mammoth-1

# Fix list: mammoth-1

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/mammoth-1, the October 2026 research run, assessed 21 October 2026. Grade D, 47.7 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 mammoth-1: 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. Reach, 40 out of 100, up to 23.1 more on the total

Why it scored 40: Memory 10 of 30, the only build is BF16 at 54.7 GB, 65.7 GB with a fifth added, which fits the reference machine but nothing smaller. Runtimes 0 of 25, the trained attention mask and readout run only on OneJev's qev runtime at a pinned commit with the repository's patch, and released runtimes would run the weights causally. Builds 0, no quantised build. Terms 15, MIT named in the card's metadata. Download 10, not gated. Hosted 5, Baseten's own Model API serves it.

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

Whether an agent can run it, and where.

- 0 to 30, memory. The smallest build from the publisher or a runtime's own library that keeps 4-bit precision or better, at its file size plus a fifth for the runtime and a working context. 30 for 16 GB or less, 25 for 24 GB, 20 for 32 GB, 15 for 64 GB, 10 for what fits our reference machine (the GPU share of 128 GB of unified memory), 5 for 256 GB, 0 above.
- 0 to 25, runtimes. 5 for each of llama.cpp, Ollama, vLLM, SGLang, LM Studio and Hugging Face Transformers that runs the architecture in a released version, from the runtime's own code, library or documentation. A publisher's fork or an unreleased branch doesn't count. 25 at most.
- 0 to 15, builds from the publisher. 15 for quantised builds at two precisions or more (FP8, 4-bit, GGUF), 10 for one, 0 when only other people's quantisations exist.
- 0 to 15, terms of use. 15 for an OSI-approved licence with nothing added, 10 for a custom licence that lets anyone use it commercially for free below a threshold or under a use policy, 5 for research or non-commercial use only, 0 for no licence or unclear terms.
- 0 to 10, download. 10 when the files download with no account and no terms to accept, 5 for a click-through that needs an account, 0 when a person approves each request.
- 5, hosted too. The publisher's own API serves it, or at least two inference providers do, so an agent without the hardware can still call it.

When our reference-machine runs start, speed on that machine (tokens per second and time to first token at the build that fits) joins Reach.

## 2. Risk, 50 out of 100, up to 15.4 more on the total

Why it scored 50: File safety 15 of 25, safetensors only, but loading it as trained runs the repository's setup.sh, engine.py and serve.patch. Provenance 15, Baseten's own pages link to the organisation and every file has a hash. Safety evaluation 0, none published. Agent guidance 5, general advice on thresholds and escalation. Licence risk 10, the card went from proprietary and evaluation only to MIT on release day with no note. Security programme 5, Baseten's company-wide disclosure contact.

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

What could go wrong when you run it, and what the publisher has done about it. A higher score means less risk.

- 0 to 25, file safety. 25 for weights in safetensors or GGUF only, 15 when pickle files (.bin, .pt, .pth) sit beside them, 0 for pickle only. 10 off when the documented way to load it or to build its prompts runs code from the repository (`trust_remote_code`, or the publisher's scripts in place of a template).
- 0 to 15, provenance. 10 when the files come from the publisher's own organisation, the one its site or docs link to, and 5 for a checksum on every file.
- 0 to 20, safety evaluation. 20 when results are published and someone other than the publisher ran them or can rerun them, 10 for the publisher's own results, 0 for none.
- 0 to 15, guidance for agent use. 15 for documented prompt-injection guidance or an instruction hierarchy the model was trained to follow, 10 for a statement of what its trained refusals cover and the safeguards to add around it, 5 for general cautions, 0 for none.
- 0 to 15, licence risk. 15 to start, 5 off for each of these, a use policy or field-of-use limits beyond the law, terms the licensor can change or revoke, a revenue or user threshold, and a licence that changed without a dated note or whose link is broken. Not below 0.
- 0 to 10, a security programme. 5 for a way to report a flaw (security.txt, a disclosure policy or a security contact), 5 for advisories handled in public.

When our reference-machine runs start, measured robustness (prompt injection with AgentDojo, over-refusal with XSTest) joins Risk.

## 3. Agent interface, 55 out of 100, up to 6.9 more on the total

Why it scored 55: Read as a decision model, by what it takes and returns. Chat template 10 of 20, the repository's template is the base's and the decision prompt lives in code. Tool-calling format 10, the request and answers are given only in engine.py and the runtime's schema. Trained tool calling 20, several questions answered together from one prefill. Structured output 5, the card claims probabilities without saying how to ask. Reasoning control 0, it has no thinking mode. Context 10, 262,144 tokens in the config, though the fine-tune trained at up to 16,384.

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

What a model and a developer need to talk to it. Speech, image and other models with no chat interface read the tool-calling and structured-output lines for what the model takes and returns (voices, timestamps, streaming, typed outputs), and the note says how.

