ScaleDown

# Fix list: ScaleDown

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/scaledown, the October 2026 research run, assessed 14 October 2026. Grade D, 48.6 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 ScaleDown: 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. Reliability, 20 out of 100, up to 16 more on the total

Why it scored 20: Read as a hosted model API. No status page found. status.scaledown.ai and status.scaledown.xyz don't resolve, and the site and docs link to none (0). No readable incident history (5). No rate limit with numbers. The site promises generous fair-use limits and the terms (2.3) let ScaleDown change limits on reasonable notice (0). Only the Compress reference lists 429 with advice to back off and retry. Classify, Decisions, Extract and Summarize list 502 or 500 and 504 without retry guidance. A call has no side effects, so a retry is safe (5 of 15). No SLA published. llms.txt mentions SLAs only for enterprise and self-hosted contracts (0). The models carry no beta or preview label and are listed on OpenRouter and Vercel AI Gateway (10).

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.

## 2. Security & auth, 48 out of 100, up to 9.1 more on the total

Why it scored 48: Read for a model API, on the lines used for Jev. A plain API key in `x-api-key`, or as a bearer token on chat completions. We found no scopes, expiry or rotation documented and didn't check revocation, since the dashboard is behind sign-in and disallowed in robots.txt (18 of 30). The terms (6.3), the DPA (20 July 2026) and the FAQ agree that customer data isn't used for training. The DPA says request data is held only in memory during inference, operational logs without customer data are kept up to 90 days, and processing stays in the United States. llms.txt says prompts are never logged, which the DPA's operational logs qualify (17 of 20). Classify and Extract judge text that often comes from users. We found no warning or mitigation for injected instructions (3 of 15). Each Decisions response reports tokens and cost. llms.txt puts audit logs on enterprise plans only (6 of 15). No security.txt (404), disclosure policy or bug bounty found. The DPA commits to breach notice within 48 hours. Certification claims conflict (DPA SOC 2 Type II and ISO 27001, homepage SOC 2 Type 1 badge, llms.txt in progress), and we found no trust centre or report to settle them (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.

## 3. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 on ScaleDown's endpoints or through the gateways that list it (0). Per-token price published without a login, $0.05 per million input tokens with output free (20). The homepage says new accounts get 50 million free tokens with no credit card. The docs introduction says to contact sales for a key, and we didn't sign up, so the free tier rests on the homepage (20). A person signs up in a browser and creates a key in the dashboard (0).

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. Schema & documentation, 69 out of 100, up to 5 more on the total

Why it scored 69: An OpenAPI 3.1 document (ScaleDown API 1.0.0) covers /compress/raw/, /classify and /v1/scaledown. Its /classify schema has only `text` and `labels`, without `chunks`, `document`, `multi_label` or the label caps, and Extract and Summarize on /v1/chat/completions aren't in it (15 of 25). llms.txt indexes 28 pages, each with a Markdown twin (10). Overviews say when to use each model and how to write rubrics, and a table says when to choose /classify over Decisions. No known-limitations page (14 of 20). Typed question kinds behind a `type` field, with 20 labels, 16 in multi-label mode, 50 chunks and 2 to 10 score levels stated in prose, and Extract driven by a JSON schema with `$ref` refused (11 of 15). curl, Python and TypeScript examples with sample responses, and an error table per endpoint (422, 402, 502, 504 and others), though 429 appears only for Compress and the error body is shown only by our 401 (12 of 15). Pinned model IDs (`classify-1`, `extract-1`, `summarize-1`). No public changelog, but the docs source on GitHub has dated commits (7 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.

## 5. Transparency & trust, 54 out of 100, up to 4 more on the total

Made of editorial 56, provenance 52.

Why it scored 54: Closed hosted models under published terms of service (20 July 2026), a privacy notice (3 June 2026) and a DPA, naming Carbon ScaleDown, Inc. and Delaware law. The legal pages render only in the site's JavaScript and the server answers them with 404 (15). The terms, DPA and FAQ agree on no training, and the DPA states retention (none for request data, 90 days for operational logs). llms.txt's claim that nothing is logged and the conflicting certification claims count against agreement (20 of 30). No deprecation policy or dated notices. The terms promise commercially reasonable efforts (5 of 20). The DPA lists four subprocessors (AWS, Google Cloud, Modal Labs, Supabase), all in the United States, and commits to 30 days' notice of any change that would break its retention or training commitments (16 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):

- Domain age: scaledown.ai, registered 2025-03-28 (1 year) (3 of 15)
- Terms of service: published, but our reader couldn't read it (7 of 10)
- Privacy policy: published, but our reader couldn't read it (7 of 10)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 6. Agent ergonomics, 76 out of 100, up to 3.9 more on the total

