The words Minealyze uses

The vocabulary below is the same everywhere: on the site, in the dashboard, in this doc, in your Discord. One word, one meaning.

One rule sits behind the whole product: every piece of content carries its origin. Measured means counted from your real data. AI-written means a sentence the AI wrote from those numbers: it never computes a number itself, and any sentence that isn't anchored to a real figure gets removed before it reaches you.

For each word: what it is, why it matters to you, a worked example, and, for anyone who wants to verify, the technical term analysts use ("Pros call it:").

Your server, day to day

Players today · Players this month

How many distinct players logged in today, and this month. Each player counts once. If "today" drops while "this month" holds, your players are spacing out their visits: that's the very start of slipping away.

Example: "24 players today, 80 this month."

Pros call it: DAU, MAU.

Loyalty

Of your month's players, the share who show up on a given day, shown as "X out of 10." It's the difference between a drive-by server and a community that's alive: it moves before the total even drops.

Example: "80 players this month, 24 a day on average: loyalty 3 out of 10, solid for a survival server."

Pros call it: stickiness (DAU/MAU).

Players who come back

Of a week's new players, how many replay the next day ("Still playing after 1 day"), the week after ("after 7 days"), or the month after ("after 30 days"). This is THE sign your welcome makes people want to stay: a server rarely dies overnight, it dies from new players who never come back.

Example: "Of last week's 12 new players, 5 replayed this week."

Pros call it: D1/D7/D30 retention.

The arrivals

The new players who showed up the same week, tracked together over time, like a graduating class. Comparing this week's arrivals to last week's tells you if your welcome is improving or slipping.

Example: "The May 12 arrivals: 12 new players, 5 still around a week later."

Pros call it: cohort.

Typical session

The session length right in the middle of all of yours: half last longer, half shorter. If it shrinks, your players are logging in "to check" and leaving, a sign of boredom before they go.

Example: "Typical session: 42 min. Last month: 51 min. Your players are staying less."

Pros call it: median session.

Peak hours

The map of when your server is alive: sessions by day of week and hour. This is when to schedule events and know when to show up yourself.

Example: "Your peak: Saturday 9pm, 31 sessions. Your quiet spot: Tuesday morning."

Pros call it: punchcard.

Server health

A score out of 100 that summarizes several measures (comebacks, loyalty, new players, sessions...), with a bar for each. One number to watch week to week; the bars tell you WHAT to fix.

Example: "72/100: solid. Lowest bar: 'new players stay' (2 out of 10). Start with the welcome."

Pros call it: composite health score.

Who's slipping away, and why

Slipping away

The central word of the product: a player spacing out their visits until they stop coming back. Minealyze compares them to THEIR OWN habits (login rhythm, session length, friends still active), never a single rule for everyone; a player on vacation isn't "slipping away."

Example: "Zylkor used to play every 2 days; hasn't been seen in 9 days. He's slipping away."

Pros call it: churn.

Fading fast · Keep an eye · Doing fine

The risk verdict: the answer in words to "will they come back?", always with its context, never a bare score on its own.

Example: "Zylkor: fading fast. 7 out of 10 players in his situation don't come back."

Pros call it: risk score, propensity.

Who's slipping away (the list)

The list, rebuilt every night, of players who are slipping away: the name, the top reason, the verdict, and the button to act. Sorted by urgency, never by money.

Example: "This week: 4 players are slipping away, 2 of them store supporters."

Pros call it: watchlist (the word stays in the API, never on screen).

Top reason

For each flagged player, the number one cause of their slipping away, in one sentence with real units: never a variable name, never a bare number.

Example: "Not seen in 9 days (usually: every 2 days)."

Pros call it: dominant factor.

Player groups

Pillars, becoming regulars, new, slipping away, can't lose them, gone quiet. Every player lands in a group based on three simple questions: seen recently? comes often? supports the store? A player moves between groups the moment their behavior changes: these aren't castes.

Example: "80 players: 12 pillars, 9 becoming regulars, 4 slipping away, 41 gone quiet."

Pros call it: RFM segments (Recency / Frequency / Monetary).

Store support

The total a player has put into the store. An optional sort criterion, never the default sort: a pillar who never paid is still worth nudging back.

