Introduction
ZORAA is being developed as a research and decision-support system for understanding startups. Its purpose is to organise evidence, make comparisons more consistent and reveal the questions that deserve further investigation. This public edition describes the approach without claiming finished models or established performance.
Why startup evaluation is difficult
Startups operate with limited histories, changing strategies and uneven disclosure. A strong pitch can coexist with incomplete evidence. Two businesses at different stages may report similar numbers that mean very different things. Research must account for those differences rather than assume that one ranking explains every company.
Research philosophy
Evidence should be traceable. Assumptions should be identified. Unknowns should remain visible. The system should separate company-reported statements, externally supported observations and analytical interpretations. None should silently stand in for another.
Evaluation categories
The proposed framework considers founder and team, market, product, technology, business model, financial health, traction, competition, customers, growth, operations, regulatory environment, funding, scalability, risk and industry conditions.
Relevant categories and signals will vary by stage, sector and available information. Missing information should not automatically be interpreted as poor business performance.
Data architecture
The intended architecture begins with company-submitted material and information from permitted public, regulatory, research and licensed sources. Each record should retain its origin, collection date, applicable permissions and relevant verification status.
Listing a source category does not mean ZORAA currently has access to a particular provider. Public availability alone does not establish permission for every type of reuse.
Signal extraction
Information is mapped to defined observations, such as reported revenue, customer concentration or relevant team experience. Extraction should preserve context, units, time periods and source attribution. Conflicting statements should be flagged instead of quietly averaged away.
Scoring framework
Any eventual score should summarise a defined analytical framework, with visible category drivers and limitations. Stage and sector context should guide comparisons. Exact weights and proprietary implementation details are not published here.
The sample 78/100 on this website is a fictional interface example. It is not a validated assessment, investment recommendation, probability of success or promised return.
Confidence modelling
Confidence is intended to communicate how well an interpretation is supported. Relevant factors include source quality, agreement between sources, freshness, corroboration and coverage. A high-confidence observation can still be wrong.
No confidence model or threshold has been validated for public use through this website. The “High” label in the sample is illustrative.
Data completeness
Completeness asks how much of the relevant information is available. It should be measured against an explicitly defined set of fields appropriate to the research scope, rather than every possible fact about a company.
Completeness is not a quality score. Many populated fields can originate from one unverified source. The sample 91% is fictional and has no production denominator.
Risk modelling
Risk indicators should point to specific concerns: customer concentration, funding dependence, inconsistent disclosures, operating constraints or missing evidence. Severity and uncertainty need separate treatment. A risk flag is a prompt for investigation, not a finding of wrongdoing.
Explainability
An output should identify the observations that influenced it, the sources that support those observations and the limits of the interpretation. Useful explanations should include important counterevidence and what additional information could change the result.
The goal is an auditable research rationale. An explanation generated by AI is not proof that the underlying claim is correct.
Source verification
Identifying a source, checking a document and independently confirming its contents are different actions. The intended system should label these distinctions explicitly. Information from a company website is a primary statement about the company, but is not automatically independently verified.
Model limitations
AI can misread documents, make unsupported inferences and reproduce biases. Data can be incomplete, stale or inaccurate. Historical patterns may fail when business conditions change. Startup outcomes are uncertain and cannot be guaranteed by a score.
ZORAA has not published a performance benchmark, accuracy guarantee or validated outcome prediction. Such claims would require documented testing.
Human review
Human reviewers should examine consequential inconsistencies, contested facts and high-impact interpretations. Founders should have a route to request corrections. The scope and availability of product review processes remain part of development.
The current website accepts questions and correction requests through its contact form. It does not provide live research review or regulated financial services.
Continuous improvement
Changes to the framework should be recorded and evaluated against consistent criteria. Future versions should make it possible to distinguish changes in a company’s evidence from changes in the research method.
Feedback from founders, researchers and investors will help refine the questions. Join early access or contact ZORAA to contribute.