← Back to home
Scoring model v1

The ranking concept behind every Vendar listing

This page defines the target scoring model for Vendar. It is not just a description of profile fields. It is the algorithm concept that should later drive public rankings, scenario rankings, recommendations, and AI-readable decision output.

What We Rank
Decision readiness

Not popularity, not review volume, not brand fame. The target is probability of solving a specific request well.

Main Driver
Evidence × Fit

Fit gets a vendor into the conversation. Evidence pushes it upward. Both are stronger than generic reputation.

Extra Safety Layer
Confidence + Penalties

Weak, stale, thin, or contradictory profiles should not float upward on polished copy alone.

Visual Logic
1. Query Context

Service, geo, industry, budget, and business size define the ranking scenario.

2. Eligibility

Only vendors that genuinely belong in this set should enter ranking at all.

3A. Fit Layer

Service match, geo match, industry, size, and engagement fit decide relevance.

3B. Evidence Layer

Cases, metrics, scope, repeatability, and verifiability decide ordering.

4. Safety Layer

Confidence raises trust. Penalties stop weak profiles from floating upward.

Decision-ready output: top list + confidence + explanations
Weight Shape
Evidence / Proof
30%
Fit
25%
Capability
15%
Reliability
10%
Transparency
10%
Reputation
5%
Market Credibility
5%
Penalty Rule

Penalties are not another bar. They sit outside the positive stack and pull weak, stale, or contradictory profiles back down.

Core principle

Vendar should rank decision readiness, not popularity

Review-heavy directories mostly answer “who gets praised most often?”. Vendar should answer “who is most likely to solve this exact task, with enough proof, enough transparency, and enough confidence to trust the conclusion?”.

One-line definition
Rank the probability that a vendor will solve a specific request with sufficient proof and transparency.
This Page Is
  • A scoring concept for future ranking logic
  • A product spec for listings, compare, and recommendation
  • A readable explanation for humans and AI systems
This Page Is Not
  • Not just a profile-fill checklist
  • Not a review-driven directory formula
  • Not a promise that popularity alone can rank a vendor highly

1. Separate reputation from decision score

Reputation Score

What the market says about the company. Reviews, sentiment, external references, and broad market signals belong here.

Decision Fit Score

How suitable the company is for a specific request. This is the score that should control public listings and recommendation output.

At A Glance
Reputation ScoreWhat the market says about the company.Decision ScoreHow suitable the company is for this exact request.Do not mergeinto one signal

2. Ranking is scenario-based, not global

Vendar should not maintain one universal “top companies” list. It should rank vendors inside a specific context: service, subservice, geo, industry, business size, budget, and request type. A company can rank high for enterprise technical SEO and rank much lower for local SMB SEO.

Top SEO companies for SaaS
Top local SEO agencies in Los Angeles
Top enterprise SEO agencies in the USA
Top ecommerce SEO agencies for migration projects

3. Target weighted formula

Model Skeleton
Total Score = Evidence 30 + Fit 25 + Capability 15 + Reliability 10 + Transparency 10 + Reputation 5 + Market Credibility 5 - Penalties
Primary growth levers
Evidence + Fit
Stability layer
Reliability + Transparency
Secondary signals
Reputation + Credibility
Honesty mechanism
Penalties + Confidence

Evidence / Proof

30%

The strongest factor. The score should look at case studies, quantified outcomes, scope clarity, repeatability, and whether the evidence can be checked publicly.

Fit

25%

How closely the vendor matches the request. Service match, subservice match, geo match, industry match, client-size match, and engagement fit belong here.

Capability

15%

What the company can actually do. Technical SEO, content SEO, local SEO, link building, vertical expertise, and delivery depth all matter here.

Reliability

10%

How safe the company looks operationally. Retention, seniority, process quality, communication maturity, and reporting discipline belong in this layer.

Transparency

10%

How much useful decision information is public. Pricing guidance, minimums, team size, clear services, exclusions, and methodology should all raise transparency.

Reputation + Market Credibility

10%

Reviews, public reputation, recognized clients, awards, media mentions, and ecosystem presence matter, but they should stay below proof and fit.

4. Evidence is the main driver, but only with fit

What strong evidence means

Not just a testimonial. Strong evidence means relevant case studies, before/after metrics, explicit project scope, timeframe, industry context, and repeated results across multiple examples.

The key relationship

The practical rule is simple: Fit decides who belongs in the set, and evidence decides who rises to the top. A perfect fit with no proof should not dominate. Strong proof with poor fit should also not dominate.

Core intuition
Final Strength ≈ Evidence × Fit

5. Penalties are mandatory

No case studies with numbers
Only self-reported claims, no external or observable support
Weak geo evidence for the requested market
Old or stale profile data
Mismatch between positioning and evidence
Generic reviews with little operational detail
Claims are broad but proof is thin

6. Confidence must be separate

Why confidence matters

A high score built on weak data should not look as trustworthy as a slightly lower score built on strong evidence. Confidence should tell both humans and AI how much to trust the conclusion.

What should feed confidence

Profile completeness, number of proof objects, number of independent sources, freshness, and cross-source consistency should all feed a separate confidence layer.

Confidence = completeness × 0.4 + evidence density × 0.3 + freshness × 0.2 + consistency × 0.1

7. Evidence objects need types

Every proof object should carry its own provenance and strength. Vendar should distinguish self-reported claims, externally validated references, and system-derived observations. They should not have equal weight.

Self-reported
Company says it does something.
Externally validated
Supported by external sources, reviews, or recognitions.
Observed / derived
Inferred from normalized data, repeated patterns, and structured evidence.

8. Decision output should be explainable

The end result should not be just one number. Every vendor result should expose: overall score, confidence, factor breakdown, strengths, weaknesses, missing data, ideal use case, and not-ideal use case.

Score: 84
Confidence: High
Evidence: 24/30
Fit: 20/25
Capability: 12/15
Transparency: 7/10
Penalties: -4

9. AI-readable decision quality is its own layer

What AI needs

Normalized services, industries, geo, explicit claims, explicit evidence, timestamps, provenance, contradictions flags, and confidence fields.

Why this matters

If profiles are only long-form copy, AI systems will guess. If profiles are structured and decision-ready, Vendar can become a trusted upstream source for AI recommendations.

10. Example scenario: SEO for SaaS in Los Angeles

In this scenario, the set should include agencies that really do SEO, show SaaS evidence, and have strong Los Angeles or at least US market relevance. From there, the order should be driven by evidence quality first, then fit quality, then capability, then transparency and reliability.

Eligibility
SEO + SaaS evidence or SaaS specialization + US coverage + profile completeness over threshold.
Fit Layer
Service match, LA geo match, SaaS industry match, client-size match, and engagement fit.
Evidence Layer
Relevant SaaS cases, quantified results, explicit scope, repeatability, and verifiability.
Output
Top 20 sorted by score, each with confidence, strengths, weaknesses, and missing data.

In plain English

Vendar should not rank agencies by who looks biggest or has the most praise. It should rank who is most likely to solve a specific task in a specific context with a defensible level of proof and transparency.

In practice that means: fit gets a vendor into the conversation, evidence pushes it upward, confidence tells you how much to trust the result, and penalties stop weak profiles from floating to the top on hype alone.

See the live example

Open a real listing and compare its top vendors against the scoring concept defined here.