Scoring Criteria

This document defines how the Integrity Index scoring system works. It is written for two audiences: members of the public who want to understand how a score is produced, and the AI agent responsible for recurring profile reviews, evidence classification and score recalculation.

The Integrity Index is an evidence-based public-interest accountability tool. It uses publicly available, source-linked information to assess the public integrity risk associated with South African public figures. The score is not a declaration of guilt, a court finding, or a statement about a person’s private character. It is a structured risk score based on documented public evidence.

The Integrity Index uses an Integrity Risk Score from 0 to 100.

Scoring direction: A higher score means a higher level of publicly documented integrity risk. A lower score means a lower level of publicly documented integrity risk.

Colour logic: The website visual bar must run from green at the low-risk end to red at the high-risk end: green at 0, yellow/orange in the middle, red at 100. A triangle or marker should show the person’s exact score on the bar.

A score may only be created or changed where there is source-linked evidence. The agent must not score rumours, anonymous social-media claims, opinion without evidence, or unsupported political attacks.

Each evidence item must be recorded with the following fields:

• Public figure name
• Date of evidence or publication
• Evidence category
• Matter or incident name
• Short factual description
• Source title
• Source URL
• Source type and reliability tier
• Status: alleged, investigated, charged, convicted, finding issued, or resolved in the latest review
• Score contribution
• Reviewer note explaining why the score was applied

The score is not a simple article-counting system. It is a severity-based decision-tree model. The agent must identify each separate matter or incident first, then apply the highest relevant evidence category to that matter.

The decision tree is:

• Check whether there is a relevant criminal conviction. If yes, assign 100 and classify the profile as Severe Risk.
• If there is no relevant criminal conviction, check whether there are formal criminal charges. Score each separate charged matter at 50, subject to no double counting and the total score cap of 100.
• For matters without formal charges, check for adverse official findings or being named adversely in court judgments, commission reports or similar official documents.
• For matters without an adverse official finding, check for official investigations by bodies such as the NPA, Hawks, SIU, Public Protector or Parliament Ethics Committee.
• Check for ethics breaches, disclosure failures and conflict-of-interest findings.
• Check for repeated credible media allegations or serious public-interest concerns supported by reputable sources.
• Check for serious conflict-of-interest concerns, procurement concerns, family-linked benefits or undeclared interests that do not already fall under a higher category.
• Add only separate matters. Do not count the same incident more than once.
• Cap the final score at 100.
• Recalculate the score during every 30-day review cycle using the current evidence record.

The following categories are used for scoring. The “cleared, withdrawn, disproven, corrected or dismissed” category is not a scoring category. Cleared or dismissed matters should be noted in the profile where relevant, but the score is recalculated every 30 days from the active evidence record rather than using negative points or subtraction.

No double counting is one of the most important rules in the methodology. The score is matter-based, not source-based.

The agent must first group evidence into separate matters or incidents. Each matter receives only the highest applicable score category. Other evidence relating to the same matter is recorded as supporting evidence but does not add additional points.

Example: If a person is formally charged with fraud relating to a tender, that matter scores 50. A Hawks statement, an SIU investigation, a News24 article and a Daily Maverick article about the same fraud case are supporting evidence for that same matter. They do not add 25 + 15 + 15 on top of the 50.

Separate unrelated matters may be added together. Example: a pending fraud charge relating to one tender scores 50. A separate and unrelated parliamentary disclosure breach may score 20. The total score becomes 70, subject to the 100 cap.

A relevant criminal conviction is the strongest evidence category and results in an automatic score of 100. This is because a court of law has found the person guilty of an integrity-related offence.

Relevant offences include, but are not limited to:

• Corruption
• Fraud
• Theft
• Bribery
• Dishonesty-related offences
• Money laundering
• Racketeering or organised crime linked to public office or public resources
• Procurement corruption
• Abuse of office
• Public-finance or tender-related criminality

The agent must distinguish between relevant integrity-related convictions and unrelated or minor offences. Only relevant integrity-related convictions automatically trigger the maximum score.

