Our Methodology
The methodology MaishaThabiti uses to separate entities, evidence, dated knowledge and decision logic.
Our methodology separates subjects, evidence, structured facts and decision logic so that useful information does not become indistinguishable from opinion.
The four-part knowledge model
Entities
The real-world things a user may need to understand or compare: opportunities, careers, skills, courses, qualifications, financial products, services, institutions and other defined subjects.
Evidence
The source material that supports or challenges a claim: official documents, institutional pages, research, datasets, standards, credible reporting and other appropriate sources.
Knowledge Records
Small, specific, dated facts or findings derived from evidence, with enough context to understand what the claim means, when it applies and what its limitations are.
Decisions
Rules or decision structures that combine explicit user inputs with suitable Knowledge Records to explain options, trade-offs and next steps.
1. Define the subject clearly
We first determine what kind of thing is being described, its practical scope and the identity that distinguishes it from similar subjects. Different labels can refer to the same thing, and similar labels can refer to different things. Clear identity reduces duplication and improves relationships across the knowledge base.
2. Match the source to the claim
Source quality is contextual. Government and recognised authorities are usually the strongest sources for laws, official requirements and public programmes. A university is authoritative about its own course offering. Independent research may be better for outcomes or market evidence. Community experience may be useful for lived experience, but it should not replace an official rule.
3. Separate evidence from interpretation
A source may establish a fact without establishing every conclusion a user might draw from it. We therefore distinguish what the source directly supports from what is inferred, estimated, compared or recommended.
4. Make facts atomic and time-aware
Where practical, important knowledge is stored as small, specific records rather than one large undifferentiated narrative. That makes it easier to attach a source, date, geographic scope, limitation, confidence level or review status to the fact that actually needs those attributes.
5. Preserve uncertainty and disagreement
Incomplete evidence does not automatically make a subject unusable. A useful record can remain incomplete while targeted research continues. When credible sources conflict, the disagreement should be investigated or shown; it should not be erased by averaging incompatible claims or choosing whichever answer is most convenient.
6. Keep recommendations traceable
Where MaishaThabiti provides decision support, the reasoning should be connected to explicit user answers and suitable, dated knowledge. A recommendation should be explainable in terms of the factors that mattered, the evidence used, the trade-offs identified and the limitations that remain.
7. Review, correct and supersede
The system is designed to evolve. Stronger evidence can replace weaker evidence. A new rule can supersede an old rule without erasing the historical record. Editorial review should improve the knowledge base rather than preserve an earlier conclusion merely because it was published first.
8. Use automation without surrendering authority
Automation and AI can assist discovery, extraction, classification, comparison, drafting and quality checks. They do not become the source of truth simply because they work quickly. Provenance, validation, review and evidence remain the basis of authority.
A living knowledge base
The aim is not perfect completeness on day one. The aim is a system that can become more complete without filling gaps with fabricated facts or losing the provenance of what is already known.
Last updated: 6 September 2026.