- 0 to 20, the chat template. 20 for a template in the repository (Jinja, in the tokenizer config or its own file), 10 when the prompt format is only given as code you run.
- 0 to 25, the tool-calling format. 25 when the publisher documents how tools are declared and how calls come back, with the parser to use in a named runtime, 10 when only the template or a runtime's own recipe says it.
- 0 to 20, tool calling the model was trained for. 20 with several calls in one turn, 10 for one call at a time.
- 0 to 15, structured output. 15 when JSON or JSON Schema output is documented for the model with how to ask for it, 5 when the card claims it without saying how.
- 10, reasoning control. Thinking that can be turned off or budgeted per request, with where the reasoning text comes back.
- 0 to 10, context. 10 for 128K tokens or more trained, 5 for 32K to 127K. Context reached only by scaling the position encoding counts one band down.

## 4. Documentation and upkeep, 55 out of 100, up to 6.9 more on the total

Why it scored 55: Model card 5 of 20, a 295-byte card with the architecture and size and nothing else. Training data 5, the report describes the mix but gives no cut-off and the data isn't published. Last change 20, the card changed today. Responsiveness 20, Base Labs answered the one discussion the same day. Version policy 0, the weights were replaced in place on 20 October with no note. File health 5, the licence metadata contradicts itself with no LICENSE file.

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

- 0 to 20, the model card. Intended use, limitations, architecture, parameters, context length, languages and knowledge cut-off. 20 for all seven, 3 off for each that's missing.
- 0 to 20, training data. 20 for the data published or listed with sources and sizes, 15 for a data card that names the sources or the mix, the cut-off and the filtering, 10 for a general description with a cut-off, 5 without one, 0 for none.
- 0 to 20, the last change to the files, the template or the reference code. 20 within 30 days, 15 within 90, 10 within 180, 5 within a year, 0 older.
- 0 to 20, responsiveness. The publisher's replies to the 40 most recent questions and reports in the repository's discussions or issues. 20 for most, 10 for some, 5 for a few, 0 for none.
- 0 to 10, version policy. 10 for earlier versions kept in their own repositories with a dated note when one is replaced, 5 for earlier versions kept without notes or a first release, 0 when files were replaced in place without a note.
- 0 to 10, file health. 10 when config, tokenizer, template and card agree and no report of a broken file waits without a reply, 5 when one does, 0 when the files contradict the card.

## 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.

- Terms are read as MIT because the card's metadata names it, as settled for licences named only in metadata. With no licence text, a license: other tag, a same-day change from all rights reserved, and the index still calling it proprietary, a stricter reading would score terms 0 for unclear terms
- The Apache-2.0 base's licence text and notices aren't in the repository, which Apache-2.0 asks a redistributor to include
- Runtimes 0 assumes released runtimes don't apply the block-bidirectional mask from config.json's modeling_overrides. A looser reading that credits Transformers, which qev runs on, would give 5. Not run, so not checked
- Context 10 follows the config and the base. The fine-tune trained at up to 16,384 tokens, and whether decisions hold beyond that wasn't checked
- Which weights Baseten's Model API serves, the scored revision b9f2f2c or main, isn't stated
- The index excluded one private new-domain test for mammoth-1 on signs of training-data overlap, while the report says there was zero contamination. Recorded, not deducted
- The baselabs datasets mammoth-1-decision-index and elephant-1-decision-index need a person's approval and weren't requested

## Weaknesses

- Runs as trained only on OneJev's qev runtime at a pinned commit with the repository's patch
- No quantised build, so about 66 GB is the floor
- Licence named only in metadata, changed from proprietary on 21 October 2026 with no note and no LICENSE file
- The main branch's weights differ from the revision the Jev Decision Index scored
- A 295-byte card with no intended use, limitations, context or cut-off, and no safety results

## 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.

- Run `bash setup.sh <workdir>` in the pytorch/pytorch:2.8.0-cuda12.8-cudnn9-devel image on one CUDA GPU. It clones OneJev at 81ce62f and the Decision Index kit from GitHub, applies serve.patch and pip-installs them, so read it first
- Load through engine.py's DecisionModel (qev DecisionEngine). Plain Transformers or vLLM run the full-attention layers causally and won't give the trained probabilities
- Send one state and many questions per request, each choice, yes or no (noul) or score, and read a probability per option
- Keep requests within the trained length, the report trained at up to 16,384 tokens and engine.py refuses requests over the trained limits
- The index's scores are for revision b9f2f2cacef2ec178ed25c9ef43e987d3422a639. Pin that revision to get the scored weights, or main for the later ones
- Text only, the vision encoder is the base's, frozen and untrained for decisions

## What the review panel asked for

- versioned weight releases
- a model security contact
- tagged revisions with change notes
- a LICENSE file
- Put the request schema on the card
- Version the replaced weights
- A quantised build
- A LICENSE file

## 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.