Trialmark

Model acceptance tests for language models. Test models on your own work, decide in advance what result is good enough, and keep a record that anyone can rerun and check.

Choosing a language model is a trade between accuracy, cost and speed. Trialmark makes that choice systematic:

1. Your own test set. Cases from your own work, each with a reference answer written by a person and tagged by kind, such as "factual" or "multi-part". The set is hashed, so a later run can prove it used exactly the same cases.
2. A pass mark set before the run. For example: "choose the cheapest model within 3 points of the best on correctness, as long as at least 95% of its answers are grounded". The rule is stored with its hash and the time it was set, so it cannot be quietly changed after the results are in.
3. Repeats, with the range shown. Models do not give the same answer every time. Each model runs several times, and the report shows the lowest and highest result so that a difference inside the range is read as no difference.
4. A report by kind of case. A small model may be as good as a large one on single facts and worse on multi-part questions. The breakdown by tag shows where.
5. A run record and a release gate. Every run writes one JSON file with the test set's hash, the models and their settings, the judge, the pass mark and every result. trialmark gate exits with 0 or 1, so a pipeline can refuse a model change that fails the pass mark.

Trialmark runs on your own machines. Your test sets, reference answers and records stay with you.

Where it comes from

Trialmark was spun out of the model benchmark in Forge, a governed layer through which organisations delegate work to AI. That benchmark compares models on a fixed set of business-document questions, through Forge's own answer path, with three repeats and accuracy, cost and speed side by side. Trialmark makes the same kind of test available to any developer or team, with or without Forge.

Status

Trialmark is at an early stage.

Until the model runners arrive, you connect a model with the built-in python runner: a function of your own that calls the model and returns its answer.

Install

pip install trialmark

Python 3.10 or later. The core has no dependencies.

Try it

The examples folder holds a five-question test set and a configuration that compares three stand-in models, so a run needs no API key. Run it from the repository's root folder:

trialmark validate examples/testset.json
trialmark run examples/config.json --by-tag
trialmark report examples/config-run-record.json --format markdown
trialmark gate examples/config-run-record.json --model perfect-model
trialmark plugins

examples/run-record.json is an illustrative record of three made-up models whose scores are invented to show the format. They are not measurements of any real model.

Test your own model

Write a function that calls your model:

# my_model.py
from trialmark import RunOutput

def answer(model_id, case_input, settings):
    reply = call_your_model(model_id, case_input)      # your code
    return RunOutput(output=reply.text, cost_usd=reply.cost,
                     input_tokens=reply.input_tokens, output_tokens=reply.output_tokens)

Then name it in a configuration, with the pass mark fixed before the run:

{
  "test_set": "my-testset.csv",
  "pass_mark": { "metric": "correct", "tolerance_points": 3, "choose": "cheapest" },
  "repeats": 3,
  "concurrency": 4,
  "models": [
    { "id": "large-model", "runner": "python", "settings": { "callable": "my_model.py:answer" } },
    { "id": "small-model", "runner": "python", "settings": { "callable": "my_model.py:answer" } }
  ],
  "scorers": ["exact-match"],
  "output": "run-record.json"
}

trialmark run config.json --by-tag

Exact match suits short answers. Free-text answers need a judge, which is planned.

Use it from Python

from trialmark import ModelEntry, PassMark, load_with, make_scorer, render, run_trial

record = run_trial(
    load_with("my-testset.csv"),
    PassMark(metric="correct", tolerance_points=3, floors={"grounded": 0.95}, choose="cheapest"),
    [ModelEntry(id="model-a", runner="mine"), ModelEntry(id="model-b", runner="mine")],
    runners={"mine": MyRunner()},          # any object with run(model, case) -> RunOutput
    scorers=[make_scorer("exact-match"), MyGroundednessScorer()],
    repeats=3,
    concurrency=4,
)
record.save("run-record.json")
print(render(record, "text", by_tag=True))

Customising Trialmark

There are four customisation points: runners (call a model), scorers (turn an answer into metrics), loaders (read a test set from a file format) and report writers. ARCHITECTURE.md gives each interface and explains what stays fixed in the core and why.

You do not need to fork Trialmark to customise it. Install it from PyPI and add plugins:

• For a one-off model call, use the python runner, as above.
• From Python, pass your own runner and scorer objects to run_trial.
• To share a plugin, publish it as its own package.

Writing a plugin

A plugin package declares an entry point in one of the groups trialmark.runners, trialmark.scorers, trialmark.loaders or trialmark.reports:

# trialmark_acme/__init__.py
from trialmark import RunOutput

class AcmeRunner:
    def __init__(self, endpoint="https://api.acme.example"):
        self.endpoint = endpoint

    def run(self, model, case):
        reply = call_acme(self.endpoint, model.id, case.input, **model.settings)
        return RunOutput(output=reply.text, cost_usd=reply.cost)

# pyproject.toml of the plugin package
[project.entry-points."trialmark.runners"]
acme = "trialmark_acme:AcmeRunner"

Once it is installed, trialmark plugins lists it, and a configuration uses it with "runner": "acme", with options under "runners": {"acme": {"endpoint": "..."}}. A plugin cannot take the name of a built-in plugin.

Formats

Test set

{
  "name": "contract-questions",
  "version": "2026-11-05",
  "cases": [
    { "id": "q1", "input": "What is the notice period?", "reference": "Sixty days' written notice.", "tags": ["factual"] }
  ]
}

The hash covers the cases (their ids, inputs, references and tags), not the name or version.

Pass mark

Metrics are whatever your scorer produces, each between 0 and 1. A model's score on a metric is the mean within each repeat; the report shows the mean of those and their range.

Licence

Apache-2.0. See LICENSE.