Choose the axis
Identify the single experimental factor you want to vary while holding the rest of the setup as stable as practical.
Save Time & Tokens Testing Primitives.
Alpha 1 · Single-axis diagnostic
A test-axis witness for experimental design.
Don’t let AI expand experiments endlessly. A good harness can loop experimental runs, but can it tell you when yet another round of testing the same axis on a candidate primitive may not add useful information? Save those tokens for something more meaningful.
Experimental, advisory, and early-stage. It does not establish causation, statistical significance, or production readiness.
condition A
condition B
Change one axis. Observe whether the selected metric changes with it.
What it does
A controlled comparison, not a verdict.
Every result is shaped by the experiment you supply: the axis, runner, metric extractor, conditions, and sample design.
Provide the experiment
Supply a runner that executes each condition and a metric extractor that converts each run into a comparable result.
Compare the structure
Just a Marble runs controlled variations and checks whether the selected metric separates the axis conditions.
Read the advisory label
The output is an MDC classification: a compact indication of how much distinguishable information appeared under the supplied design.
How it works
One axis. Five explicit steps.
- 1 Define the test axis Name the controlled factor being changed.
- 2 Run controlled variations Execute each condition through the supplied runner.
- 3 Extract a metric Reduce each run to the feature you chose to observe.
- 4 Compare outcome structure Look for separation, overlap, or weak distinction.
- 5 Return a classification Report an advisory MDC label with supporting output.
Current labels
MDC classification
The labels describe the observed distinction produced by the supplied experiment and metric. They do not certify the experiment, prove an effect, or replace statistical analysis.
MARBLE
The selected metric does not meaningfully distinguish the tested axis conditions in the supplied runs.
LOW_MDC
The metric begins to show condition-sensitive structure, but the distinction remains limited or weak.
MODERATE_MDC
The outcome structure shows a clearer relationship to the tested axis under the current design.
HIGH_MDC
The supplied metric strongly distinguishes the tested conditions within the current experiment and sample design.
Interpretation boundary:
Classification quality depends on the runner, metric, conditions, sample count, and controls provided by the user.
Alpha 1 examples
The metric decides what the experiment can see.
Random walk · completion only
Same axis, non-sensitive metric
axis = step_size
metric = run_completed
result = MARBLE
Step size changes, but the extractor records only whether each run completed. Because completion does not distinguish the conditions, the output is MARBLE.
Random walk · maximum excursion
Same axis, axis-sensitive metric
axis = step_size
metric = max_abs_position
result = LOW_MDC
Maximum absolute position can respond to step size. The metric begins to separate the conditions, producing an advisory LOW_MDC classification.
FAQ
What Just a Marble is, and what it is not.
These answers keep the Alpha 1 boundary explicit: Just a Marble is an experimental diagnostic software surface, not an automated authority.
What kind of software is Just a Marble?
Just a Marble is experimental axis-distinction diagnostic software. It can be used as a lightweight screening step before expanding a simulation, benchmark, or experiment, and as a local sensitivity probe for one candidate axis. It is not a full sensitivity-analysis platform, benchmark suite, or production middleware system.
Is Just a Marble a sensitivity-analysis tool?
It is sensitivity-analysis-adjacent, but narrower. Just a Marble tests whether changing one supplied experimental axis produces distinguishable outcome structure under the metric chosen by the user. It does not perform full global sensitivity analysis, estimate causal effects, prove statistical significance, or replace established statistical methods.
Is Just a Marble a screening-design tool?
It can support screening-style use when someone wants to determine whether one axis appears informative enough to justify broader testing. Alpha 1 does not provide a complete screening-design suite, multi-factor experimental design, or automated variable selection.
Is Just a Marble a benchmark?
No. It can be placed around benchmark-like or simulation workflows, but it is not the benchmark itself. The user supplies the experiment runner, changed axis, metric extractor, conditions, and sample design. Just a Marble compares the supplied outcomes and returns an advisory classification.
Is Just a Marble middleware?
Not currently. Alpha 1 is a standalone, single-axis Python reference implementation. It may have future middleware potential, but middleware would require stable schemas, conformance fixtures, multi-axis behavior, compatibility boundaries, integration design, and defined support commitments.
What is an axis-distinction tool?
An axis-distinction tool asks whether changing one experimental input produces meaningfully distinguishable results under a selected measurement. It does not determine whether the axis caused the result. It evaluates whether the supplied experiment and metric reveal useful distinction between conditions.
What does substrate governance mean here?
Substrate governance means evaluating whether an experimental axis appears useful enough to keep testing, refine, or examine more closely before consuming additional compute, time, or experimental complexity. The current MDC result is advisory. It does not automatically issue commands such as stop, retest, expand, pass, or fail.
Does Just a Marble prove statistical significance?
No. Its classifications depend on the supplied runner, metric, conditions, sample count, and sample design. They are diagnostic observations, not proof of causation, statistical significance, production readiness, or general validity.
What is MDC?
MDC is the advisory distinction classification returned by Just a Marble. Alpha 1 currently uses MARBLE, LOW_MDC, MODERATE_MDC, and HIGH_MDC. These describe the observed distinction under the supplied test configuration. They are not universal judgments about the underlying axis.
Who is Just a Marble for?
Potential users include simulation developers, benchmark designers, experimental software researchers, game-system researchers, developers comparing configurable model behavior, and teams deciding whether an experimental variable deserves broader testing.
Can Just a Marble reduce unnecessary testing?
It may help identify when a selected axis and metric produce little useful distinction, which can inform whether broader testing is worth considering. It does not automatically determine that testing should stop, and it does not replace expert judgment or formal experimental design.
What does Alpha 1 support?
Alpha 1 provides an independently runnable, single-axis Python reference implementation with advisory MDC classifications and standalone tests. It does not yet provide a CLI, multi-axis mode, package release, stable external API, stable schema, production certification, or compatibility guarantees.
Current release state
Alpha 1
A compact Python reference implementation for independently running a single-axis diagnostic experiment.
- ImplementationSingle-axis Python reference
- Test status26 standalone tests passing
- RoleAdvisory diagnostic tool
- Statistical claimNot statistical proof
- CertificationNot production certified
- CLINot yet available
- Multi-axis modeNot yet available
- Package releaseNot yet available
License summary
Source-available, not open source.
This summary is for orientation only. The full license text controls if any wording differs.
Request commercial licensingAllowed without a commercial license
- Personal use
- Teaching and internal noncommercial academic research at accredited universities and colleges
- Genuinely free, noncommercial games or tools created and publicly shared by individuals
- Private modification by otherwise permitted users
- Publication of research results
Requires written permission or a paid license
- Business, nonprofit, government, or contractor use
- Sponsored work or client work
- Revenue-generating use
- Redistribution of JAM source code or modified forks
Built using Just a Marble by Stephen Montgomery
Contact