«Why this number?» explainability

«Why this number?»: the breakdown that shows you where each point comes from

Nine indicators that open one by one: exact definition, count, rate, saturated value, and points against their weight. With the total at the bottom.

Internet love tests give you a number and stay silent. This one opens the books and lets you check them.

In the WappLoveScore report, just below the Love Score, there’s a section called «Why this number?». Inside are the nine indicators that form the score, each in its own expandable section. When you open any of them, you see four raw data points: the count (how many times it actually occurred in your chat; for the three indicators that are not a count, the relevant data: median seconds, active days, or percentage), the rate (normalized, almost always per 1,000 messages), the saturated value (from 0 to 1, the count already compared with the maximum threshold), and the points obtained against that indicator’s weight. At the top, in the header of the expandable section, a bar tells you at a glance how much of that weight you’ve utilized.

Before the numbers comes the exact definition of what has been measured, without euphemisms. «Direct expressions of affection» is not «nice messages»: these are messages containing «I love you», «I adore you», «I’m crazy about you» or their English equivalent, discarding negated phrases («I don’t love you anymore»), periphrases («I want to tell you something») and third-party quotes. «Good-morning/good-night ritual» refers to active days when either person says good morning between 5:00 and 12:00 in their first three messages, or good night after 20:00. And the definition includes the threshold for reaching the maximum, so you know what you are being compared against.

At the bottom of the section is the total: how many points each indicator contributed and what the total is out of 100. Below that, a phrase the app explicitly states: «The score is mathematical and objective. AI is not involved in this calculation». The engine is deterministic and runs in your own browser, so the same file always yields exactly the same number.

And there’s a second expandable section, «How I read the chat», which audits the step prior to calculation: the format detected in your export (iPhone or Android, 12- or 24-hour clock, dates in dd/mm or mm/dd), how many messages and how many multimedia items were counted, how many deleted and edited messages, what percentage of lines could not be recognized, the date range analyzed, the detected chat language, and the applied lexicon. Plus a timezone warning: WhatsApp export does not include it, so the timezone of the phone that exported the chat is assumed.

Benefits of «Why this number?» explainability

Nine expandable sections, one per indicator

direct affection, nicknames, love emojis, good-morning/good-night ritual, response synchronicity, consistency, reciprocity, shared humor, and active interest.

The weights are visible and total 100

expressions of affection are worth 20 points, and the other eight indicators are worth 10 each. Each header shows «points obtained / weight» and a bar indicating what has been utilized.

Four figures per indicator

count, normalized rate, saturated value from 0 to 1, and points. No rounded percentages without origin: with these numbers, you can reconstruct the count manually.

Ventaja 4

The exact definition of what has been measured, including exclusions. For «I love you», negated phrases, periphrases, and third-party quotes are discarded; for the median response time, waits longer than 12 hours are excluded, because that is no longer responding.

The maximum threshold, written in the definition itself

8 «I love you» per 1,000 messages, 25 nicknames, 40 love emojis, 60 laughs, 12 questions of interest, good-morning/good-night ritual on 60% of active days, and activity on 85% of days in the range.

The sum at the bottom and the explicit notice

«The score is mathematical and objective. AI does not intervene in this calculation». The calculation is deterministic and runs in your browser, so it is repeatable.

«How I read the chat»

detected format, messages and multimedia counted, deleted, edited, percentage of unrecognized lines, date range, chat language, applied lexicon, and the assumed time zone.

The limits, indicated where appropriate

the dictionaries are in Spanish and English and have a weight of 50 out of 100 points, so if your chat is in another language, those indicators are marked as incomplete instead of faking a number, and the report’s reliability drops a notch.

Noa Ferrer

A number you can audit

Upload your chat export, see your Love Score, and break it down indicator by indicator. Free, and the chat is analyzed on your own mobile.

Why this number? How the Love Score is calculated, indicator by indicator