Lewati ke konten utama
TransparencyMethodology

How Mindber scores AI tools

Full documentation of the Mindber scoring system — engagement signal weights, score formulas, confidence levels, red-flag rules that block publication, and the conditions that route a record to human review.

Version v1.0Updated

Score architecture

Every product receives one headline composite — Mindber Score™ — computed from three sub-scores on a 0–10 scale, each rounded to 1 decimal:

mindberScore = round(
  innovationIndex × 0.40  // Mindber Innovation Index™
  + functionalityScore × 0.50  // Mindber Functionality Score™
  + activityScore × 0.10  // Mindber Activity Score™
, 1)

Detailed formulas for each sub-score are on the Scoring methodology page. Sub-score breakdowns are visible in the score popover on every product page.

Engagement calculation

The Mindber Activity Score is a 0–100 number derived from six independent public probes. Each signal is first normalized to 0–10, then multiplied by its weight. A 14-day decay biases the displayed value toward the most recent observations.

activityScore_raw = Σ( normalize(signal, 0, 10) × weight )
activityScore_displayed = activityScore_raw × decayFactor(14d)
Traffic
Daily–weekly by tier
30%
Social (YouTube + LinkedIn)
Weekly
20%
Hiring (ATS endpoints)
Weekly
15%
HTTP homepage probe
Daily–monthly by tier
15%
DNS (A / MX / NS)
Daily–monthly by tier
10%
Pricing page
Daily–monthly by tier
10%

Supplemental signals (GitHub API, Reddit public API, RSS/Atom feeds) are collected but carry no standalone score weight in the current model. See Data sources for the full source list.

Confidence levels

Confidence reflects how many of the three sub-scores fed the composite calculation. It is not a statistical confidence interval — it is a data-completeness label.

High
All three sub-scores present
Innovation 40% · Functionality 50% · Activity 10%
Medium
Two of three sub-scores present
Weight of missing input redistributed proportionally
Insufficient
Fewer than two sub-scores present
No Mindber Score shown — page renders "Score pending"

A record with confidence Insufficient routes to the human review queue regardless of any individual sub-score.

Red-flag rules

The following conditions trigger a hard block on content publication. The admin override flag defaults to false. A blocked record is written to dirArticleQCFails with the check ID and reason; the publish endpoint returns HTTP 422.

#01Pricing claim without source

Article references pricing (tiers, amounts, discounts) with no source URL captured in frontmatter

Source required: Vendor's official pricing page URL

#02Unverified company name

Vendor or company name not found in the canonical entity list

Source required: Official company domain or LinkedIn company page

#03Uncited legal claim

Lawsuit, regulatory action, or legal proceeding mentioned without inline citation

Source required: Court record, official regulatory filing, or verified news report

#04Stale financial claim

Revenue, valuation, or funding figure with source older than 90 days

Source required: Press release, SEC filing, or verified announcement within 90 days

#05Broken external link

Any outbound URL in the article returns a non-200 HTTP response

Source required: Updated URL or removal of the reference

#06Copyright risk

More than 15 consecutive words reproduced verbatim from an external source

Source required: Original source with explicit license, or paraphrase the passage

#07Sub-score floor breach

Any scoring sub-score (Innovation, Functionality, Activity) falls below 4 / 10

Source required: Human review required — editor must approve before publish

#08Confidence below threshold

Composite score confidence below 0.6 (fewer than two sub-scores available)

Source required: Human review required — additional scoring inputs or score withheld

Human review queue

Records are routed to the admin review queue when any of the following conditions is true. A queued record is not published until an editor approves it.

  • Any sub-score (Innovation, Functionality, Activity) falls below 4 / 10
  • Composite score confidence falls below 0.6
  • A banned-word safety filter replaces generated content
  • Any red-flag rule (#01–#08 above) triggers

Editors may override individual flags via PATCH /admin/articles/:slug/qc-override with a required reason field. Every override is logged and auditable.

Disclaimer

Scores reflect publicly observable signals at the time of measurement. They do not assess product quality, company health, financial viability, or legal standing. Scores may lag reality between cadence cycles. This content is not legal, financial, or investment advice. Mindber accepts no liability for commercial decisions made on the basis of these scores.

How Mindber Scores AI Tools — Full Methodology