Behavioural model for detecting inhuman totem pop reaction times after damage.
0.0%
Precision
Minimizes false bans.
Percentage of flagged players who are
actually cheating. High precision means legitimate players stay safe.
0.0%
Recall
Minimizes missed detections.
Percentage of actual cheaters
on your server who get caught. High recall means fewer cheaters escape.
0.0%
F1 Score
The harmonic mean of Precision and Recall.
Represents the
overall accuracy and reliability of the detection engine.
How to read these numbers
Three figures describe every detection model, and they answer
different questions. Only one of them is about your players being treated fairly.
Precision
How often a flag is right
Of every player the model flagged, the share that really was cheating.
99.9% precision means false positives are virtually non-existent. This is the number that matters to the legitimate player
on your server, and the one Ecstacy is tuned to protect.
Recall
How many cheaters are caught
Of every player who really was cheating, the share the model flagged.
Deliberately the lower figure: pushing it up always costs precision. On a
Minecraft server the two mistakes are not equal. A missed cheater is a bad
afternoon. A wrongly banned regular is a player you do not get back.
F1
The two together
The harmonic mean of precision and recall, useful for comparing one model
against another in a single number. It is not a percentage of anything real,
so do not read it as an accuracy figure.
What a flag actually costs a player
Precision is a per-flag figure, and a single flag never punishes anyone. Every
detection family accumulates a violation level and acts only when it reaches
fail.max: twenty flags by default, forty on Auto Totem. The distance
between one flag and the threshold is exactly what turns a high-precision signal
into a decision you can stand behind.
On top of that, punishment is opt-in per family: the shipped Auto Totem
configuration has an empty command list and does nothing until you fill it in.
Servers that want to watch before enforcing set punish.enable: false
and keep every flag, alert and dashboard entry. The
checks overview covers how that fits
together.
What these numbers do not tell you
They are not a promise about your server. A model measured on
high-tier competitive play behaves differently on a casual SMP, in both
directions. Your own flag rate over a week is better evidence than any figure on
this page.
They do not cover movement cheats. Ecstacy detects combat
automation. Fly, speed and reach are a different problem. Run a movement
anticheat alongside it, which is what the
compatibility page is about.
They are not a live counter. These are evaluation figures for
the models, not a rolling tally of today's bans.
They do not account for your settings. The
precision tier you choose in checks.yml changes how
much evidence a check needs before it flags at all, and therefore changes the
volume you see.
Server impact
Scoring runs on Ecstacy's cloud cluster rather than on your hardware. The plugin
captures packets, raises alerts, drives the GUIs and applies punishments; it does
not run the analysis. In benchmarking that holds to no measurable TPS
impact up to 5,000 concurrent players.
The structural consequence matters more than the number: because the expensive half
lives cloud-side, the plugin's per-tick cost stays flat as detection improves. A
better model does not become more work on your main thread, and it arrives without
a jar swap.
Measuring it on your own server
Three things give you your own version of these figures, without waiting for a ban
to be disputed:
/ecstacy suspicious: players scoring near the bar without
crossing it. A populated list means detection is live even on a quiet day.
/ecstacy violations <player>: whether a violation level
climbed across sessions or spiked inside one fight. Those read very
differently.
/ecstacy feedback, from the [Report] suffix on an
alert. Mark a flag TRUE, FALSE or DUNNO.
Verdicts feed back into how detections are evaluated, and cost you nothing.
Benchmark FAQ
What is Ecstacy's Crystal Aura detection precision?
Crystal Aura detection reaches 97.0% precision and 91.1% recall, an F1 score of 93.9%, measured per-window on held-out high-tier Crystal PvP match data.
What is Ecstacy's Anchor Macro detection precision?
Ecstacy's Anchor Macro model achieves 99.0% precision and 95.2% recall (97.1% F1 score).
Does Ecstacy affect server TPS?
No measurable TPS impact up to 5,000 concurrent players. Inference runs on Ecstacy's cloud cluster, not on the game server, so the plugin's per-tick cost does not grow as detection improves.
What does 97.0% precision mean for my server?
Precision is the share of flagged players who really were cheating, so 97.0% means about three flags in a hundred are false positives. Punishment is not applied on a single flag either: every check accumulates a violation level and acts only at its threshold, twenty flags by default.
Why is recall lower than precision?
It is a deliberate trade. Recall is the share of real cheaters caught, and pushing it higher always costs precision. On a Minecraft server a missed cheater is a bad afternoon, while a wrongly banned regular is a player you do not get back, so the models are tuned to protect the second case.
