Skip to main content
DETERMINISTIC SCORING ENGINE · MODEL V1.0.0

Scoring Methodology & Mathematical Rules

How is the DevReality score calculated?

Scores are calculated by fixed rules in code, not by AI. They measure public presentation quality, not programming ability. Every point earned or deducted maps directly to observable public repository signals or PageSpeed Web Vitals.

What the Score Measures
  • •Documentation completeness: README presence, quickstart guides, byte depth, and visual UI screenshots.
  • •Repository hygiene: Clear project descriptions, topic tags, license files, and non-fork original ratio.
  • •Portfolio quality: Core Web Vitals (FCP, LCP, CLS), responsive mobile viewport, and accessibility landmarks.
  • •Discoverability: Profile pinned projects, clear naming conventions, and working live deployment URLs.
What the Score Does NOT Measure
  • •Real programming skill or system design: We never execute private code or assess algorithmic problem-solving.
  • •Interview readiness or employment worth: Presentation scores do not predict job performance or engineering impact.
  • •Private enterprise contributions: Developers with confidential enterprise repos naturally show lower public activity.
  • •Team collaboration or soft skills: Public metrics reflect showcase presentation only.

GitHub Presentation Categories (100 Points Total)

Auto-generated from src/config/scoring.ts. Updated automatically with every score engine build.

CategoryMax WeightWhat Is Checked & Evaluated
Profile Completeness15 ptsBio presence, profile README existence, location, contact/website links, and avatar configuration.
Repository Presentation20 ptsNon-fork original project ratio, clear repository descriptions, topic tags, and tech stack clarity.
Documentation & Readability25 ptsREADME coverage across repositories, byte-size depth, installation commands, quickstart guides, and screenshots.
Maintenance Signals15 ptsRecency of latest repository push, active non-archived projects, and sustained commit consistency.
Technology Consistency15 ptsCohesive primary languages, modern framework usage, and non-trivial codebase sizing (>2KB code bytes).
Discoverability & Pinned Repos10 ptsShowcase pinned repositories on public profile, descriptive repository names, and live homepage links.

Portfolio Audit Categories (100 Points Total)

Evaluated using server-side DOM parsing and Google PageSpeed Insights API. Auto-generated from source code config.

CategoryMax WeightWhat Is Checked & Evaluated
Performance & Web Vitals20 ptsMobile & desktop PageSpeed metrics: First Contentful Paint (FCP), Largest Contentful Paint (LCP), and layout stability (CLS).
Accessibility (WCAG 2.2)15 ptsSemantic HTML landmarks (<main>, <nav>, <header>), image alt attributes, contrast ratios, and accessible form labels.
Search Engine Optimization15 ptsDescriptive <title> tags, meta descriptions, OpenGraph social preview tags, viewport tag, and valid heading hierarchy.
Security & Best Practices10 ptsHTTPS enforcement, modern HTML5 doctype, secure external link rel attributes (noopener/noreferrer), and clean DOM.
Project Showcase & Demos25 ptsRecognizable project showcase section, working live demo URLs, repository source code links, and narrative depth.
Mobile & Responsive Experience15 ptsMobile viewport scaling, responsive layouts, media query structures, and touch-target friendly navigation.
Missing Category Normalization Algorithm

If a category cannot be inspected—such as when a portfolio URL is omitted, or when an external website blocks automated PageSpeed audit bots—DEV//REALITY does not penalize you with an arbitrary zero. Instead, the final score dynamically normalizes across all verified categories using the mathematical formula below:

Normalized Score = (Earned Points / Sum of Max Available Category Points) × 100

Example: If only GitHub presentation is audited (100 max points available) and a candidate scores 82 points, their overall score is 82/100 (not diluted by the missing portfolio).

Confidence Levels

Every scored observation and recommendation is tagged with a transparent confidence level:

HIGH CONFIDENCE

Directly Measured Data

Derived from definitive API fields or HTTP headers. Example: Repository count, README byte length, or PageSpeed First Contentful Paint.

MEDIUM CONFIDENCE

Structural Pattern Matching

Derived from passive HTML/DOM inspections. Example: Presence of a <nav> landmark or matching regex patterns for live demo links.

LOW CONFIDENCE

Incomplete / Truncated Data

Derived from edge cases where an endpoint timed out or returned truncated metadata. Stated clearly so recruiters avoid unfair bias.

Score Bands & Qualitative Labels

RangeLabelDescription
90 – 100Strong PresentationOutstanding public technical presentation with clear documentation and attention to detail.
75 – 89Solid FoundationGood developer presence with a few areas that would benefit from polishing.
60 – 74Needs RefinementCore projects and profile are present, but presentation weaknesses dilute first impressions.
40 – 59Major Improvements NeededNoticeable documentation, organization, or portfolio gaps that hinder recruiter discovery.
0 – 39Time for a RebuildPublic technical presence is sparse, outdated, or incomplete.
What the AI Does and Does NOT Do

What Gemini AI Does:

  • • Writes an entertaining roast based strictly on verified evidence IDs.
  • • Simulates how a hiring manager scans your profile in 30 seconds.
  • • Structures recommendations into an actionable redemption roadmap.

What Gemini AI NEVER Does:

  • • Never calculates, modifies, or influences numerical scores.
  • • Never invents repositories, features, or metrics outside verified evidence.
  • • Never makes subjective character or personal trait judgments.

Want to see realistic reports in action?

Explore our sample reports or learn how the data collection pipeline functions.