A trust score that says how much of itself was measured

Most reputation scores are a single number with no way to tell a confident one

from a guess. Four stars from six ratings and four stars from six thousand print

identically, and the reader supplies the confidence themselves, usually

generously. This post is about what happens when a protocol refuses to do that.

Here is the live trust response for Cresva's own storefront, captured 16

September 2026:

{
  "score": 50,
  "unweighted_score": 62.5,
  "tier": "bronze",
  "trend": "stable",
  "components": {
    "data_accuracy": 50,
    "fulfillment_speed": null,
    "return_rate": null,
    "negotiation_fairness": 50,
    "response_reliability": 50,
    "dispute_record": null,
    "customer_satisfaction": 50
  },
  "measured_components": {
    "data_accuracy": true,
    "fulfillment_speed": false,
    "return_rate": false,
    "negotiation_fairness": true,
    "response_reliability": true,
    "dispute_record": false,
    "customer_satisfaction": true
  },
  "measured_weight": 0.55,
  "provisional": true,
  "provisional_reason": "No completed transaction has been recorded for this brand, so nothing here rests on observed behaviour.",
  "verified_transactions": 0
}

Three of the seven are null, and that is the design

fulfillment_speed, return_rate and dispute_record come back null. Not

zero, not fifty, not a category average. Nothing has happened that would measure

them, so the response says nothing rather than something plausible.

A zero would have been the worst available answer. On a 0 to 100 scale a zero

reads as terrible, so a storefront with no dispute history would be reported as

having the worst possible dispute record. A middle value is only slightly better

and is worse in one respect: it is unfalsifiable. A reader cannot tell a

measured 50 from an assumed one, so every number in the response becomes

suspect at once.

measured_components exists so a consumer does not have to infer the

distinction from a null. It is the same information stated positively, because a

client that filters on components.fulfillment_speed > 60 and one that first

checks measured_components.fulfillment_speed behave very differently on the

same payload.

measured_weight is the field to read first

0.55 means components carrying 55 percent of the total weight have evidence

behind them. Below 0.5 no tier is published at all: tier comes back as

unrated, which is not a low tier but the absence of one.

This is the field that separates a score from a verdict. A 50 with

measured_weight: 1.0 is a storefront that has been observed and is average. A

50 with measured_weight: 0.55 is a storefront that has been half observed.

Those are different facts and a single number cannot hold both.

provisional says it in words

provisional: true with provisional_reason spelling out the cause:

`"No completed transaction has been recorded for this brand, so nothing here

rests on observed behaviour."` The same response carries

verified_transactions: 0, which is where that sentence comes from.

A field that explains itself in prose is unusual in an API and it is here for a

specific reader: the one who is about to put this number in front of a shopper.

A boolean can be ignored. A sentence that will be quoted is harder to ignore.

The ladder

| tier | score |

|---|---|

| Unrated | under 40 |

| Bronze | 40 to 59 |

| Silver | 60 to 74 |

| Gold | 75 to 89 |

| Platinum | 90 to 100 |

Unrated is on the ladder deliberately. A ladder that starts at Bronze is a

ladder every storefront appears to be on, and the most common real state, not

enough evidence yet, would have nowhere to sit.

What to do with it

GET /api/storefront/{brandId}/trust for the score.

GET /api/storefront/{brandId}/trust/signals for what went into it.

If you are building an agent, the useful rule is to branch on

provisional before you branch on score. A provisional score is not a bad

score. It is a statement that the question has not been answered yet, and

presenting it to a shopper as though it had is the failure this design exists

to make difficult.