Question before tool
The project begins with a question that determines which methods and tools are appropriate.
Research Standards
Serious investigation before polished presentation.
A compelling demo can still rest on a weak question, contaminated evaluation, or work the student cannot explain. Pristone's research standards put method, evidence, ownership, and honest limitations first.
Free to ask · replies within 2 hours
The Pristone Method
One capability loop across every track.
- Diagnose
- Prioritize
- Build
- Test transfer
- Measure
- Re-prioritize
Student research at Pristone is an investigation the student can explain and defend. It may produce code, a model, a report, or a presentation, but the artifact does not substitute for a sound question and method.
Written and operationally reviewed by Edward Mabonga, founder of Pristone Academy.
Minimum standard
What the work must make visible
Baseline before improvement
An improvement claim needs a defensible point of comparison.
Separation and leakage control
Training, validation, and testing remain meaningfully separated where the method requires it.
Repeatability
Another person should be able to understand the data, procedure, assumptions, and major decisions.
Failure analysis
Unexpected and negative results are examined rather than hidden.
Source integrity
Claims, datasets, code, and borrowed ideas are attributed accurately.
Student authorship
The student must be able to explain and defend every major part of the work.
Honest reporting
Limitations are part of the result, and publication or competition outcomes are never promised.
AI and machine learning
Additional technical checks
Data provenance
The source, license, collection process, and known limitations of the data are documented.
Evaluation contract
Metrics are connected to the research question and chosen before results are interpreted.
Leakage review
Features, preprocessing, model selection, and repeated testing are checked for information leakage.
Baseline comparison
Complexity is justified by comparison with a simpler deterministic, statistical, or heuristic approach.
Reproducible workflow
The environment, random behavior, transformations, and major experimental settings are recorded.
Human authorship
AI-assisted code or writing is disclosed where appropriate, verified, and understood by the student.
Claims
What a research program should not promise
Guaranteed publication
Publication depends on external standards, review, venue fit, and factors no mentorship program controls.
Guaranteed admissions value
A project may demonstrate initiative, but no single activity guarantees an admission result.
Novelty without review
A project is not called novel merely because the student has not seen the idea before.
Results without limitations
A strong result still requires a clear account of uncertainty, scope, and where it may fail.
Continue through the Pristone system
Begin with the question and the readiness evidence.
A research diagnostic tests fit, preparation, ownership, and whether the proposed question can support serious work within the available time.
Founder-led by Edward Mabonga (MIT EECS) · no guaranteed scores, admissions, publications, promotions, or jobs