Why it scored 76: Read as an API an agent calls for a decision, the reading used for Jev. Decisions answers are a few numbers per question, and the reference's sample answers a three-way choice from 62 input tokens with 1 output token. Extract adds 500 characters of context on each side of every value by default, which `context_chars` can turn off, and inputs reach 1M tokens per the site FAQ (20 of 25). `threshold`, `top_n`, `context_chars`, multi-label mode, the `reasoning` switch and up to 50 chunks a request sized against one billed label set (17 of 20). Error tables per endpoint, with 402 for insufficient credit and a plain `detail` message on 401. Error bodies aren't documented (13 of 20). Calls have no side effects. No retry guidance outside Compress, and the langchain-scaledown client doesn't retry (14 of 20). No SDK of ScaleDown's own. OpenAI SDKs work against /v1/chat/completions with only the base URL changed, a Jev client works against /v1/scaledown, and langchain-scaledown, LiteLLM and OpenRouter cover Python and TypeScript callers. `model` is optional on Decisions (12 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, 56 out of 100, up to 3.9 more on the total

Why it scored 56: Read for a closed model, on the lines used for Jev. Classify, Extract and Summarize reached Vercel AI Gateway and OpenRouter on 13 October 2026, and the docs changed on 14 October (30). Pinned `-1` model IDs exist, but the terms (2.2) promise only commercially reasonable efforts to give notice of material adverse changes, and there's no deprecation policy (5 of 12). One version of each model so far. The docs dropped the native /extract and /summarization/abstractive endpoints and the batch pages from the reference on 13 October, while the site and LiteLLM's guide still name those paths (2 of 8). No public changelog. Dated commits in the docs repository, a contact address and a Discord bot repository, and issue replies weren't sampled (6 of 15). No SDK of ScaleDown's own. langchain-scaledown 0.2.0 (MIT, 12 October 2026) is published by the scaledown-team organisation, and OpenAI SDK compatibility covers other languages (8 of 15). langchain-scaledown ships GitHub workflows and unit and integration tests, released two days ago (5 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.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- Conflicting compliance claims on ScaleDown's own pages, read 14 October 2026. The DPA (20 July 2026) says ScaleDown maintains a SOC 2 Type II attestation and ISO 27001 certification, the homepage shows SOC 2 Type 1, HIPAA and ISO 27001 badges, and llms.txt says GDPR, HIPAA and SOC 2 compliance is in progress. At least one is wrong. No report or trust centre found to settle it (-2) (https://scaledown.ai/dpa, https://scaledown.ai/llms.txt)

## 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 sign-up flow, the dashboard and whether keys can be scoped, rotated or revoked, since /dashboard, /settings and /getapikey are disallowed in robots.txt and we made no account
- Whether the free tier needs a card. The homepage says no card and 50 million free tokens. The docs introduction says to contact sales. The lead's 'first 50 hours' limit wasn't found on any ScaleDown page
- Which compliance claim is current (SOC 2 Type II and ISO 27001 in the DPA, SOC 2 Type 1 badge on the homepage, in progress in llms.txt)
- Whether the native /extract and /summarization/abstractive endpoints still answer after the docs dropped them on 13 October 2026. LiteLLM's guide and the site still use them
- Rate limits, whether 429 carries Retry-After, and whether failed calls are charged
- The lead's DecisionBench figures (77.2 per cent against Jev's 72 per cent) weren't found on ScaleDown's site, docs or the LangChain post
- TrustedRouter, RouterPlus and an Arthur AI integration weren't confirmed on a first-hand page. Arthur appears only as a logo on the homepage
- Classify isn't in OpenRouter's model list and Extract and Summarize aren't in Vercel AI Gateway's, so the gateways each carry part of the family

## Weaknesses

- No status page, no published rate limits and no SLA outside enterprise contracts
- No public changelog. The docs dropped the native /extract and /summarization endpoints from the reference on 13 October 2026
- Compliance claims disagree between the DPA (SOC 2 Type II, ISO 27001), the homepage badges (SOC 2 Type 1) and llms.txt (in progress)
- No security.txt, disclosure policy or bug bounty, and no guidance on hostile text in classified input
- The OpenAPI document covers /classify, /v1/scaledown and Compress only, without the documented limits

## 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.

- Send several questions about one item in one `/v1/scaledown` call. Each becomes its own Classify call and the usage is summed
- Pin `classify-1`, `extract-1` or `summarize-1` once you've tuned thresholds on your own labelled cases
- Set `context_chars` to 0 and use `top_n` on Extract unless you need the 500 characters of context each side
- Set `reasoning: false` on /classify when you only need the label. The reasoning call bills its input at 2x
- Back off and retry on 429, 502 and 504. No limits are published and calls have no side effects

## 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.