Example: "Zylkor: $74.50 in store support since January."

Pros call it: LTV (lifetime value).

Why that week's arrivals left

The comparison between the new players from a batch who stayed and the ones who left: what did the ones who left NOT do? A lead is only shown if the gap is clear, always with "a lead, not a certainty." If the data isn't enough, the page says so and invents nothing.

Example: "The May 12 arrivals: the 7 who left hadn't joined a faction; the 5 who stayed had. Lead: joining a faction."

Pros call it: cohort autopsy, correlation ≠ causation.

The first 5 minutes

A new player's path, step by step (arrive → move → place a block → talk → come back tomorrow), and the exact spot where it breaks. Most new players decide within the first 5 minutes.

Example: "Of 100 new players: 86 move, 54 place a block, 38 talk, 24 come back the next day."

Pros call it: funnel, drop-off.

The hook

The step in the first 5 minutes most tied to coming back, on YOUR server. It's the most reliable sign, not a guarantee: getting someone to place a chest doesn't save them for sure.

Example: "Your hook: placing a first chest. Those who do come back twice as often (68% vs. 31%)."

Pros call it: aha milestone, phi correlation.

Alerts

When a server number swings sharply out of its ordinary range (economy, event volume, disputed payments), Minealyze flags it; you decide. Never a ban, never an automatic action. An empty screen is good news, and says so.

Example: "Tonight at 3:40am: emeralds earned per hour jumped to 4.2x normal. Lead: an unregulated auto-farm."

Pros call it: anomaly, z-score, EWMA, IQR.

Chargeback

A store payment reversed after the fact by the player's bank: the money is taken back from you, often with fees. A wave of chargebacks triggers an alert.

Example: "3 chargebacks this week out of 41 payments: worth watching closely."

Pros call it: chargeback rate, dispute.

Act

The nudge

A message sent to players who are slipping away, with an optional small comeback reward (cosmetic or social, never a gameplay advantage), over Discord or in-game. It's the same thing you already do by hand when you type "when are we seeing you again?" to a friend.

Example: "May 12 nudge: 24 players slipping away, 21 nudged, 3 set aside."

Pros call it: winback.

Preview → Confirm → Log

The mandatory sequence for every nudge: you see everything (recipients, message, reward, who's set aside), you confirm it yourself, and everything is written to the log. Nothing ever goes out behind your back.

Example: "Log, May 12, 9:03am: nudge confirmed, 21 nudged, 3 set aside."

Pros call it: preview/confirm/log, audit trail.

Do not disturb

A player who asked for it never receives another nudge: they're removed up front, before any send, manual or automatic.

Example: "24 targeted → 23: 1 player on 'do not disturb', excluded automatically."

Pros call it: opt-out.

The caps

Hard limits that keep your community from getting hurt: no two nudges to the same player within the window, no nudge on a tiny group, and for auto-nudges a monthly cap on actions and budget. Too many messages drives players away faster than no message at all.

Example: "Armed rule: max 40 actions and $20 a month. On the 41st, it refuses and logs why."

Pros call it: frequency cap, minimum segment size, fail-closed.

Never pay-to-win

Any reward generated or sent by Minealyze is cosmetic or social: it gives no gameplay advantage at all. A questionable offer is blocked and explained, never quietly swapped for something else. It's Mojang's rule (the EULA), and ours.

Example: "A cosmetic 'Ember' key: accepted. An enchanted sword kit: refused, with the explanation."

Pros call it: non-P2W, command allowlist.

Auto-nudges

A rule YOU arm ("when a regular is gone for 7 days, send them this message"), which Minealyze then runs on its own, under caps, with a "Kill switch" button that always works.

Example: "'Regulars absent 7d' rule: armed on May 3. Since then: 4 runs, 11 nudged, 2 refused (cap)."

Pros call it: capped semi-automation, kill-switch.

Prove

Nudged · Set aside

When you nudge a group, Minealyze keeps a few of them aside, chosen at random, without sending them the message: that's so you can compare afterward. They're neither punished nor forgotten: they're eligible again from the very next nudge.

Example: "24 players slipping away: 21 nudged, 3 set aside."

