Website security · Raksha guide

What Is a Website Security Score? How It Is Calculated

A website security score is a summary of the findings included in a particular assessment. Raksha starts at 100 and subtracts severity penalties adjusted by confidence. It is not a probability of compromise, a certification or a measure of everything the scanner could not see.

By Raksha · Reviewed 16 November 2026

Raksha’s calculation

For the evidence-reviewed security findings in a run, multiply each severity penalty by its confidence weight, add the deductions and subtract that total from 100. Round the result to the nearest integer and clamp it between 0 and 100.

Score = round(100 − sum of severity penalty × confidence weight), bounded to 0–100.

Base deductions per finding
SeverityPenalty before confidence weighting
Critical22 points
High11 points
Medium5 points
Low2 points
Informational0.5 points
Confidence multipliers
ConfidenceMultiplierMeaning
Confirmed1.00Full deduction; observed evidence must support the finding.
Likely0.70Reduced deduction for a less certain signal.
Possible0.45Reduced deduction for a tentative observation.

Worked calculation: 84 out of 100

This is an illustrative calculation, not a live result or Raksha’s own current score. Suppose a reviewed run includes one confirmed high finding, one likely medium finding and one confirmed low finding.

  1. High: 11 × 1.00 = 11 points.
  2. Medium: 5 × 0.70 = 3.5 points.
  3. Low: 2 × 1.00 = 2 points.
  4. Total deduction: 16.5. The unrounded score is 83.5; the displayed score is 84.

The exact displayed result here is 84/100. Changing confidence or the reviewed finding set changes the result. A zero means the deductions reached the score floor—not that a breach has been proven.

Why the score and the finding list must agree

Severity counts and the security score derive from the same reviewed security set. Search, AEO and GEO observations stay in Web Health and do not belong in the security ledger. A limited free view can show fewer finding details than the full report; use coverage and report-access limits to understand what you are seeing.

Scores across different tools are not interchangeable. Tools use different checks, deductions and evidence thresholds. Even two Raksha runs may not be directly comparable if targets, scope or completed checks changed.

Security score vs AI visibility score

AI visibility is separate: it combines citation coverage, answer-engine optimisation, generative-engine optimisation, SEO, entity coverage and content authority. A missing llms.txt file can affect web-health observations but is not a confirmed security vulnerability.

Publishing llms.txt does not guarantee that an AI service reads or cites your site. Its format is a proposal, not a universally supported standard.

How to use a score responsibly

  • Prioritise observed impact and confidence, not just the numerical deduction.
  • Read the evidence and coverage before accepting or dismissing an issue.
  • Re-check after a fix with comparable scope and check completion.
  • Do not infer that 100/100 proves safe authentication, database rules or business logic.
Read the evidence-grading methodology

Start with a site you are authorised to test

A free passive check needs no card. Sign-in, email confirmation, ownership rules and usage limits apply. It is not an unrestricted penetration test.