A formal criminal charge is treated as serious because the matter has moved beyond allegation and into formal prosecution. It does not mean the person is guilty. It means the public integrity risk is materially elevated because prosecutors or law-enforcement authorities have acted on the matter.

Official investigations carry less weight than charges because an investigation may or may not result in a finding or prosecution. However, official investigations still carry meaningful public integrity risk when the investigation concerns corruption, fraud, abuse of office, procurement irregularities, conflicts of interest, or similar matters.

Media allegations carry the lowest score range among the major categories. They may still matter where they come from reputable sources, are repeated by multiple credible outlets, or are supported by documents, court papers, official records, named sources or investigative reporting. Media allegations must be clearly labelled as allegations.

Cleared, dismissed, withdrawn or corrected matters are not used as a subtraction category. The score is not reduced through negative points.

Instead, every profile is reviewed on a recurring 30-day cycle. During each review, the agent must check whether any matter has been withdrawn, dismissed, disproven, corrected, cleared, resolved or superseded by new information. If so, the active score must be recalculated based on the current evidence status.

Where a matter was previously listed but later cleared or dismissed, the profile may include a factual note explaining the update, especially where the matter remains relevant to understanding the public record. The agent must not continue scoring a matter as active where the available public record shows that the matter has been conclusively cleared or withdrawn.

The Scoring Criteria Sources document defines the baseline source list used by Integrity Index. For scoring purposes, sources are grouped into reliability tiers:

Every listed public figure must be reviewed at least once in every 30-day cycle. The review must include both identity verification and evidence verification.

The agent must perform the following review workflow:

• Confirm the person’s current office, title, party and status from official sources.
• Search the baseline Scoring Criteria Sources list for new information.
• Run targeted searches using the person’s full name, known name variations and office title.
• Group all findings into separate matters or incidents.
• Classify each matter according to the scoring categories.
• Apply the highest applicable score to each matter and avoid double counting.
• Check whether any previous matter has been withdrawn, dismissed, cleared, corrected or superseded.
• Recalculate the total score from the current active evidence record.
• Update the profile score, evidence table, scoring breakdown and date last reviewed.
• Flag high-risk changes for human review before publication where the change involves criminality, corruption, fraud, serious allegations, or a material score increase.

Each public-figure profile should make the score transparent and easy to understand. The profile page should include:

• Current Integrity Risk Score
• Risk band and colour bar
• Short summary of why the score was assigned
• Scoring breakdown by separate matter
• Evidence table with source links
• Current status of each matter
• Date last reviewed
• Methodology note
• Correction and right-of-reply information

The profile must clearly distinguish between conviction, charge, official finding, investigation, allegation and concern. These terms must not be used interchangeably.

The agent must follow these rules when applying the scoring system:

• Never score an unsourced claim.
• Never treat an allegation as a conviction.
• Never double count the same incident across categories.
• Always group evidence by matter before assigning points.
• Always use the highest applicable category for each matter.
• Always cap the final score at 100.
• Always preserve source links and source classifications.
• Always mark the date last reviewed.
• Always recalculate from current evidence during the 30-day review cycle.
• Always flag sensitive or high-risk claims for review before publication.
• Always record when a matter has been cleared, dismissed, withdrawn or corrected.
• Always keep the wording factual, neutral and evidence-based.

Integrity Index scores are based on publicly available evidence. A higher score means a higher level of publicly documented integrity risk. The score is calculated using a severity-based model that gives the greatest weight to relevant criminal convictions, then formal criminal charges, official findings, official investigations, ethics and disclosure issues, credible media allegations and serious conflict-of-interest concerns.

The scoring system is matter-based. Multiple articles or documents about the same incident do not create multiple scores. Separate unrelated matters may be added together, and the total score is capped at 100. Every profile is reviewed on a recurring 30-day cycle so that new information, corrections, withdrawals, dismissals or cleared matters can be reflected in the current score.

The Integrity Risk Score is not a court judgment and is not a declaration of guilt. It is a structured public-interest risk indicator based on available public evidence and transparent scoring criteria.