Pros call it: holdout, control group.

The real effect

What the nudge actually changed: the extra comebacks among the nudged group, compared to the set-aside group. Shown in players first, the percentage second, always with its uncertainty range.

Example: "57% came back among the nudged, 33% among the set-aside: about +24 points."

Pros call it: lift, 90% confidence interval.

Proven · Not enough data

Our honesty rule: Minealyze only writes "proven" if the gap is too big to be luck. Otherwise: "not enough data," and that number counts in no total. It's neither a failure nor a penalty: it's the same rule that stops us from selling you a fake result.

Example: "+24 points on 24 players: small group, wide range. +6 points on 300 players: proven."

Pros call it: statistical significance, p-value.

Proof (the page)

The page that answers "did my nudges actually work?": every nudge, its real effect, its verdict, and the honest total. "Revenue kept (proven)" only adds up proven nudges, and always comes after the total in players who came back.

Example: "3 nudges this quarter: 1 proven (+5 players), 1 with no measurable effect, 1 not enough data."

Pros call it: holdout report.

What Minealyze tells you

Morning recap

Every morning, 3 to 5 things to do today, in your Discord. If nothing's urgent, it says so.

Example: "Morning recap: 2 players slipping away (1 a pillar), an emerald alert at 4.2x at 3:40am."

Pros call it: daily brief, next-best-action.

Weekly review

Once a week, your server's story in plain words: what moved, why, and what to do next week. The morning recap makes you act; the weekly review makes you understand.

Example: "Week of May 12: 61 players (+4). This batch of arrivals is holding better. Weak spot: Saturday night."

Pros call it: weekly briefing.

Copilot

Ask your question in plain English, it answers with YOUR numbers, cited, and can suggest a nudge, never send one. If it doesn't have the number, it doesn't have an opinion.

Example: "'Who's slipping away?': 4 players this week, including Zylkor. Suggest a nudge?"

Pros call it: natural-language queries, read-only analytics views.

Measured · AI-written

"Measured" means counted from your real data. "AI-written" means the sentence that puts that number into words. The AI never computes a number itself, never decides a score, never picks a target.

Example: "This batch's retention: 5 out of 12 [measured]. 'Your welcome is improving' [AI-written]."

Pros call it: deterministic vs. generative.

Why am I seeing this?

The link present on every calculation or AI output (verdict, alert, hook, real effect) that opens the calculation in plain words, then in technical terms. It turns a black box into an advisor.

Example: under "Zylkor: fading fast" → "he used to play every 2 days, nothing for the last 9."

Pros call it: explainability (XAI).

Trust and data

Where you stand

How your server compares to Minealyze servers of the same kind and size, anonymously. Opt-in participation, no usernames involved, no one sees your server by name.

Example: "Your 7-day comebacks: better than 7 out of 10 survival servers your size."

Pros call it: benchmarks, percentiles.

Anonymous because the group is big enough

A shared-pool statistic only shows up once enough servers take part that no one could be recognized. Below that: nothing, and Minealyze says so.

Example: "Survival servers, 50-150 players: 4 contributors. It takes 8: the comparison isn't published yet."

Pros call it: k-anonymity, threshold N.

Coded usernames

Before anything reaches an AI, player usernames and IDs are swapped for codes. Chat, signs, and books aren't collected by default.

Example: "The AI sees 'player P-4821 logs in every 9 days,' never 'Zylkor.'"

Pros call it: pseudonymization, per-tenant HMAC.

The plan meter

What your plan allows this month (nudges, copilot questions), shown before you act, never as a surprise afterward.

Example: "Nudges: 3 of 4 used this month. Copilot questions: 3 of 15."

Pros call it: quota, entitlement.

The two keys

The server key (one per server) connects your plugin to your Minealyze space: required, shown once. Your AI key (optional, Network and Hosting plans) lets you plug in your own AI account: writing runs and gets billed on your side. It's never required to use Minealyze.

Example: "Onboarding: paste the server key into config.yml. Settings (Network): add your AI key if you want to pay for AI directly."

Pros call it: ingestion API key, BYOK (bring your own key).

Next step

Check the API reference for the exact request/response shape, or the FAQ for common questions.