Scoring Methodology

Last updated: March 1, 2026

Purpose: This page documents the exact algorithm used to compute Reputation Scores on My Blog. We publish this openly so every score can be understood, verified, and challenged.

1. Overview

Each account listed on My Blog receives a composite Reputation Score between 0 and 100. The score is a weighted sum of six independent component scores, each also ranging from 0 to 100. Components are evaluated using only publicly contributed data—no private or platform-internal information is used.

The formula is:

Score = ∑ (Component_i × Weight_i), clamped to [0, 100]

2. Component Weights

Weights reflect the relative importance assigned to each signal. They sum to 1.0 and are configurable by site administrators, though defaults are shown below.

Component	Weight	Description
Report Volume	0.25	Number of independent, approved reports filed against the account.
Reporter Credibility	0.20	Weighted average reputation of the reporters who filed those reports.
Evidence Strength	0.20	Quality and quantity of evidence attached to reports.
Behavior Consistency	0.15	Degree of agreement among reports on the type of behavior observed.
Account Age Anomaly	0.10	Whether the account’s age-to-activity ratio is anomalous.
Platform Confirmation	0.10	Whether the native platform has independently taken action.

3. Component Scoring Details

3.1 Report Volume (25 %)

Uses a logarithmic scale to diminish the marginal impact of each additional report. This prevents score inflation through mass-filing while still rewarding independent corroboration.

Component Score = min(95, 30 × ln(report_count + 1))

Approximate values:

Reports	Component Score
1	≈ 21
3	≈ 42
5	≈ 54
10	≈ 72
20+	Approaches cap of 95

3.2 Reporter Credibility (20 %)

Computes the simple average of the reputation scores of all reporters who filed approved reports against the account. New reporters begin with a baseline reputation of 10 (out of 100). As reporters submit accurate, evidence-rich reports their reputation increases; rejected or flagged reports reduce it.

This mechanism ensures that reports from consistently reliable contributors carry more weight than reports from new or unreliable sources.

3.3 Evidence Strength (20 %)

Each report receives an evidence score based on the type and quantity of evidence provided:

Evidence Type	Points Per Item
Archive.org link	+30
Screenshot	+20
Direct post URL	+15

Each report’s evidence score is capped at 100. The component score is the average evidence score across all approved reports for that account.

Archive.org links receive the highest weight because they provide independently verifiable, timestamped snapshots of content that may otherwise be deleted.

3.4 Behavior Consistency (15 %)

When multiple reports independently describe the same type of behavior (e.g., spam, coordinated amplification), the consistency strengthens the signal. The component score equals the maximum agreement ratio across behavior types multiplied by 100:

Component Score = (max_count_for_any_behavior_type / total_reports) × 100

If all five reports tag the account as “spam,” the consistency score is 100. If three tag it as “spam” and two tag it as “impersonation,” the score is 60.

3.5 Account Age Anomaly (10 %)

Evaluates whether the account’s follower count is anomalously high relative to its age—a common signal for bot-farm accounts that gain followers rapidly through artificial means.

Component Score = min(100, (follower_count / age_in_days) × 2)

When account creation date or follower count is unavailable, a neutral default score of 25 is used so that missing data neither inflates nor deflates the composite.

3.6 Platform Confirmation (10 %)

Checks whether the social-media platform itself has taken independent action on the account. Platform status is recorded via a taxonomy and scored as follows:

Platform Status	Component Score
Banned	100
Suspended	75
Confirmed (by platform)	60
Disputed	10
No action / Unknown	0

4. Score Bands

The raw composite score is grouped into interpretive bands to aid users who are not familiar with the numeric scale:

Range	Band Label	Color
0–19	Insufficient Evidence	█ Gray (#9CA3AF)
20–39	Low Suspicion	█ Yellow (#EAB308)
40–59	Moderate Suspicion	█ Orange (#F97316)
60–79	High Suspicion	█ Red (#EF4444)
80–100	Confirmed Bad Actor	█ Dark Red (#7F1D1D)

5. Confidence Indicator

Every score is accompanied by a confidence level that reflects how much data supports it. Confidence is based on a composite “data points” count: the number of approved reports plus bonuses for the presence of evidence and multiple independent reporters.

Data Points	Level	Meaning
≥ 5	High	Multiple independent reports with evidence.
≥ 3	Medium	Some corroborating data available.
≥ 1	Low	Limited data; score may shift substantially.
0	None	No data yet; score is a placeholder.

6. Top Contributing Factors

Each score breakdown displays the three components with the highest weighted contribution to the final score. This lets users see at a glance why an account received its score rather than relying on the number alone.

7. Score Recalculation

Scores are recalculated automatically when:

A new report is approved against the account.
An existing report is removed or its status changes.
A moderator triggers a manual recalculation.
An administrator runs a batch recalculation for all accounts.

Historical scores are not retroactively recalculated unless explicitly triggered. A score-history log records every recalculation with a timestamp and trigger type for audit purposes.

8. Known Limitations

Community-sourced data. All inputs come from user submissions. The system cannot independently verify claims and is susceptible to false positives, coordinated manipulation, and stale data.
Logarithmic saturation. The report-volume component saturates at high counts, meaning additional reports beyond ~20 have diminishing effect. This is intentional but may under-weight accounts with very large report volumes.
Reporter baseline. New reporters start with a baseline reputation of 10. Until they build a track record, their reports receive relatively low weight.
Missing metadata. When account creation date or follower count is unknown, components default to neutral values. This means accounts with incomplete metadata may be under-scored.

9. Model Evolution

The scoring model is versioned and may be updated over time as data grows and edge cases emerge. Changes to weights, formulas, or component definitions will be documented on this page with a revised “Last updated” date. Material changes will also be noted in the Transparency Statement.

10. Contact

Questions or feedback about the scoring methodology: admin@botidentifier.com.