Methodology
How ratings, moderation and the Matrix are calculated. Published in full, because a hidden formula is indistinguishable from a manipulated one.
How the rating is calculated
The displayed rating is the average of published reviews, weighted by two factors: the verification level of the reviewer and how long ago the review was published. There is no statistical correction in the displayed rating — it has to reflect what reviewers actually said.
weight = verification_weight × recency_weight
Weight by verification level
| Verification level | Weight | Signal |
|---|---|---|
| Published | 0.70 | Moderated, no employment link verified. |
| Verified | 1.00 | Corporate email confirmed. |
| Verified with evidence | 1.30 | Proof of use checked by our team. |
Weight by recency
The weight decays with a half-life of 30 months and never falls below 0.45. A review from three years ago does not describe today’s supplier — but it is not worth zero either.
| Months since publication | Weight |
|---|---|
| 0 | 1.000 |
| 6 | 0.871 |
| 12 | 0.758 |
| 24 | 0.574 |
| 36 | 0.450 |
| 60 | 0.450 |
How the ranking is calculated
The ranking uses a different quantity from the displayed rating: a Bayesian mean with a prior. Without it, a supplier with a single 5.0 review would lead the category ahead of one with 4.6 and two hundred reviews. That score is never shown as if it were the supplier’s rating.
C = 8.0 (prior weight)
M = 4.527 (published global mean)
How the Matrix is calculated
The Matrix places the suppliers in a category on two axes. The axes are deliberately different from each other.
Satisfaction axis — absolute
Derived from the same Bayesian score used for ranking, rescaled from 1–5 to 0–100. It is absolute: a supplier with 4.6 has high satisfaction regardless of the category. Normalising it against peers would make the best supplier in a weak category look excellent.
The prior pulls a score toward the global mean, and that pulls upwards anyone sitting below it. So satisfaction is capped at the average reviewers actually gave: the prior can lower a position built on thin evidence, never raise one built on a poor rating.
Market presence axis — relative to the category
A weighted composite, normalised against the strongest supplier in the same category. Presence only means something against peers: 200 reviews is a lot in Testing Labs and very little in Slots. A square root compresses the tail so that one outlier does not flatten everyone else against the left edge.
| Factor | Weight |
|---|---|
| Published review volume | 35% |
| Unique profile views (90 days) | 25% |
| Markets served | 15% |
| Verified licences | 15% |
| Profile completeness | 10% |
Where the lines fall
The satisfaction cut is fixed at 62.5 on the 0–100 scale (equivalent to 3.5 stars). The presence cut is the category median, recalculated every quarter and recorded in the snapshot.
- Leaders High satisfaction and strong market presence.
- High Performers High satisfaction, smaller presence than the category leaders.
- Contenders Strong presence, satisfaction below the category cut.
- Niche Focused presence, satisfaction below the category cut.
Minimum critical mass
A category only gets a Matrix with at least 5 suppliers holding 5 or more published reviews each. Below that, the Matrix would be statistical noise presented with authority — the worst possible outcome.
A published snapshot never changes
Each quarter is frozen when published, together with the parameters and the raw inputs used. A quadrant that changes retroactively destroys the credibility of every badge already issued from it — a supplier that put "Matrix Leader" on their site needs that to stay true.
Merit badges come out of the published snapshot by criteria over data. None are granted by hand and none can be bought.
How moderation works
Every review goes through automated triage. Triage decides nothing: it raises signals and presents them, with the reason, to a human moderator. A review with a combined risk of 40 or more is never published without a person seeing it.
Self-reviews and competitor reviews are explicit policy violations. Both are detected through the corporate email domain.
Signals raised in triage
| Signal | Risk weight |
|---|---|
self_review | 100 |
competitor | 90 |
duplicate_ip | 45 |
fresh_account | 20 |
free_email | 15 |
thin_content | 25 |
undisclosed_incentive | 60 |
extreme_only | 10 |
How disputes work
A supplier never deletes a review. They can reply publicly, as many times as they want, and open a formal dispute pointing to an objective policy violation. Moderation decides within 5 business days and the decision is recorded with its reasoning — a decision with no recorded reason is indefensible months later, which is exactly when it gets questioned.
An upheld dispute removes the review for a policy violation. Never for being negative.
What money does not buy
- Money never changes a rating, an organic position or a quadrant in the Matrix.
- Sponsored placement is a separate layer, labelled on every appearance.
- The free profile is permanent and receives reviews like any other.
- Incentives, where they exist, come from the platform, are never conditional on the rating, and are declared on the review itself.