examples, a troubleshooting page and an environment variable reference. Error types are not listed in one place (11). Tagged releases for both packages and a public changelog by year, but the 2026 changelog stops at 4.2.0 of 25 August and has no entry for 4.2.1 to 4.2.8 (11).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 6. Transparency & trust, 63 out of 100, up to 3.2 more on the total

Made of editorial 73, provenance 52.

Why it scored 63: Apache-2.0 for the whole repository (30). The docs state what telemetry collects, and Confident AI publishes terms, a privacy policy, a DPA and a sub-processor list with regions. They disagree in places. The privacy policy says the DPA is available on request while it is published at /dpa, and the DeepEval docs say data sits in a private cloud on AWS while the sub-processor list names Supabase, ClickHouse and Railway for storage and hosting (20). The changelog has Breaking Change sections and the source warns at run time about the reversed metrics and the legacy key file. No deprecation policy or notice period was found (9). Telemetry through PostHog is on by default and `DEEPEVAL_TELEMETRY_OPT_OUT=1` turns it off. The source also sends the judge provider and model, the runtime kind, the CLI command and test case counts, which the docs list omits (14).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: confident-ai.com, registered 2023-08-15 (3 years) (7 of 15)
- Terms of service: not found (0 of 10)
- Privacy policy: not found (0 of 10)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 83 out of 100, up to 1.5 more on the total

Why it scored 83: Python 4.2.8 was tagged on 2 October 2026, seven days before the check (30). 14 Python releases tagged since 9 August, from 4.1.6 to 4.2.8 (20). 124 commits in the 30 days to 9 October, most by three maintainers, and a Discord community. 322 issues and 387 pull requests are open and reply times were not read (14). Current official packages in Python and TypeScript, the latter at 0.9.21 on 30 September (15). A `poetry.lock` file and CI on pull requests. No Dependabot configuration was found and the core test workflow is failing on main (4).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: the trust centre at trust.oneleet.com/confident-ai, which is drawn by script. Its robots.txt request returned the application's HTML shell. SOC 2 Type II and HIPAA are the vendor's statements on its own site
- unchecked: whether Confident AI has a status page. No link to one was found on the pages read
- unchecked: Confident AI's REST API reference, OpenAPI file, MCP tool reference, audit log and data retention pages. The hosted platform is not the surface graded, and robots.txt disallows /docs-openapi
- unchecked: GitHub release notes for 4.2.1 to 4.2.8. The tags page is closed by robots.txt, and the dates come from the cloned tags
- unchecked: reply times on issues and pull requests. Only the first page of open issues was read
- unchecked: the Node.js versions the TypeScript package supports. No `engines` field was found in `typescript/package.json`
- Confident AI's REST API and hosted MCP server at https://mcp.confident-ai.com/mcp are a separately documented surface. They may merit a listing of their own
- The Payments onboarding line is scored on the local framework, which needs no account. Scored on Confident AI it would be 0, and Payments would be 40
- Confident AI's terms forbid using its services for competitive analysis. No clause on automated access was found. This matters before any probe of the hosted platform is run
- The robots.txt request answered 400 on rdap.verisign.com and 404 on api.npmjs.org and pypistats.org, so those hosts publish no rules and were read on that basis
- The terms and privacy links in `provenance` are Confident AI's and cover the optional hosted platform. The framework itself is governed by the Apache-2.0 licence

## Weaknesses

- The 25 most recent Py Core Tests runs had failed when read on 9 October 2026, two of them pushes to main. Issue #3372 of 26 September reports the same
- No SECURITY.md, no published advisories and no security.txt on deepeval.com or confident-ai.com. The trust centre is drawn by script and was not read
- Release 4.2.0 reversed the score direction of four safety metrics in a minor version. The changelog marks it as breaking and the code warns at run time
- The 2026 changelog stops at 4.2.0 of 25 August. Releases 4.2.1 to 4.2.8 have no entries in it
- Usage telemetry is on by default and sends the public IP address. The source also sends the judge model, runtime kind and CLI command, which the docs list omits
- Confident AI's terms forbid using its services for competitive analysis. This matters before any probe of the hosted platform is run

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Set `DEEPEVAL_TELEMETRY_OPT_OUT=1` before the first run if usage events and the public IP address should not go to PostHog
- Set a judge model key such as `OPENAI_API_KEY`, or use the non-LLM metrics. Most metrics call an LLM judge and bill the owner's provider account
- Review thresholds for `BiasMetric`, `HallucinationMetric`, `MisuseMetric` and `ToxicityMetric` when upgrading past 4.2.0. Higher scores now mean better
- Read results from `.deepeval/.latest_run_full.json` or a `results_folder`. `deepeval inspect` opens a terminal interface meant for a person
- Pass an existing key with `deepeval login --api-key` in CI. Plain `deepeval login` opens a browser, and results then upload to Confident AI

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.