← The Diagnostic

Ridgeline Interactive is invented, and so is every number below. The shape is real. This is what lands on your desk at the end of a Diagnostic: four findings, each one carrying the check that could have killed it, one experiment registered before its data exists, and a plain statement of what the read cannot tell you.

No studio, game, player or purchase below is real, and no figure on this page came from anyone's telemetry.

Every measured quantity here carries a diagnostic that could have broken it, and the result of running that diagnostic. A finding delivered without its probe is advertising.

Finding 01

18.4% of your week-8 “active” accounts are propped.

They log in, they play, and they read as retained. Their session initiation has migrated almost entirely onto system prompts — daily reset, expiring pass progress, squad invites they answer and never originate. Remove those and the model puts their survival at eleven days. This is churn that has already happened and has not yet posted.

The probe · run 06 Mar

If these accounts are genuinely propped, withholding one prompt class for a fortnight should separate their exit hazard from the matched control. It did, by 4.1 points. Had the two curves stayed together, the 18.4% would have been an artifact of how we defined a prompt, and the finding would have come to you withdrawn.

Finding 02

Your churn curve is two curves. 61% of it is satiation.

Of the players who left between weeks 6 and 12, roughly three in five left the way someone finishes a book. The rest left the way someone leaves a room. The two shapes want opposite fixes, and a single retention number averages them into a lever that moves neither.

The probe · run 09 Mar

Friction-shaped exits should sit near friction events. Within 48 hours of a moderation flag, a rage-quit or a lopsided loss streak, they appear at 3.4 times the rate of the satiation group. A ratio near one would have meant the split was noise wearing two names.

Finding 03

Who a player is placed with moves 12-week retention by 9.6 points.

Players in the top third by exposure to squads that held together and worked retain at 45.7%. The bottom third retains at 36.1%. The gap survives conditioning on skill, session count and install cohort, which is the form of the claim we will defend.

The probe · run 11 Mar

The obvious alternative explanation is selection: good players get placed with good players and would have stayed anyway. Conditioning on skill and session count shrank the gap from 13.2 points to 9.6 and left it standing. Had it collapsed toward zero, this would read as a finding about your matchmaker's sorting and not about squads at all.

Finding 04

Composition lifts spend by a third — and decides whether it stays bought.

With a high spender in the squad, the other members buy at 9.4% against 7.1% without, net of playtime. The half that matters arrives ninety days later: purchases made inside a working squad are still standing at 81%, and purchases made inside a squad in blame or over-reliance at 46%. The second group is revenue this quarter and a lost player next.

The probe · run 12 Mar

A squad “with a high spender” carries that player's own purchases, and counting them manufactures the lift out of nothing. Excluded here by rule, registered before the number was computed. With them left in, the lift reads 14.1% — which is the figure to distrust if you ever see it quoted without this sentence attached.

Two fields were absent from the slice, and their absence is not a rounding error. Without re-queue-together we cannot see whether a squad re-forms on its own once your system stops assembling it, which is the cleanest single signal that a tie is real. Without a within-match timeline we read each match as a set of totals, so a squad that fell apart in the last four minutes and one that was never together look alike.

Beyond the fields: these are associations until the registered experiment runs. Nothing here licenses a claim about an individual player's future, and nothing here is a statement about what your matchmaker is trying to do. We measured what your data records. Intent is not in the data.

The Diagnostic is a fixed-scope read on a slice of your telemetry, under NDA, at a published price. You get the four-finding shape above, each with its probe, one experiment registered before its data exists, and a statement of the ceiling. If you want to know first whether your fields can carry it, the readiness audit answers that in four minutes and asks nothing of you.