Kev by Jared Palmer

# Fix list: Kev

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/jaredpalmer-kev, the October 2026 research run, assessed 1 October 2026. Grade B, 67.4 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 Kev: 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, 49 out of 100, up to 8.9 more on the total

Why it scored 49: Read as software you run. No account. `kev.serve` binds to 127.0.0.1 and is open by default, and one optional bearer key (`KEV_API_KEY`) guards `/v1/*`. The deploy skill tells agents to always set it on Modal (15 of 30). A decision model has no write actions, so there's nothing to approve, and an answer is only as safe as what the caller does with it (15 of 20). Caller text is tokenised so it can't produce Kev's delimiter tokens, questions can't read each other, and the playground has presets for fake delimiters. Nothing documents how hostile text in the state can move an answer, and the cards say not to make consequential decisions about people without human review (9 of 15). Each response carries a request ID, token usage and model time. The server keeps no log of calls (6 of 15). No SECURITY.md, disclosure policy or advisories. Weights are pinned by Hub revision and the release tarballs carry SHA-256 checksums. The pointer head is a pickled `head.pt` loaded with `torch.load`, which PyTorch 2.6 and later, the versions Kev requires, read in weights-only mode by default (4 of 20).

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, 73 out of 100, up to 5.4 more on the total

Why it scored 73: Scored on the local-package checklist, since Kev is a model you run. It isn't on a package index, because the PyPI name `kev` belongs to an unrelated 2021 project. It installs from the repository with uv and a lockfile, with Python 3.12 and 3.13 stated, or from release tarballs with SHA-256 checksums (10 of 20). Public CI on GitHub Actions runs the unit tests and the playground build on every push, and the latest runs on main had passed when we looked on 2 October. The model parity and API tests need weights and aren't in CI (20 of 25). One open issue, #8 from 20 September on date arithmetic, which the README lists as a limitation with a workaround, `KEV_DATE_FACTS=1` (23 of 25). The Kev 1.0 release notes say what changed, including the server refusing over-long states with a 422 where it used to cut them silently, but the Python package has stayed at 0.1.0 and there's no changelog file (10 of 15). Kev 1.0 fixes the four checkpoints as one versioned family, while the package classifier still says alpha (10 of 15).

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: Free Apache-2.0 software with nothing to buy from Kev, so 20 + 20 + 20 for pricing, free use and no sign-up. No payment protocol (0). GPU time is yours, on your own hardware or on Modal at the hourly rates the deploy skill lists.

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. Transparency & trust, 49 out of 100, up to 4.5 more on the total

Made of editorial 70, provenance 27.

Why it scored 49: Apache-2.0 for the code, adapters and heads, on Apache-2.0 Qwen bases. The model cards give the training recipe, and the document and skill training partitions are on the Hub. Kev-27B's base is Qwen's post-trained release, whose training data is unknown, and the card says so (30). Self-hosted, so inputs stay on your hardware, which the model card states. There's no other data statement, and nothing on what a Modal deploy logs (20 of 30). Earlier versions stay at Hub tags, and the retired `kev-family` release points to `kev-1.0`. No deprecation policy (10 of 20). We found no telemetry code in the `kev` package and the docs say nothing either way. Weights download through the Hugging Face Hub client (10 of 20).

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):

- Legal entity named: not found (0 of 20)
- Domain age: github.com/jaredpalmer, no registry record we could read (0 of 15)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 5. Schema & documentation, 77 out of 100, up to 3.7 more on the total

Why it scored 77: Read for a model you serve yourself. The server is FastAPI with Pydantic request models (`SystemOneRequest`, `Noul`, `Choice`, `Score`) and follows TypeSafe's published System One contract, but Kev publishes no spec file of its own (18 of 25). No llms.txt. The README, model cards, release notes and two agent skills are Markdown in the repository (5 of 10). Each model card has intended and out-of-scope uses, and the README and release notes say where Kev trails Jev, such as knowledge questions, date arithmetic and option order (18 of 20). Three question types with 1 to 255 options or levels, and a 65,536-token state limit enforced with a 422. State is free-form by design (12 of 15). curl and Python examples, a sample response, and the 422 and 401 cases described. No full error table (12 of 15). Hub tags pin every version (`v1.0`, `v1`, `v1-lora`) and the release notes are dated, with no separate changelog (12 of 15).

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. Agent ergonomics, 78 out of 100, up to 3.6 more on the total

Why it scored 78: Read as an API an agent calls for a decision, as with Jev and Clef. Answers are a probability per allowed option, so output stays small, and the server caches the state, so more questions about the same document pay only for the questions. Validated context is 8,192 tokens on the three smaller models and 65,536 on Kev-27B (20 of 25). The caller sets the output shape, with any number of questions a request, and `/v1/systemone/separate` and `/v1/systemone/permute` check question isolation and option order (18 of 20). A 422 names the state's token count and the limit, and a missing key gets a 401 saying how to send it. Other errors aren't documented (15 of 20). Calls are stateless and safe to retry. There's no retry guidance and no rate limiting in the server (15 of 20). TypeSafe's Python SDK works unchanged per the README, and `kev-latest` is the default model, but setup is a clone, a uv sync and a GPU or a Mac (10 of 15).

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.

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

Why it scored 83: Read for an open-weight model. Kev 1.0 released on 1 October 2026 (30). Kev-27B v2 and Kev-9B v2 on 30 September and Kev 1.0 on 1 October, after the first family release on 24 September (20). One open issue, from 20 September, answered in the README. Outside pull requests have been merged (#14, #108, #175), but Jared Palmer wrote 312 of the 333 commits we cloned and a Devin bot 12 more (18 of 25). No SDK of its own. Kev reuses TypeSafe's Python SDK, and agent skills handle deploys and fine-tunes. Not an MCP server, so no registry entry applies (8 of 15). CI passes on main and dependencies are locked with uv, with torch capped below 2.9 (7 of 10).

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: GitHub stars and forks. Our reader returned a stale repository page (7 stars, 0 forks and an old description), so popularity is blank
- Hugging Face downloads and likes for kev-9b read differently in two fetches (3,474 and 59 on the model page, 989 and 36 from the API), so we left them out
- unchecked: reply times on closed issues and pull requests
- Kev-9B on a Mac. The README says it's expected to fit a 32 GB Mac but hasn't been measured
- The accuracy and calibration figures are the author's, on the author's suites and on the Decision Index's datasets. We haven't run them, and Kev-9B's Hub tags list BANKING77 among its training datasets

## Weaknesses

- No package. `pip install kev` installs an unrelated 2021 ORM, so Kev runs from a Git clone with uv
- Kev-0.8B, 4B and 9B are validated to 8,192 tokens of state, though the server accepts 65,536
- Jared Palmer wrote 312 of the 333 commits we cloned
- No SECURITY.md, disclosure policy or advisories, and the server is open unless `KEV_API_KEY` is set
- Below 27B it trails Jev on knowledge questions and date arithmetic, with MMLU-Pro at 0.59 for Kev-9B against Jev's 0.84

## 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.

- Install from the repository. The `kev` package on PyPI is an unrelated project
- Pin a checkpoint with `@v1.0`, as in `jaredpalmer/kev-4b@v1.0`, so tuned thresholds keep their meaning
- Keep states under 8,192 tokens on Kev-0.8B, 4B and 9B, or use Kev-27B for long documents
- Set `KEV_DATE_FACTS=1` when a decision depends on the gap between two dates
- Expect a 422 naming the token count when a state passes 65,536 tokens. The server refuses it instead of cutting it

## What the review panel asked for

- package versions tracking releases
- a written deprecation policy
- throughput per GPU table
- cost per 1,000 calls in README

## 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.